By now, it feels like ancient history, but there have been numerous attempts to transform how we interact with AI through novel hardware. Remember the Humane AI Pin? The Rabbit R1? These gadgets promised to integrate large language models seamlessly into our daily lives with voice conversations, embedded cameras, and even the ability to project visuals onto our hands. The vision was compelling: a wearable AI assistant that doesn’t require a phone. The execution? Let’s just say the reviews were harsh.

Despite these setbacks, a consensus has emerged: smartphones, for all their versatility, may not be the ideal platform for multimodal AI. The phone is trapped behind a screen you must constantly unlock, navigate, and tap. The other prevailing view holds that the real future lies in mixed reality glasses, devices you’d wear all day, blending AI seamlessly with your visual field.

But that technology is not ready yet.

We’re asking for a device that’s as light as sunglasses, as powerful as the latest iPhone, capable of blending 3D objects with reality in real time, equipped with cameras and cellular connectivity, lasting all day on a single charge, and priced like a smartphone. The Apple Vision Pro, impressive as it is, remains a hefty “face brick” tethered to an external battery pack with a price tag exceeding $3,500. To achieve true practicality, mixed reality glasses would need to shrink to swim-goggle proportions, a leap that could be a decade away.

The Meta Ray-Ban Display Glasses were a pleasant surprise. Their form factor is reasonable enough to wear without looking like a cyborg cosplayer, and the functionality, while limited to a single overlay display rather than full 3D augmented reality, hints at what’s coming. But we’re not there yet.

So where does that leave us?

The Hardware Already in Your Drawer

Instead of waiting for the all-in-one device of our dreams, why not make the most of what we already own? Consider this combination:

A smartwatch with always-on connectivity and its screen for quick glances. Wireless earbuds with built-in microphones for voice interaction. A clip-on camera like the Insta360 Go or similar wearable recorder for visual context.

Together, these three devices form an ideal multimodal AI interface: display, audio, video capture. You already own at least two of them.

The missing ingredient was never the hardware. It was the software that could unify them.

Imagine you’re attempting a tricky recipe, hands covered in flour, when your clip-on camera catches you reaching for the salt. OpenClaw interrupts through your earbuds: “Hold on, the caramel hasn’t reached the right color yet, give it another forty seconds.” A timer appears on your watch. As you wait, you ask aloud what temperature the oven should be. It answers, then adds: “By the way, you’re almost out of vanilla extract. Want me to add it to your grocery list?“ Done.

What makes this interesting is that OpenClaw already has most of the ingredients, it only needs custom skills to orchestrate across those 3 devices.

The Implications

Here’s what the Humane AI Pin missed: the next breakthrough in AI interfaces may not come from new hardware. It will come from software that finally makes our existing hardware better integrated.

This has several implications worth considering:

First, it democratizes access. A smartwatch, earbuds, and a simple cam are exponentially cheaper than a dedicated AI wearable. If the interface is software, the hardware becomes commoditized.

Second, it creates optionality. Don’t like your earbuds? Swap them. Prefer a different smartwatch? Fine. OpenClaw is agnostic, it works across platforms, operating systems, and messaging channels.

Third, it inverts the innovation dynamic. Instead of waiting for Apple or Meta to ship perfect glasses, thousands of developers are building skills and integrations for OpenClaw right now. The pace of iteration is accelerating weekly.

The ‘Device’ of the Future is Already Here

A year ago, I would have imagined this article ending with a speculative call to action: someone should build the software that ties these devices together. Now that software exists, and it’s open source.

The ideal multimodal AI gizmo isn’t a single device. It’s a constellation of devices you already own, unified by an agent that can quickly figure out how to orchestrate them in creative ways.

The Humane AI Pin tried to replace your phone. OpenClaw doesn’t replace anything, it recruits everything.

Over two decades ago, as a young enthusiast, I embarked on a peculiar quest. I joined millions worldwide on a journey that, despite never leaving our computer desks, sought answers from the farthest corners of the universe. It was the late nineties, and the search for extraterrestrial life was ongoing through various initiatives.

SETI@Home, a program that lets our computers contribute idle CPU time to analyze cosmic radio signals, was our "spaceship." Even though, in the end, my computer, or any other one, didn't discover green aliens, we became part of the largest distributed computing project in the world.

The initial ambition of the Berkley SETI Research Center was to get 50,000 to 100,000 participants. Instead, it has gathered over 5 million contributors. Fast forward to today, I found myself reminiscing about the SETI@Home days and how it had created a foundation that could soon transform the world of artificial intelligence and large language models (LLMs).

The Current State of AI and the Need for Democratization

By the end of last year, around the week ChatGPT was first released, I wrote a piece titled 'Democratizing the Exciting New Wave of AI' where I pondered the potential consequences of AI technology being controlled by a handful of tech giants. The goal was to explore ways of democratizing AI through open-sourcing and decentralization – making it accessible to all and enabling widespread participation in building new models.

Since then, the open-sourcing movement has flourished with several language models, such as Falcon by TII, MPT by Mosaic ML, and the recent Llama 2 by Meta. These models, ranging from 7 billion to 70 billion parameters, have abilities nearing those of GPT 3.5 while being free to use and customize. Despite this progress, a significant roadblock remains – running these LLMs requires high-end GPUs with large memory - an investment beyond the reach of most users. Additionally, training the large version of these models from scratch requires thousands of GPUs in an industrial-grade data center and tens of millions of dollars of investment.

Now, imagine the radical evolution that could occur if our everyday computers' idle processing power was harnessed into a massive, collective AI supercomputer. A distributed computer enabling anyone, anywhere, to run and fine-tune advanced AI models. This vision is much closer to reality today than just six months ago --it's precisely what PETALS has partly accomplished with their latest framework. It brings to mind the era of Napster and BitTorrent, but there's a crucial distinction: this endeavor is legal, open-source, and for the collective benefit.

Just as a torrent disperses file sharing across a network of users, PETALS segments each model into manageable blocks, spreading them across consumer-grade computers worldwide. This enables even modest devices with entry-level GPUs to contribute to the network, thus facilitating the execution of the largest AI models without cutting-edge infrastructure.

At this early stage, the technology is still limited to inference and fine-tuning a handful of LLMs. It doesn't yet allow the complete model training from scratch, and the inference performance is still below what's achievable with dedicated servers. However, it's a significant validation of the approach and a great showcase of the potential of what such a system could become in the near future.

PETALS' other remarkable accomplishment lies in its user-friendliness. Despite its complex, distributed architecture, using it is as straightforward as interacting with a centralized API like OpenAI. Here's a code example loading the largest new Llama 2 model and then using it to complete a short sentence:

Revolutionizing AI with Blockchain

Stepping into the next challenge, the successful operation of this distributed system hinges on a simple yet crucial factor: the willingness of individuals to donate their idle CPU/GPU processing power and to keep their machines running and connected to high-speed Internet around the clock. A certain level of mutual benefit exists — those who lend their resources can also use the collective AI supercomputer when needed.

However, not everyone with robust hardware needs to run or fine-tune AI models. Consider the gamers whose high-performance machines optimized for high resolution, high frame rate play are dormant outside the gaming sessions. The potential could extend even further if platforms like PlayStation 5 and Xbox Series X allowed the integration of such a system.

With its peer-to-peer sharing, BitTorrent provided a naturally balanced supply and demand network, sustained by the regular exchange of stored media files and bandwidth. Yet, this balance can't be assumed for an AI inference and fine-tuning distributed system, where the incentives for contributors and users may not organically align. This is where introducing a new mechanism — a blockchain and crypto token model — could marry perfectly with the structure PETALS has established. Imagine the appeal of earning crypto tokens – let's use the imaginary name 'NeuraCoin' — in exchange for providing GPU processing power, bandwidth, and electricity, essentially creating a new form of mining that doesn't suffer from the inefficiency and waste of Proof of Work systems. This system would allow users to employ these tokens to access the collective supercomputer or exchange them for other coins or fiat currencies. Users can also have the choice to either directly purchase ''NeuraCoins' to start using the supercomputer immediately or earn them over time by providing access to their idle processing capacity. Such a system does not require a brand-new Blockchain network; it can be built on top of existing ones like Cardano or Solana.

This kind of ecosystem's dynamics can lead to a more balanced demand and supply, provided certain conditions are met. A higher demand for AI computational resources could, theoretically, motivate individuals with idle processing power to contribute these resources in exchange for 'NeuraCoins,' thereby expanding the network's computational resources. On the other hand, an increase in supply could, under ideal conditions, lead to a decrease in the cost of computational resources due to competition among contributors. This dynamic could potentially make the AI-distributed supercomputer more accessible to users who need it.

The key factors influencing an individual's incentive to contribute to such a system primarily revolve around comparing Internet bandwidth and electricity costs against the potential earnings from tokens and their respective market value. This equation predominantly depends on geographical factors. A possible solution could be integrating these considerations into the token economy. For instance, users who contribute their processing power from regions with higher operational costs would receive a more significant token compensation. However, such as setup would effectively cause those utilizing the distributed capacity to pay a premium for accessing nodes from higher-cost locations. To optimize this dynamic, an algorithm could be implemented that initially allocates the most cost-effective nodes and incrementally taps into the pricier ones as network capacity reaches saturation. Such a prioritization mechanism could help create a more equitable, efficient, and scalable distributed AI ecosystem.

Huggingface, the popular hub for hosting and running AI models, may be ideally positioned to architect that kind of system. That is, provided it doesn't conflict with their prevailing centralized business model.

So, what could go wrong?

One of the most positive pieces I read about the topic is ‘Why AI Will Save the World’ by Marc Andreessen. Although I agree with many of the optimistic premises of the article, it is bordering on utopian idealism and needs a pinch of reality for balance. During an interview, Marc casually dismisses the threats posed by AI's potential dominion, assuming it would have to design and fabricate its own GPUs and servers independently.

The reality is that there's already plenty of cloud computing capacity accessible through APIs. It could allow an AI, via a few lines of generated code, to remotely instantiate virtual machines, transfer data, and ultimately replicate itself in a quasi-organic manner. But, given that cloud infrastructure is controlled by central private corporations with dedicated cybersecurity teams, one could reasonably assume that the risk is still relatively containable. 

On the other hand, let's consider the implications of constructing a vast, massively distributed computer system. Imagine countless machines of disparate security levels scattered across the globe, all linked together. We may unintentionally open Pandora's box for the ideal hardware platform that could enable the execution and spread of malicious AI code across the globe like a digital wildfire impossible to control.

In my view, this risk isn't merely hypothetical—it's a looming shadow we must take into account as we blueprint the architecture and inter-workings of such a supercomputer. It should be equipped with some sort of ‘red kill switch.’

It is a challenging feat to achieve, given the decentralized nature of the network, yet absolutely essential. Additionally, many governance questions remain open beyond the technical aspect: Who gets to press the emergency button? How many people should have this power? How do we pick them? What are the triggers for a 'shutdown decision'?

Final Thoughts

To close on an uplifting note. I still prefer to envision a world where we are not dependent on a handful of large centralized and closed AI providers, a world where our devices don't merely sit idle, awaiting our commands, but actively contribute to a massive global decentralized AI supercomputer. Every self-driving car at rest or hooked to a charging station, every gaming PC and console during its off-hours, every office desktop machine at night - all could potentially be powering beneficial AI operations, becoming integral to the fabric of our digital future.

Still, it is critical to seriously consider the risks and design protective mechanisms from the early versions of such a network. Therefore, we must deeply reflect on: How can we strike the delicate balance of aiming at decentralized operations while holding onto the thread of safety and control?

Share

Additional Reading

Meet Petals: AI That Can Run 100B+ Language Models At Home | Aneesh Tickoo

Petals: Collaborative Inference and Fine-tuning of Large Models | BigScience

Petals on GitHub | BigScience

Videos

Torrent for Running Huge GPT-LLMs | Prompt Engineering

The digital revolution has shaken up the media world like never before over the past few decades. It significantly changed how we consume news, music, and video. The shift began with the dematerialization of print and other tangible media then continued with the unbundling of articles from newspapers, songs from albums, and video from cable networks. This progression has been a rollercoaster ride for media companies, forcing them to reinvent their business model in a fully digital world. Yet, just when they thought they had it figured out, a new wave of turbulence is on the horizon, one that will be more profound and brutal than the first.

The rise of ChatGPT has been phenomenal, to say the least. The staggering estimate of 100 million users, just two months after its launch, testifies to the unparalleled level of interest. This kind of scale leaves little room for doubt:

A few exchanges are all it takes to realize - there's simply no going back. The direct responses to questions, presented in a pristine manner, feel like a breath of fresh air in a polluted web with SEO junk.

In the present world of written media, distribution has become highly dependent on standalone articles and the ability of their headlines to generate clicks. As a result, the search engine became a strong driving force behind media exploration. The value is no longer derived from the worth of an audience within a branded publication, but instead from the number of clicks each article could generate. As users increasingly turn to pre-trained AI chat tools for their queries, one may wonder: what will happen to the online publishing business?

There are essentially two types of search queries. The first type involves queries that seek direct answers or solutions to specific issues. The second category contains queries that aim to uncover source websites for more in-depth exploration and research. It may appear that the latter type is less impacted at the moment, as people tend to revert to traditional search methods to access additional content. But is this truly the case?

Despite its popularity, ChatGPT has a couple of key limitations. On the one hand, it's confined to the knowledge available before its training data cutoff date, which was at the end of 2021. On the other hand, it operates in an isolated environment from the Internet, which prevents it from accessing real-time updates and information.

That said, some companies have already begun offering solutions to overcome these limitations. Microsoft Bing, Neeva, and Perplexity.ai are a few of the players trying a combination of traditional search with AI-powered chat.

The current hybrid approaches may sound appealing at first. Still, they are just transition solutions as companies are trying to experiment with AI-powered search without destroying the existing business models. In other words, these crossbreed offerings are attempting to safeguard "the link" at the cost of a subpar user experience. Do you really want your search results to be a mishmash of both answers to your query and a wall of blue links? It's like a lifejacket for the status quo, sacrificing the smooth sailing of the user experience for a false sense of security.

Now, compare the screenshot above with this one:

As a starting point, which experience would you rather have?

What we see Microsoft and other players now launching is the typical initial attempts of trying to fit "the old" inside "the new." Looking back at the early iPad days, Apple went through the same process. For example, when they released Newsstand, they tried to put fancy PDF versions of paper magazines inside an App.

The skeuomorphic design imitating the appearance of a real-world stand looked cool, but the approach of creating isolated digital replicas of print magazines ultimately flopped.

Apple eventually realized they had to rebuild the concept from scratch and tailor it for the new digital reality. So, four years later, they replaced Newsstand with Apple News. Through a single subscription, users could access their favorite magazines with a far better experience that feels more native to the digital format. The App also offered a personalized "For You" feed featuring a blend of articles from various publications.

As we examine how the mixture of search and AI chat is done in the present solutions, it appears we may be having a "Newsstand moment."

I believe, there are far superior ways to implement a combination of traditional search with AI chat. What we have seen so far feels like an unappetizing cocktail with ingredients not blending together.

The No-Web Future

Imagine a future where chat-style interfaces are at the forefront. No more scrolling through endless results bloated with ads. Instead, these interfaces would abstract away the external search by distilling the information from results down to its core, then presenting it in an easy-to-digest form inside an uncluttered chat interface.

Here's how it could work:

Based on the user's query, the system will determine whether to utilize its AI language model or a traditional search engine. If the pre-trained AI model can answer the question, it'll operate similarly to the current ChatGPT. But if the inquiry requires an external search, such as "What's the latest news from the elections?"the system will tap into a standard search engine API. However, rather than polluting the user experience with a wall of blue links, it'll first send them to the AI language model for further analysis and summarization before presenting the distilled information to the end user. This approach would offer the best of both worlds, providing a refined experience without sacrificing quality and accuracy.

Now, here's the dilemma. Even though such a service will be best for consumers, it will also have a devastating impact on the publishing industry. It will dismantle "the link," ultimately causing a drop in traffic to the source publications, leading to a decrease in display ads and a drying up of the financial lifeline for content creators.

Looking at the decline in newspaper ad revenue over the past two decades, we see that even as the industry has shifted to digital ads, it remains a far cry from its peak in 2000. The rise of Google, and later Facebook, has dramatically altered the ad market by creating intent-based and personalized targeting, allowing brands to reach their audiences with greater precision.

It's an ironic twist of fate that the very tech giants that stole the advertising business from traditional publications have also become their saving grace by driving traffic and keeping them afloat. Imagine the magnitude of the revenue drop if newspapers didn't have the traffic from Google. It appears the moment has arrived where we'll witness this scenario starting to play out.

Even though traditional pages may still be around in the future, they will significantly decrease in relevance. Chat-based interfaces will curate and present information in a personalized, conversational format. Instead of browsing pages, users will spend more time in Q&A mode, and we'll progressively move to a No-Web world.

Monetizing Content in a No-Web World

In that new reality, publishers will have to solve the puzzle of how to remain relevant, keep their content in front of their audience, and generate revenue. The question on everyone's mind is:

The harsh truth is that a mere mention at the end of a chat session is unlikely to drive a level of traffic anywhere close to a top Google search result. If users can already get what they need, why bother clicking on that link?

At this stage, it is hard to predict exactly how the ecosystem will evolve but we can explore a couple of potential developments that might occur in combination.

People consume content in two distinct ways: "pull mode" and "push mode." In "pull mode," individuals actively hunt for information using keywords and search engines. This is the mode that AI-powered front-end chat interfaces will ultimately disrupt. In "push mode," individuals passively receive information, relying on the suggestions of curation services such as Apple News, Flipboard, and, more recently, Artifact. These personalized feeds offer a tailored experience, and publishers have the opportunity to double down on this mode by removing their content from the open web and making it available only through a subscription.

It is essential though that a large number of publications are accessible through a single subscription for the end user. That's the model that has saved the Music industry.

As mp3s dematerialized CDs, sales collapsed. A glimmer of hope emerged with legal download services such as iTunes. Still, it wasn't until music streaming offered the entire catalog with a single subscription that the industry could reclaim its former glory. Music has now jumped back to its peak level of revenue1, last seen in the late 1990s.

The alliances with the news aggregation services outlined earlier play a vital role in the "push mode." Nevertheless, for publishers to truly succeed in the new No-Web landscape, they must also figure out a strategy to prevent the "pull mode" from being eaten by AI-powered chatbots.

Consider this perspective: What if AI-powered clients were actually a golden opportunity waiting to be leveraged to great effect rather than a looming danger to be feared?

Instead of standing still and watching their traffic and ad revenue dry up, publishers have a unique opportunity to forge lucrative B2B deals with tech giants like Microsoft and Google. By removing their pages from the public web and granting special access to their prized content to these AI chat clients, publishers can tap into a new revenue stream without having to sell to millions of individual consumers. This could be a far more effective monetization strategy, as they would only need to secure a limited number of large deals with select tech companies. The responsibility for charging end users with a bundled subscription will then fall upon the search and AI providers.

Today, those kinds of arrangements are not feasible. Search engines have no incentives to pay. The ongoing deal is "free traffic" in exchange for "free content." But if "free traffic" were to disappear, something else would need to be offered as compensation for content.

Final Thoughts

The dismantling of the open web publishing business model, as we know it, is most likely inevitable. What remains unclear is the speed at which it will happen and what the final outcome will look like.

In this article, the focus was explicitly on written digital content. It is essential to highlight, though, that the shift we're seeing towards AI chatbots will not be just limited to the realm of publishing. It's a change that's set to ripple across the entire web, from booking and retail to ticketing. Websites currently serving as these services' hubs will progressively migrate towards API-based platforms that AI chat clients will tap into. Yet, the distinction between this scenario and content publishing is that a drop in traffic to these services' traditional websites won't at all affect their business models and revenue streams. For instance, if a ticketing platform generates revenue through commissions, it doesn't matter if the order originates from a website or an AI assistant. To these services, AI chatbots will only serve as another channel, providing consumers with more options and versatility.

If we try to guess how things may end up at a broader level, it's possible that the Internet will undergo a radical transformation. It might emerge as a more invisible version of its current form. The bulk of websites and apps that we use today may be deconstructed down to the two components below:

The seemingly innocuous Spotlight Search and Siri are often ridiculed as they cannot do much. Spotlight is primarily good at in-device search, and Siri is mostly good at setting timers. But what if these unremarkable tools were armed with the power of a GPT-like AI backend, capable of distilling results from full-fledged search and accessing a multitude of API-based services? The ramifications would be immense, as billions of iOS devices around the globe would suddenly gain the ability to open the tap on the new No-Web reality and, at the same time, switch off the Google-dominated world as we know it2.

Share

The death of Google at the hands of ChatGPT may be a popular narrative, but the truth is far more nuanced. The destabilization of the Google’s Search empire has been brewing for years, and the emergence of language models represents just the latest development in a complex web of trends. In this post, we’ll first uncover the 9 forces putting pressure on Google’s Search reign, then speculate on the 9 strategic moves Google could make to navigate these turbulent waters.

My frustration with the search quality started somewhere between 2017 and 2018. The returned links were not strictly following the queries anymore. Instead, they would be a couple of steps broader in scope. Initially, I thought it was probably some recent algorithm experimental updates that weren't working out as intended; Google would eventually correct the problem. But things kept getting worse every year.

It was refreshing to see the now-famous article by Dmitri “Google Search Is Dying” and the intense level of engagement around it. I was far from being the only one.

It’s also worth reading the full twitter thread from Michael Seibel:

In short, why did Google Search results become so bad for some users?

SEO Exploitation

In the never-ending game of cat and mouse, SEO marketers are constantly trying to crack the code and rise to the top of the page, while Google continuously updates its algorithm to keep low-quality content at bay. By now, it appears the SEO gang has won.

The SEO aspect has two sides. On one side, you have the legitimate goal of boosting quality content or products to gain visibility and attract potential customers. But on the flip side, there's the overwhelming spread of spam, where the only intention is to generate income through AdSense placements by tricking Google into thinking it's relevant content. Sadly, Google has a vested interest in this problem as it profits from ads displayed on these fraudulent pages.

This situation leads to a weird place where the outcome is bad for the user experience but pretty good for Google's business. When incentives are not aligned, things go downhill.

Ads Escalation

The pressure of constant growth requirements led the company to keep increasing the number of sponsored links. As a result, it now reached a point where, for some keywords, the first page looks like an overstuffed sandwich made of ads with only a few crumbs of organic content.

Unsolicited “Smart Search”

This is well put by Dmitri Brereton from the same post:

Google increasingly does not give you the results for what you typed in. It tries to be “smart” and figure out what you “really meant”, in addition to personalizing things for you. If you really meant exactly what you typed, then all bets are off.

Even the exact match query operator (“ ”) doesn’t give exact matches anymore, which is quite bizarre.

For the query operator (" ") issue, Google later responded, arguing this is incorrect and the operator still serves its purpose.

My annoying experience with (" ") usage is not that it changes the query, but it's that it often leads to the surprising: “Your search did not match any documents”, like in this example:

While it might be true, I'm having a hard time believing that the entire Internet doesn't have even a single page with that exact sequence.

The other forces squeezing Google Search

So far, we looked at:

1. Search quality degradation

Force #1, that may impact Google's dominance in the future.

There are, however, 8 other dynamics at play. Some are already taking place, and others are likely to emerge in the future:

2. Amazon Search

More and more people are bypassing Google and taking the express lane straight to Amazon to search for product information. It is estimated that 61% of product searches now start on Amazon. This shift in behavior is due to a couple of reasons. First, the issue discussed earlier of the declining reliability of Google's results. Second, the efficiency and convenience of Amazon's one-stop shop for comparing prices, reading reviews, and making purchases.

Interestingly, Amazon started its sponsored search ad business about ten years ago. It reached an annual revenue of $38 billion and grew by 19% from the previous year. In comparison, Google Search's revenue in Q4 2022 dropped 2% from the prior year.

3. Instagram and TikTok Search

For the younger generation, seeking out the ultimate eatery isn't a matter of checking Google Maps. Instead, they turn to TikTok and Instagram to discover visually appealing dishes. Likewise, they don't bother with a Google Search when it comes to finding a cosmetics product review. Instead, they watch videos of the product in action from people they can relate to directly on their favorite social media apps.

TikTok and Instagram typically work best for leisure-related content, as users seek authentic observations from people just like themselves rather than sterile search engine-optimized results. However, these platforms also have the challenge of influencer-manipulated popularity, but that's a topic for another day.

For now, Instagram and TikTok are eating into Google’s core products as covered by TechCrunch. It’s worth reading the entire piece.

4. Reddit

This is best summarized with this extract from Dmitri’s original post:

Reddit is currently the most popular search engine. The only people who don’t know that are the team at Reddit, who can’t be bothered to build a decent search interface. So instead, we resort to using Google, and appending the word “reddit” to the end of our queries.

And

Why are people searching Reddit specifically? The short answer is that Google search results are clearly dying. The long answer is that most of the web has become too inauthentic to trust.

5. Microsoft Bing

I'd give Bing a shot in the past only when Google's answers left me feeling irritated. Yet, over the years, it began to bear a striking resemblance to the search giant in both appearance and functionality.

The problem with Bing was not that it was a bad product but that it was very undifferentiated. What was the point of using it?

Even after nearly a decade and a half, their global market share is only at about 3.5%.

But

The dynamics changed in 2022 with the unexpected massive traction of ChatGPT, and the announcement of a $10 billion investment by Microsoft in OpenAI. The speculation that Microsoft would integrate the technology into Bing became rampant and finally materialized, perhaps sooner than people expected. It was officially launched at an an event on February 7th.

The hybrid search and chat approach combined with the seamless integration with the Edge browser creates a seemingly compelling experience. Here's a review in less than 10min.

As a response, Google rushed to schedule its similar event the day after, showcasing their chatbot Brad, I mean Bard. The demo was somewhat underwhelming, and the episode was tarnished when the results returned an incorrect answer.

The reactions were quick to follow:

To be fair, disseminating false information by language model-based bots is not a unique problem to Google. But if such a large corporation is set to embark on a high-stakes announcement, it's baffling that no one bothered to do a basic fact check. Maybe by running a Google search?

The funny thing is that even the Microsoft demo was riddled with inaccuracies, but no one noticed until a recent post, “Bing AI can’t Be Trusted” exposed all the glitches. So, it turns out the Microsoft team is also guilty of neglecting to run manual checks on their fancy demo.

But regardless of the drawbacks inherent to LLMs, it appears that Bing is set on a new differentiated and experimental trajectory that makes it at least more interesting than Google for the time being.

Microsoft's advantage does not lie in technology since Google can develop something comparable, and both companies will face the same LLMs drawbacks. The main edge is that the potential business risk for Microsoft is much lower. They won't shed a tear even if they lose half of the ad revenue.

In 2021, Google ad revenue (including YouTube) represented about 80% of the total. For Microsoft, Bing amounted to only 5%. Being the underdog gives Microsoft a safer playground to experiment more aggressively with self-disruption.

For now, the stock market is mirroring that perception. Google's stock price took a hit after the Bard error, but Microsoft is still unaffected even after the Bing Chat meltdowns keep getting more hilarious.

6. Apple

It's not a secret anymore that Google pays Apple a hefty amount to keep its search engine as the default in Safari. The amount is not officially disclosed, but analysts estimated the payment to be $15 billion in 2021 and to reach $18 to $20 billion in 2022. Google Search ad revenue was $162 billion that year, not including YouTube and AdSense.

Let's envision a hypothetical low-case scenario where the current deal between Google and Apple is modeled around a 'virtual revenue share' with a 55/45 split. In this case, the $18 billion allegedly paid to Apple would translate to approximately $40 billion worth of search ad revenue for Google, roughly 25% of their total search ad revenue [1]. That's a substantial chunk of the market, and it's all controlled by one company with the power to shut it down at a moment's notice [2].

Now, why would Apple terminate that deal? It's lucrative even for a company at the scale of Apple.

There are a couple of reasons why it could happen. First, Apple is marketing itself as a privacy-centric company. Sending search traffic to Google doesn't fit well with the privacy narrative. Second, Apple actually has its own search engine, Spotlight. It is mainly known for local device search, but it has discretely evolved since iOS 15 by showing more and more external web results with direct links and images, bypassing Google. The interesting thing about Spotlight is that it's a browser-less experience tightly integrated with the operating system.

Apple also has Siri, which has significantly improved speech recognition and synthesis but is still totally useless for natural language processing. However, it is inevitable for Apple to build its GPT-like language model at some point, empowering Siri to finally fulfill its original role as a smart virtual assistant. When that happens, the marriage of Siri and Spotlight will be like a power couple, offering a similar or better experience than what's currently available but entirely outside the browser.

Moreover, Apple has been expanding its services portfolio with offerings such as TV+, Fitness+, Music, Arcade, and iCloud. The company's next logical step could be introducing an LLM-based premium Siri+ with a monthly subscription. Given the billion+ installed base of iOS devices, Apple should have no trouble compensating for and surpassing any lost revenue from the termination of its deal with Google.

7. Other search engines

DuckDuckGo, the privacy-focused search engine, got trendy at some point but is still at only ~0.6% global market share.

Newcomers such as Neeva, co-founded by a Google veteran, are more interesting, though. Neeva is not only more private but is ad-free. In addition, they offer subscription plans with unlimited search and additional privacy with a bundled VPN. Neeva has also introduced NeevaAI, a hybrid search and language model engine, addressing the two common criticisms of ChatGPT - the lack of sources and blindness to post-2021 information-.

The advantage of their business model lies in their perfectly aligned incentives. Neeva gets paid by the users they actually serve instead of third-party advertisers.

Perplexity.ai, another newcomer with a similar approach, is also worth keeping an eye on.

8. ChatGPT and other language models

By now, it is clear that ChatGPT has taken the throne as the superior way to address certain queries, despite its known imperfections. One of the critical aspects OpenAI got right was the early positioning. ChatGPT was introduced as an experimental research release, which made users much more tolerant of its faults.

It only takes a handful of experiences for anyone to be hit with the strong sentiment that:

It's such a relief to see the unpolluted direct responses to your queries nicely structured. Also, the ability to ask follow-up questions sets the chatbot apart from conventional search engines.

The sentiment has been widely echoed in the tech scene.

The reported 100 million users two months after the launch is an unprecedented uptake and a confirmation that the interest is far beyond the typical early adopters and tech enthusiasts.

What remains to be seen is:

The search queries are broadly of two kinds. First, the questions that seek direct answers or solutions to specific problems. Second, the requests that intend to look for the source websites for further navigation. The latter type is relatively immune for now. The Google revenue risk will be determined by the ratio between these two types of queries. That number is not publicly disclosed, but Google does for sure have the data. Given their recent panic reaction, we could speculate that the first kind of request represents a significant portion.

Besides, ChatGPT and other chat-based language models represent just the first wave that still relies on centrally run engines.

There's an inevitable second wave that will arise at some point where language models can run locally on devices. This evolution will mirror the trajectory of AI image generation. At first, server-side processing was necessary, as seen with services like DALL-E. But the release of Stable Diffusion brought a more compact model with comparable abilities that could run on a consumer-grade GPU. For instance, the Stable Diffusion optimized version for Apple silicon can work on an iPad, producing a photo in about 40 seconds.

For now, language models are still highly demanding in computing resources. Still, the curves of the increasing power of consumer chips and declining model sizes will likely intersect in a few years. At that point, Google's search competitors will no longer be restricted to Microsoft and a handful of major corporations. Instead, every consumer device could run its own pre-trained ChatGPT-like version and bypass Google search.

9. The Justice Department

A few weeks ago, the Justice Department dropped the hammer on Google, suing the tech giant for monopolizing digital advertising technologies.

Google is accused of violating antitrust laws and stifling competition through its dominance in the 'ad tech stack.'  The legal action is like a storm cloud hanging over the company, casting uncertainty and doubt.

It's unclear if this lawsuit has any real shot at taking down Google ad empire, but it's a burden they could certainly do without at this moment.

What should Google do?

Let's now strap on our optimistic caps and dive into some speculative reflections on the moves Google could make to break free from this tangled web.

Despite the challenges, I am optimistic about Google's future. The company has a talented workforce, a massive cloud computing infrastructure, a mastery of the latest AI technologies, owns DeepMind, and let's not forget:

First, let's address the elephant in the room. This is what many are quietly thinking, but few are loudly talking about.

1. CEO change

Seven years is a respectable tenure, and I do not mean to undermine Sundar's achievements in any way. However, the present circumstances are shifting the landscape in an unprecedented manner, and it seems like a good time to consider a leadership transition.

If Steve Balmer was still running Microsoft, it is unlikely that the company would be where it is today.

Om Malik lately wrote Does Google need a new CEO?:

Google’s board, including the founders, must ask: is Pichai the right guy to run the company, or is it time for Sundar to go? Does the company need a more offense minded CEO? Someone who is not satisfied with status quo, and willing to break some eggs?

2. Double down on the Search API

Google already has a search API called Programable Search Engine. It allows developers to initiate Google search requests and integrate the results into their apps and websites.

Although it's unknown how much revenue this service generates, it could become more valuable than ever with the rising popularity of LLMs, which will still require real-time internet search to be fully viable. As a result, there may be an opportunity to significantly increase the price of this service if the demand meaningfully increases.

3. Get serious about YouTube Premium

YouTube is a goldmine of content. Sadly, it is hindered by its current recommendation engine. The algorithm is tailored to maximize attention by serving videos likely to attract views and serve ads. The approach works well financially for free users, with YouTube ads revenue reaching $29 billion.

There's, however, an untapped segment of customers that:

• Don’t like to see ads

• Ready to pay more if they could control what they see

On the positive side, Google's already launched YouTube Premium. I gave it a shot, and it's a fantastic relief not to be bombarded with advertisements. Unfortunately, it failed to meet the second requirement - the algorithm remained identical to the free version. As a result, I was still inundated with irrelevant videos of little interest.

The opportunity here lies in building a separate recommendation engine for Premium users. For example, the alternative algorithm could incorporate a natural language prompt to provide a customizable recommendation page that genuinely resonates with users' video preferences.

Another interesting feature would be the option to generate custom playlists with a prompt. For example:

create a playlist with the best educational videos about <topic> that are no longer than <x> min.

If Google greatly improves its YouTube Premium value proposition, it might see significant growth in paid subscribers.

4. Launch Bard Premium (B2C)

So far, both Microsoft and Google have opted for integrating their AI chat features within the existing search engines for free and rely on existing ad revenue.

Similarly to YouTube, there might be space for a paid version of Bard that won't be polluted with ads. But to captivate users, the offering must be beefed beyond mere ad elimination.

ChatGPT is currently running the experiment with the plus version at $20 a month. However, the feature set is basic for the time being. It only offers faster priority access during congestion and a sneak peek at upcoming functionalities.

In essence, it would be wise for Google to consider launching a standalone bard.com in Beta and refraining from altering the classic search on google.com. Depending on the prompt, the new experience could abstract the search aspect by having the engine decide whether to use the LLM or the classic search. When a traditional search is required, the LLM can still be utilized for structural organization and summarization before presenting the results to the user. This kind of flow would offer a refined and consistent user experience and be more differentiated from Microsoft and ChatGPT.

This shift would change the mission from:

Old Google

To:

New Google

5. Explore BB: Bard for Business

The opportunity to create tailor-made language models for various industries, such as legal, healthcare, education, customer service, retail, media, gaming, government, and management consulting, is ripe for the taking.

The current LLMs may not be ready for prime-time deployment in professional environments as they were trained on publicly available Internet data. Nevertheless, it is highly encouraging to see that they have achieved their current performance level without using private, professional data, which indicates their enormous potential.

There are three key advantages of a potential B2B offering. First, it opens up the possibility for Google to set a value-based Enterprise price level. Second, the risk of hallucination can be better contained when the output isn't meant for the end user. The respective domain experts can vet the results and take accountability before deliverables go to the end customers. And last but not least, the models should, in theory, hallucinate much less as they would have been trained on quality domain-specific data.

6. Acquire Reddit

Conde Nast currently owns Reddit, and there's no hint of it being up for sale. Yet, while Google is still swimming in a sea of cash, this could be a good time for them to make an irresistible offer.

If Google were to acquire Reddit, it would own a more authentic version of the Internet - one untouched by the poison of SEO. In turn, Reddit would gain from a tech company's expertise to rejuvenate its user interface and internal search capabilities. But should Google succumb to the evil temptation of aggressively monetizing Reddit through more ads, it would ruin everything. Instead, the objective should be to achieve Reddit's growth with a non-ad-based revenue model.

Would I be happy as a user to see that happening? Definitely not. But if I put on the Google hat, it might be a move worth considering.

7. Go all-in with Google Workspace

I always thought that the former “Google Apps” that became “G Suite” and now “Google Workspace” was a niche alternative mostly used by startups and some SMEs. So I was stunned to discover that they surpassed Microsoft Office 365 in market share, according to statista.com.

Assuming the data source is credible, it's an achievement deserving a standing ovation, especially since the product suite was launched in 2006. it's like a young challenger stepping into the ring against the decades-long dominance of Microsoft Office and beating it.

“Bard for Business”, covered in point #5, and this are like a match made in heaven.

Imagine a future where Google's app suite is infused with AI capabilities beyond the basic generation of email replies and text completion. If Google can invest in research around specialized language models, they can take Google Docs, Sheets, and Slides to new heights by creating industry-specific versions powered by foundation models trained on the unique data of each respective field. This approach would open up the possibility of further customizing these tools by offering add-on services that allow customers to fine-tune the models using their own private enterprise data. The result would be a set of truly unique tools for each business, serving as a virtual extension of their own team.

Google would then have the opportunity to command much higher pricing for those unique versions.

8.  Fix Google Cloud business.

Google Cloud is still the new kid on the block, as they were late to the party. They currently sit at about 10% market share.

Still, it generated $26 billion in revenue in 2022 and grew by ~37% from the previous year. The growth figure is an encouraging signal, but it's still bleeding cash.

The choice to place growth above profits may have been a deliberate act, at least according to Google's official narrative. Yet, some believe there might be a murkier root cause of why they struggle to achieve profitability.

If you run a medium margin business like a low margin business, you’ll have a culture that’s oddly frugal.

If you run a medium margin business like a high margin business, you’ll have a business that loses money forever.  – Zack Kanter

The view might be radical, but it cannot be denied that getting used to the juicy and “easy” pay-per-click ad margins molds a particular corporate identity and culture that totally contrasts with Amazon's upbringing with razor-thin retail margins.

Despite the underlying causes, if Google is using profits from ads to fund the Cloud business, it may not be a terrible strategy for now. But given the uncertain future of the ads business, they should aggressively set themselves to flip the position at some point:

9.  Step up the hardware game

Google dipped its toes into the hardware arena in 2010 with the release of its Nexus One phone in partnership with HTC. Since then, they've expanded with a range of products, including Pixel phones, smartwatches, earbuds, doorbells, thermostats, home assistants, and security cameras.

Till now, Google keeps its hardware earnings under wraps, bundled within the broad category of "Google Other," which brought in a hefty $29 billion in 2022. Yet, the best opportunity may not be in consumer hardware, as there are way too many competitors driving the margin down.

Google also has a differentiation challenge. Everyone has access to the same Android and Wear OS software. Their move to open-source Android made strategic sense back then to protect the search traffic on mobile. But now, it dilutes the competitive advantage of their phones.

It is high time for them to consider venturing into the chip market. They have already introduced their TPUs, designed for optimized machine learning, in 2016, but they have decided to keep them exclusive to their Cloud platform. It is worth reevaluating if that decision is still optimal in the current environment.

If we look at NVIDIA's revenue evolution, there's a major demand shift occurring in 2021-2022.

Historically, the company business was primarily driven by 3D graphics professionals and gaming enthusiasts. However, the recent exploding demand is driven by AI computing requirements. As a result, NVIDIA has now rebranded as a “World Leader in Artificial Intelligence Computing.”

Their gross margin was 57.8% in the past 12 months, which is not too shabby even by greedy advertising standards.

The obvious move could be for Google to try to buy NVIDIA, but that is very unlikely as NVIDIA has high ambitions of its own. Two years ago, they attempted to acquire ARM from Softbank. The deal didn't go through, though, due to regulatory restrictions.

So, if Google wants to get in the game, it will have to build its own lineup of AI-focused hardware, leveraging the power of its TPUs and potentially new ASICs. This move could ignite a fresh revenue stream that would bring in much higher margins than doorbells.

Final Thoughts

Google is strongly feeling the heat from competition, and trying to keep up with the rapidly evolving tech landscape. Their heavy dependence on search ads puts them in a vulnerable position.

For years, they focused on "moon shots" [3] hoping to land on a game-changing breakthrough, but they've come up empty-handed. So maybe, just maybe, they don't need to reach for the stars after all. The building blocks for a sustainable alternative to the ads business are right in front of them. Their addiction to high-revenue, high-margin ads may have clouded their vision, but the answer may be simpler than they think.

Share

Notes

[1] This does not account for any potential premium that Google may offer as a compensation for the sole purpose of stifling competition.

[2] Not considering any possible contractual obligations for a specified duration.

[3] Self Driving Cars, Project Loon, Google Fiber, Project Ara, Google X, Google Health, Google Glass, Project Wing, … Since the restructuring, some projects have been moved under Alphabet Inc.

Additional Reading

• Future of Search: A candid conversation with Sridhar Ramaswmy, an ex-Googler | Om Malik

• From Bing to Sydney | Ben Thompson

• How much Google should worry about ChatGPT? | Mario Gabriele

• The maze is in the mouse | Praveen Seshadri

• How to lose a monopoly | Benedict Evans

• Resetting the score | Benedict Evans

• Zero-Clicks Study | Marcus Tober

• Google exec suggests Instagram and TikTok are eating into Google’s core | Sarah Perez

When Apple revealed the first watch, there was great excitement surrounding its potential as a versatile mobile platform on the wrist. Developers eagerly flocked to create many apps and games, but the adoption was disappointing. The watch was not only slow, but people soon realized that shrinking a phone's user interface to a small wrist-worn screen had significant drawbacks. The touch gestures that were smooth on mobiles became cumbersome. The initial launch was a humbling reminder that the smartwatch was not a miniaturized phone. Apple subsequently repositioned the device as a health and fitness companion.

When the cellular version was launched with the Series 3 model, the possibility of using it without a phone finally became a reality. The live demo of a call during the keynote sparked a wave of excitement.

Despite the initial enthusiasm, those who opted for the cellular version used the connection mainly as a backup to stay plugged while going for a workout. The cellular mode drained the battery quickly, and the interface was still impractical for most other uses.

The recent introduction of the newest Apple Watch Ultra version opened up some exciting possibilities, though. With a brighter and larger screen, longest battery life, standard cellular connectivity, louder speakers, higher quality microphones, dedicated action button, the Ultra raises a tempting prospect: what if we ditched our phone for a smartwatch?

While still equipped with a display, a watch is smaller and less intrusive. It's a subtle antidote to the constant information overload in today's digital age. It rests elegantly on the wrist, liberating the hands to engage in other pursuits while keeping important information within reach. Although it may not be the definitive cure for phone addiction, it is a step in the right direction. It's a way to disconnect to connect more with people around. It's the perfect balance of being unplugged without being out of touch.

It's fascinating that the intention of Apple with the first version was to create a multi-purpose wrist-device, but it primarily became a health and fitness tracker. The Ultra version, however, was explicitly designed for serious sports enthusiasts. Yet, it may ultimately surprise us and become the one that actually serves as a multi-purpose device.

The Experiment

To explore the boundaries of the user interface of the watch as it ventures into the realm of mobile phone functionality, I decided to embark on what some may consider a heroic experiment:

The first few times when I intentionally left my phone at home, I frequently found myself reflexively checking my pockets, assuming I had forgotten it. It was a brief moment of confusion before realizing that the phone was left behind on purpose. Even as I walked away from my parked car, I couldn't shake the nagging feeling that I had left it inside. The mental habit was hard to break.

But the freedom of not having to constantly carry the phone or worry about it getting in the way while I was on the go was far more liberating than I could have ever imagined. It was particularly noticeable during my workouts.

What follows is not a product review but rather an exploration of how the seven sins of smartphone usage manifested on the watch during my experiment.

Let's dig in.

1.     Communication

Phone calls are incredibly effective, and feature voice commands for added convenience. The sensation is reminiscent of a futuristic fantasy. When paired with AirPods, the audio quality is best. Responding to texts is effortless with speech-to-text. If you're on the move and unable to glance at your screen, the watch can read messages aloud to you.

Short email handling is similar to instant messaging but reading and responding to longer ones is not practical. As a visual person, I want to have the ability to view large parts of a long email at once. Additionally, with the widespread prevalence of HTML formatted emails, the viewing experience is severely compromised. The solution was to resist sending responses until accessing a standard computer. The other major drawback is that while Siri can read text messages, it can't do it for emails.

Zoom does not currently offer a watch version. As a result, the only option for joining a conference is through a traditional phone call to one of the bridges. Even if direct Zoom links were available, they would only be relevant for voice interactions. The screen isn't large enough for shared presentations and video feeds. It would be like trying to catch a glimpse of the performance through a keyhole.

2.     Productivity

Jotting down notes while on the move proved to be quite usable. Interestingly, Apple doesn't include the iOS standard Notes app. Drafts and Bear are good alternatives that do provide a watch version. Dictation is easy and highly accurate whether using AirPods or not. The only downside is the lack of text-to-speech functionality, making it a hassle to review your notes. Asking Siri for assistance results in the disappointing response, 'Sorry, I can't read that on the Apple Watch.' So, taking notes is smooth, reviewing and editing, not so much.

When it comes to keeping track of to-dos, I have been using Things for many years. The minimalistic layout helps to quickly scan through the daily tasks and mark them as completed or reschedule them with just a couple of taps. Adding new tasks is quick with dictation. For what's needed to operate while away from the computer, the current handling of tasks is drama free.

3.     Content Consumption

Listening to audio-based content, such as podcasts, audiobooks, and music, is smooth and effortless. Curating and updating playlists is not quite as seamless, though. You can 'like' a song and add it to a list, but for more in-depth exploration and editing, you'll still need the big guns -- your phone or computer.

The web browser app is not explicitly included but the functionality is secretly there. For example, if you click on a web link inside an email, the related page will open up in a tiny browser. You can also directly open a website with a voice command. For example: 'Open cnn.com'. However, the page readability feels like trying to decipher ancient hieroglyphics.

I was astonished to discover that the Apple News app lacked any form of audio narration. Instead, it presents a short-written overview and an image for each article. The test was underwhelming but also amusing. For example, if you ask Siri 'Read the story' it won't read the current top news. Instead, it will recite a random story like ‘Once there were two mice who became friends at mouse college…’

Twitter, Flipboard, Instagram, and TikTok are all out of reach. But in this context, it's kind of the whole point. It's a feature and not a bug.

4.     Transport and Navigation

Since the initial version, Uber was among the first companies to make the leap on the watch. Sadly, they discontinued the app in early 2022. The watch edition had limited features. It didn't support pooling or contacting the driver, which most likely had made it non-viable for many users. Lyft ended its support even earlier in 2018. So, for now, no ride-hailing if you ditch your phone.

Walking navigation works nicely but when it comes to driving, the experience falls short. The voice and minimalistic display directions are not detailed enough for a big city with complex roadways and multiple lanes. To render driving navigation viable, adding CarPlay support is crucial. It would also make driving safer.

The lack of food delivery apps is not an issue, as they are primarily useful in a domestic or office setting.

5.     Photo and Video

With the Photos app, you can scroll through albums. It is not the best way to look at pictures, but a pleasant experience is revealed when used as a Watch Face. It displays a gallery of memories on your wrist, like a personal museum. Each time you raise your arm, it unveils a new piece of nostalgia.

If you're at a social event and the urge to capture a memory strikes, you're out of luck. The absence of the camera feels like a missing limb. A workaround could be to carry a dedicated digital cam, like in the good old days.

Unsurprisingly, the official YouTube app is not supported, but there is an alternative called WatchTube. Is it a sensible use of the watch? Definitely not, but it does demonstrate the video's potential.

6.     Payments

No surprise here. Using wireless payments feels like a natural fit for such a device. With a simple tap of the wrist, transactions can be made in a few seconds without having to pull a wallet.

7.     Scanning QR Codes

The consequence of the shift to digital menus at restaurants, as a result of COVID, was something I hadn't considered. Many establishments never resumed providing print menus. Using QR codes on a watch is unfeasible not only due to the lack of a camera but also because of the screen size. Viewing a standard menu layout would be highly uncomfortable.

The challenge is temporary, though, as most dining places will return to paper or tablet menus at some point.

The user interface evolution that could alleviate the form factor limitations

The results of my experiment revealed that certain usability limitations are inherent to the form factor. Some functionalities, such as following a shared presentation during conference calls or browsing long-form content, may never be entirely solvable. Still, many other experiences could be significantly improved with a re-envisioned user interface.

What's hindering the watch's potential is that the operating system is built with touch and display as the primary means of interaction. The OS is designed as if it were a tiny mobile phone around the wrist. Although voice interactions are supported through Siri, they are limited to basic commands such as dictation, scheduling reminders, setting timers and alarms, and launching individual apps.

For the watch to become truly versatile and possibly replace the phone, it should completely revamp the interaction model. It should holistically operate with natural language through a voice user interface. In other words:

The emergence of powerful large language models (LLMs) has opened up tremendous possibilities. Imagine each application having its own natural language conversational interface, where the OS acts as a central hub for processing general user prompts and seamlessly directing them to the appropriate apps. The replies would be in an interactive conversational manner with the ability to consolidate and retain information from previous interactions. A voice conversational user interface (CUI) baked into the OS would propel the watch's usability to a new level in specific scenarios.

To evaluate what kind of domains would benefit from such an evolution, let's plot the primary type of apps on a 2-dimensional graph with 'current user experience' on one axis and 'potential user experience enhancement with a CUI' on the other.

The blue rectangle shows the areas that would experience significant improvements by implementing a voice-based conversational user interface (CUI). These domains represent the functionalities that could reach an optimal state at some point in the future. These are the kind of apps that Apple and other device makers must strive to perfect on the smartwatch.

This meme encapsulates nicely the potential trajectory of email:

An advanced CUI doesn't have to be limited to input and outputs in a conversational manner. It could also be coupled with a summarization and auto-reply engine.

For instance, it could handle requests such as:

What important emails from the last hour should get my attention?

Draft a reply to Joe in a formal business style in a maximum of 200 words and read it back to me.

For news, the UI could offer the possibility of asking:

Create and read back a news briefing about what happened yesterday in technology and local politics in less than 3 minutes.

With those abilities, many limitations imposed by a small screen would become insignificant.

The red quadrant serves as a reality check revealing the inherently unsuitable scenarios to run on a watch, regardless of how sophisticated the user interface becomes. These should be left for computers, tablets, smartphones, and perhaps augmented reality glasses. Trying to run them on a wrist-device would always result in a bad experience.

The green quadrant includes the domains that presently have a good user experience on the watch and wouldn't benefit much from a future conversational user interface.

Final Thoughts

The watch can be a powerful tool if you want to practice Digital Intermittent Fasting, but it is not yet ready to fully replace the phone. Like a butterfly emerging from its cocoon, the smartwatch is still in the process of evolving. As technology advances, the user interface will morph and adapt, but some features will always belong to the realm of larger devices.

On my side, the one-week experiment has been a journey that helped me better engage in the present moment and savor the flavors of social interactions. The disconnection from the endless scroll of content was incredibly liberating, and the forced brevity of communication due to the interface constraints was a blessing.

Consequently, I have decided to embrace the Digital Intermittent Fasting lifestyle fully. I am committing to dedicating parts of my day to rely on the watch solely. At the same time, I will gradually work towards the goal of having phone-free weekends. Let's see how long I can sustain it.

Share

A few years back, I wrote a piece called 'When Network Effects Lose Their Armor' exploring the situation when technological shifts could destroy the supposedly unbreakable network effects built by giant companies. The initial post covered the case of ride-sharing networks and how self-driving cars could render them irrelevant. In this article, we will look at the case of social media companies, how their initial defensibility built around Social Graphs is disappearing, how they are creating a new one, and how the newly built one could also vanish.

As Facebook rose to dominance, it seemed that nothing could stop them. Their growth was fueled by the increasing density of connections between people as they 'friended' and 'followed' each other in masses. It was difficult for individuals to just leave and take their network with them.

This was certainly true for many years. For a long time, their position seemed unassailable. While competitors could potentially replicate their products, their true advantage was the vast and intricate Social Graphs they had built using the data of billions of users. These graphs were like a fortress, protecting their business and providing a seemingly unbreakable defense.

When WhatsApp launched, it revolutionized software design and social networking by breaking from two established conventions. First, it eliminated the need for a username and password by using the person's mobile number as their unique identifier and SMS for authentication. Second, it streamlined the process of building social connections by leveraging the user's existing address book rather than requiring them to create connections one by one manually.

As consumption gradually moved from the web to mobile, what seemed like a convenient and benign approach had actually weakened the importance of Social Graphs of companies such as Facebook. Indeed, with this setup, the Social Graph is no longer locked into a central platform but is owned by each user and can be easily transferred to a new service by simply granting new apps access to their address book. This significantly diminished the relevance of Facebook Messenger and ultimately led to the acquisition of WhatsApp.

Still, Facebook's Social Graph kept being central for content distribution as instant messaging was not the only reason people spent time on the platform. In addition, it allowed users to see posts shared by their friends, businesses, and news sources they follow. The feature gave an excellent distribution channel to influencers and companies to build an audience they can reach directly.

From my side, I gradually came to the realization that my Facebook feed had become nothing more than a tedious, uninspiring scroll of noise. The personal updates from friends were fine, but it was full of external content, shared by my connections, that I had no interest in. So, I decided to delete the app five years ago. Moving to private WhatsApp groups with my inner circle proved to be a far superior substitute for maintaining regular contact and a much more nourishing social experience.

Nonetheless, I still needed a place to find other posts about topics I was interested in beyond private messaging groups. Twitter was the best platform for that as it had a follow model not necessarily related to my social circle. The challenge was that I had to manually build lists of people in each domain, such as technology, investment, and comedy, and keep updating them over time. Slowly, the feed became more appealing as it matched the type of articles I wanted to see.

In hindsight, that was the beginning of the rise of the Interest Graph, even though, at the time, it was still thought of as a Social Graph of people we followed. But the real breakthrough app that industrialized the Interest Graph concept was TikTok. It created an algorithmic recommendation engine that dynamically builds a holistic view of users' interests while discovering and consuming the served videos. An interest profile is gradually developed depending on how long it takes to skip and how long each video is watched, among multiple other variables. The more people use it, the better the content matches what users like to see. This process occurs pretty fast in a matter of days without manual curation. In this model, the platform determines which content is recommended, not just the people that are followed or part of the social connections. This not only leads to a superior consumption experience that gets everyone strongly hooked but also renders the Social Graph almost irrelevant.

So, the defensibility of the Social Graph has been squeezed from 2 sides:

1- The shift to mobile address books for social connections and their portability across applications

2- The move to the algorithmic Interest Graph that is built independently from social connections

Consequently, Facebook had to implement a new similar interest-based model. The new algorithmic feed was done on Reels, the TikTok copycat, and the standard Instagram, which now shows many posts not originating from people you follow. On the traditional Facebook app, Meta settled for a hybrid solution. The 'home feed' is still made of content from friends, but they added a separate 'watch feed' that uses a recommendation engine to serve videos based on each user's interests.

Twitter also adopted a combined approach but implemented it within a single feed. Now, the algorithm presents a mix of tweets from both the people you follow and your interests, as determined by your past interactions and engagements. This change has dramatically improved my Twitter experience. Rather than manually searching for interesting people and checking who they follow, I now receive higher-quality posts directly in my feed. It also serves as a starting point for finding potential new accounts to follow.

It seems though that Twitter will soon follow Facebook’s approach and split into two separate feeds.

With companies utilizing recommendation engines to distribute their content, these algorithms became a powerful secret weapon unique to each platform. As the engines scale and their accuracy improves, they provide a new kind of business defensibility, replacing the lost protection of network effects based on Social Graphs.  

That said, one may wonder: have the ancient social media giants really managed to adapt and evade extinction? And is the new defensibility from the proprietary Interest Graphs stronger than the former Social Graphs?

The recommendation engines based on Interest Graphs are definitely superior to the older social-based model. However, the systems used by these platforms are designed to maximize engagement rather than serve the user's best interests. They not only consider the person's interests but also act like a clever magician, using all sorts of tricks and psychological manipulations to keep people hooked, even if they are not genuinely interested in a given topic. So, while the concept is great, the current implementations and business models are problematic.

Trapped by the algorithms, people waste hours daily, consumed by a never-ending feed of videos. As a result, users suffer from various side effects, such as difficulty focusing on other tasks, anxiety, poor mental health, relationship issues, and sleep problems.

Still, even though excessive use can poison users like a toxic substance, it doesn't hurt the platforms' defensibility. In fact, it makes them even more powerful. The more people are hooked, the more ad revenue pours in for the companies, and the more the algorithms learn about each person's preferences, the more addicted they become.

So, for now, older social media services seem to have managed to build a replacement armor made of recommendation engines, to substitute the lost protection from the Social Graph.

Yet, we have to wonder if the new reign of the Interest Graph is eternal? Or will it eventually crumble under the weight of its own power at some point?

In 2020, the US government filed an antitrust suit against Facebook, intending to break up the company due to its monopolistic practices and invasion of people's privacy.

Nevertheless, I do not believe that breaking up Meta into smaller companies would make much sense. Splitting Facebook, WhatsApp, and Instagram into separate legal entities would not address the underlying issues mentioned earlier. The same problems would still exist within each individual service, and the only difference would be smaller revenue for each entity. TikTok is already independent of Facebook, but it still leads to the same adverse effects.

Instead of breaking up Meta or banning TikTok, the government should release a new regulation that requires all social media providers to unbundle the content posted on their platforms from their recommendation engines. In other words, they should offer people the option to use either the built-in discovery algorithm or connect their account to an external independent engine that can prioritize their real needs and well-being.

Such a change would also require new business models to address critical questions. For instance, how the current platforms would be compensated by external engines accessing their content? And how should the independent algorithms price their service to the end user? The ecosystem will quickly fall back to the same old patterns if it's a free ad-based model. Should they be required to offer it as a paid subscription?

Once the business model questions are sorted out, implementing the proposed regulation would bring about two significant benefits. First, it would foster competition and address the government's concerns about monopolies. By requiring platforms to offer the option of using external recommendation engines, new companies could enter the market and provide alternative discovery algorithms. Second, this regulation could address issues related to manipulating user behavior and spreading misinformation. By separating content from built-in recommendation algorithms, individuals would have greater control over what they see. The reason is that independent engines would not be motivated to prioritize engagement and attention over users' well-being. As a result, users would be less vulnerable to manipulation, and the dissemination of false information would be more effectively contained.

If such regulation is adopted, social media platforms will change from puppet masters pulling the strings of user behavior and preferences through their algorithms into neutral entities for content distribution. In this scenario, the protective moat built around the Interest Graphs will disappear, leaving the platforms vulnerable and exposed to competition from newcomers.

I believe this is an inevitable scenario. It is not a matter of whether it will happen but rather when. For now, the question is, what can the current giants presently do to avoid becoming just another commodity content storage company in the future?

Share

The idea of the metaverse has been gaining momentum in recent years, becoming a catch-all buzzword used to describe different things in the digital landscape. What it means depends on who you ask. It could be VR social networks, online games with crypto assets, 3d communication spaces, or any virtual world. But is all the hype justified? In this piece, we’ll explore applications with good potential and also consider situations where the concept may not make much sense.

When I tried the first-generation Oculus Quest, I was immediately impressed by the immersive feeling it created. One of my most remarkable experiences was with a visual meditation App. As I entered the virtual world, I found myself floating in space surrounded by beautiful colors. As I breathed, the air flowed in and out of my mouth and was represented by bright, colorful beams of light. The demo lasted for 7 minutes and, to my surprise, by the end of it, not only I felt more relaxed, but my heart rate had actually dropped by 9 bpm. Visibly, the format could provide a holistic stimulation of the senses in a way that classical settings may not be able to achieve.

The metaverse often refers to creations like Roblox, Sandbox, Facebook Horizon, and Second Life. Each of these platforms, however, has a different interpretation of the concept. For example, Sandbox defines itself as a world where players can build, own, and monetize their gaming experiences using the Ethereum blockchain. On the other hand, Facebook sees it as a future where VR and AR devices will replace smartphones as the primary way people interact with each other and the world, in a 3D virtual environment.

Perhaps, the trickiest part of the term is the word 'the'. It implies that it is the one new singular thing that will supplant the Internet. Nevertheless, in reality, we are witnessing the emergence of multiple 'metaverse(s)' that exist in parallel. Consequently, it would be more accurate to refer to the metaverse as a 'type' or 'format' rather than a distinct realm that will replace the Internet.

Assuming the definition of 'format', metaverse(s) could then be ways for businesses to reimagine part of our lives in 3d virtual spaces. With this framing in mind, would this kind of structure ever take a significant portion of our attention? What type of activities would we want to spend time doing in there? And will the experience always be better than with traditional interfaces?

Virtual environments do offer a deeply immersive experience, providing a chance to explore and engage with places that may be impractical or even impossible to visit in the real world. From far-off planets to ancient ruins, we can virtually visit these locations and interact with each other. The opportunity to attend live events, such as concerts, and experience the feeling of being in a packed stadium without physically being there is another appealing side. The convenience of interactive 3D worlds also has the potential to benefit education and offer a safe space to practice for real-world situations without any risk.

One of the most impactful aspects may be the ability to interact with each other using photorealistic avatars, as recently demonstrated by Facebook. These avatars can enhance online interactions by making them feel more like in-person meetings, fostering a deeper sense of connection than a video call could ever provide. This could improve both business and personal encounters.

Once the development is completed, the virtual meeting experiences will be much more appealing than the current demos featuring cartoonish avatars without legs.

The examples above indicate that 3D worlds can bring an enhanced capability in some instances. When the hardware becomes lightweight enough, people may opt to interact in this manner for certain activities rather than using computer and mobile screens.

But will most of our daily activities work better in 3D immersive worlds?

In other words, are we moving toward the scenario from the sci-fi novel Ready Player One?

I believe, the hype surrounding the metaverse is likely misguided due to a flawed assumption. While the format may work pretty well for certain actions, the idea that every aspect of our physical and digital lives will be transferred to a three-dimensional virtual environment is unrealistic.

Take the example of shopping. It has traditionally involved going to physical stores to buy products. However, the rise of e-commerce has made it possible to purchase almost anything without leaving home. But is it necessary or even desirable to enter a virtual 3D shopping mall to buy groceries or clothes? In the case of groceries, a simple voice command to an assistant would provide a much better experience than walking around a virtual supermarket. And for clothing, the technology would be better used to show how the clothes would look in your photos rather than providing a virtual shopping environment. So overall, not only is immersion not necessary for regular shopping, but it could be detrimental.

Likewise, having to enter a virtual branch of a bank to make a wire transfer would not make sense. The same is true for tasks such as stock trading, booking flights, summoning a ride-sharing service, taking notes, managing one's schedule, creating a to-do list, writing, coding, listening to audio, learning to play an instrument, reading, paying bills, checking the weather, budgeting, ordering food delivery, tracking a package, finding a recipe, and cooking a meal or scheduling a doctor's appointment. And that’s only a sampling of what people do on a regular basis.

Virtual worlds are not suitable for activities that do not benefit from a spatial setting. But even for actions that could fit a spatial representation, there are still parts that cannot be adequately replicated. For example, while generating a convincing 3D version of a guitar or piano is trivial, it is not yet feasible to recreate the tactile effect required for the hands to learn to play the instruments. The lack of a realistic sense of touch makes many other day-to-day activities either unachievable or significantly less enjoyable. The same is true for the senses of taste and smell.

What about Augmented Reality?

Some may state that the metaverse concept would be too narrow if it were only seen as a VR space. The argument is the metaverse is also about a blended version of virtual and physical worlds through augmented reality.

For now, the available AR devices, such as Microsoft HoloLens and Magic Leap, have been repurposed to focus on a small subset of industrial domains like medical, defense, and manufacturing. So far, they have had a little uptake. After 6 years, only ~300,000 units of HoloLens were sold versus almost 15 million Oculus Quest 2. Nonetheless, AR devices will get smaller, lighter, and cheaper over time, which may render them accessible for more mainstream usage.

Pokemon Go was a fascinating experiment that got people hooked on chasing virtual creatures around their neighborhoods and didn't even require dedicated glasses. So maybe there's some appetite for augmented activities. The earlier example about learning an instrument in VR would actually work better in AR. A person practicing the piano could use a real instrument and have visual guidance overlayed on the keyboard, showing which key should be hit next. We can also imagine AR dating scenarios where users can show their profiles to others while navigating the city. This way, they might see potential matches as they go about their day.

Thus, there is room for augmented experiences where not completely disconnecting from reality may be more effective than fully immersing ourselves in a virtual world.

But the burning questions remain: will this be our preferred primary way of interacting with the world for all day-to-day activities? And will our delicate brains be able to handle a permanent visual stimulation directly projected onto our eyes?

Mobile phones give us the ability to decide when to view our screens and when to ignore them. Still, we struggle with an overwhelming avalanche of notifications and an endless stream of information. It reached a point where operating systems now offer a 'focus mode' that helps reduce the amount of data we see as well as a 'screen time' chart to monitor our usage.

This concept video shows an extreme AR scenario.

While anecdotal, it may not be that extreme and does make a good point prompting us to consider whether we would want to live in that kind of world.

As mobile devices replaced desktop computers, we traded larger screens and deeper immersion for greater accessibility. With the current issues of information overload and screen addiction, we may end up choosing to move further in that direction with even smaller screens and less immersion. Smartwatches paired with voice-based user interfaces may be the next typical setup. With artificial intelligence making significant progress in language understanding, it is possible that many user interfaces that previously required displays could be replaced by vocal conversational interactions.

So, what if instead of using augmented reality to see more, we could use technology to see even less by having smarter ways to access and interact with data? Doing so would preserve the health of our brains and our attention spans. We may also find ourselves better able to focus on tasks and engage in creative work without being constantly bombarded by a permanent external stimulus.

The metaverse, as currently imagined, may not become the center of our lives. Yet, it holds the potential to enhance traditional interactions with its immersive and augmented experiences for a subset of activities. So, will it revolutionize the way we live at some point in the future? Or will it be relegated to a mere novelty? We'll have to check again in about a decade. For now, I remain skeptical.

Imagine if you could have something similar to the genie from Aladdin. This powerful apparition could grant you any wish without limit. All you have to do is make a request, and he works his magic to make it come true.

The latest AI generative models do feel like they came out of a magical lamp; you enter a prompt describing what you want, and shortly after, a creative image representing the request appears; you ask it to write an essay about a given topic, and a series of paragraphs show up in seconds. This development is made up of diffusion models and large language models (LLMs), which have led to some impressive advances in AI.

This post is divided into two parts. The first one covers an overview of the recent progressions in AI, the possibilities they open, and their challenges. The second part explores a few thoughts on how to democratize access for all and avoid excess control by a handful of large organizations.

If you are already familiar with the space, you can skip to Part II.

Part I- The Potential of Language Models: Examining the Benefits and Challenges

The evolution of AI has been extremely slow from the field’s inception in the mid-fifties until the late nineties. However, about a decade ago, the implementation of Deep Learning, enabled by cloud computing and the availability of large data sets, led to revolutions in areas such as computer vision, speech recognition, autonomous driving, forecasting, and fraud detection. A tipping point was reached where exponential improvements seemed possible.

Nonetheless, after a few years of excitement, Deep Learning applications appeared to perform well on domains such as the ones mentioned above but couldn’t effectively solve natural language processing (NLP). A short interaction with a customer support virtual assistant should suffice to convince you that the service was close to unusable and was among the most frustrating user experiences. Even personal assistants like Siri and Alexa can only handle a predefined set of questions.

But the arrival of the Transformer Neural Networks Architecture in 2017 marked a new turning point that leaped forward the natural language processing field, in a few years, and gave birth to a new generation of models such as GPT-3, the now famous implementation by OpenAI. Those new models outperformed the previous architectures on a range of NLP tasks and can answer any question, write articles, fiction stories, and software code with mind-boggling realism. For the first time, we witnessed a language fluency closer to what was imagined in sci-fi with HAL 9000 from the movie 2001: A Space Odyssey.

It is not hard to see that the effects on society and the future of work are profound. Still, we may not fully realize all of the ways natural language processing will change our world in the long run. What we can do, at the moment, is to try to anticipate some of the short and mid-term outcomes.

The Good

The immediate benefits will come from leveraging LLMs as sidekicks for humans to expand their creativity, speed up their workflows and grow their earning capacity rather than threatening their jobs.

Several great applications have already emerged. For example, GitHub Copilot helps software developers quickly generate code; LEX assists authors in alleviating writer’s block and coming up with more ideas; resumeworded.com provides personalized feedback on resumes and LinkedIn profiles, helping increase job opportunities; open-source visual models such as Stable Diffusion have been customized and repurposed for specific use cases. For example:

Profilepicture.ai uses a set of your photos to automatically produce creative profile portraits with various styles suitable for different occasions.

InteriorAI.com lets you upload a snap of your interior and provides a remodeled version of the room based on specific styles. Looking at the examples, it seems the system works by doing an image search for similar spaces in the selected style rather than generating new designs from diffusion. Still, the results are pretty cool.

But it doesn’t stop there. LLMs can also:

Create digests of long articles or books

The functionality saves time and is an invaluable tool for people with disabilities who have difficulty reading or processing large amounts of information.

Askmybook.com by Sahil Lavingia is an interesting experimental idea where you can ask questions about the content of his book.

Generate structured legal documents

Write a standard lease agreement for an apartment.

The outcome’s quality is solid, providing accessibility to anyone at a low cost.

Write songs and poems

Write a song about the state of the world today and make it rhyme.

The world is such a mess right now
I can't even pretend to be alright
I watch the news and I just shake my head
How can we ever hope to make it right?
 
We're fighting wars and we're causing pain
The world is shaking and it's not right
We're tearing down what took years to build
How can we ever hope to make it right?
 
I see the hate and I see the pain
I see the hurt and I see the scars
I see the people who are suffering
And I can't just sit here and do nothing
 
So I'll stand up and I'll be vocal
I'll speak out and I'll be heard
I'll do whatever I can to make a change
To make the world a better place

While it’s far from perfect, it’s a pretty stunning base, generated in a few seconds, that an artist could start from to craft the lyrics of a complete song.

Evaluate business ideas

Is it a good idea to create a competitor to twitter?

There is no one-size-fits-all answer to this question, as the best way
to create a successful competitor to Twitter will vary depending on the
specific market and users involved. However, some tips on how to create
a successful competitor to Twitter would include studying Twitter's user
base and identifying areas where it falls short, as well as looking at
other successful social media platforms and learning from their successes.
Additionally, it is important to focus on creating a unique user
experience that offers something that Twitter does not in order to attract
users.

Why did Elon Musk buy twitter?

There is no definitive answer to this question, but it is speculated that
Musk may have bought Twitter in order to gain a larger platform for himself
and his businesses. It is also possible that Musk saw Twitter as a
potential investment opportunity

Those answers are as good as what a generalist person, not deeply involved in the matter, would provide.

Do a few more things

The LLMs’ possibilities go even further. The thread below contains more examples from the latest OpenAI release.

Beyond what’s already possible, what other skills could those models learn in the future?

For example, they could be trained on the entire collection of business plans, financial models, and strategic narratives contained in PowerPoint decks and Excel files from management consulting firms. The models would then allow anyone to generate business presentations and spreadsheets in seconds that could later be manually edited. The gains in productivity would be huge. Such a service may also open up alternative business models for consulting companies with software subscriptions for a new kind of offerings that could be named "Copilot for Slides" and "Copilot for Excel Models."

Another area with good potential is psychotherapy. If the models are trained on transcripts and recorded audio of sessions provided by practitioners, they could enable the creation of an affordable virtual therapist accessible 24/7. The development process may be controversial due to privacy concerns. But a coordinated effort to ensure anonymizing names and personal information before utilizing the data sets could present a viable solution.

So what could go wrong?

The common concerns that come to mind are related to the impact on jobs. In the short term, I believe the technology will bring an augmentation of human creativity and output rather than a replacement. But there’s no doubt that those evolutions will cause some disruption to the world of work at some point in the future. As the solutions keep improving, employers will realize they can achieve equivalent results and serve the same number of customers with a much lower number of employees. For example, customer support departments will rely on a combination of virtual interactive agents and automatic responses to user requests. Humans would merely approve the replies and occasionally edit the content to correct glitches. Consequently, only a fraction of the workforce would be needed.

However, just as in the previous technological revolutions, the reduction of old jobs comes with the creation of new ones that didn’t exist before. Two decades ago, Social Media Marketer, App Developer, Cloud Computing Specialist, YouTuber, Podcaster, Influencer, Video Game Streamer, AirBnB Host, Drone Operator, VR Designer, Smart Contract Developer, Growth Hacker, and Online Educator were professions that were not available. Likewise, we may be astonished by the new opportunities that arise from the present AI wave.

Beyond the impact on employment, the current LLMs do raise a few concerns. While they can be very broad and effective, as shown in the earlier examples, one worry is that they can capture undesirable biases which may be in the training data. Thus, they could amplify existing social prejudices as people use them.

Another cause of concern is that LLMs sometimes “make stuff up” that sounds real but is actually fake. For example, when I requested an essay about myself, I expected the model to return a few stitched paragraphs from my LinkedIn profile, assuming the professional social network public pages have been part of the training data.

Here’s what I received instead:

It is a wild piece of fiction of what my life could have been in a parallel universe.

The challenge with LLMs is they can be linguistically correct but factually incorrect. The reason goes back to how they are designed to work and how they are trained.

Christopher Manning gives a good analogy in his presentation of how LLMs are built. First, they play the Mad Libs Game and try to predict the next word within a particular context. The training occurs by reinforcing the rating of good guesses and downgrading the ones from wrong predictions. Once the model is trained, those final ratings, called weights, are used to make predictions of the most likely next word in the milieu of content the user provides. Subsequently, when LLMs lack input about a specific topic, they will still generate a combination of words that have a high probability of being pooled together from a language perspective, even if they represent fictitious facts.

InstructGPT is already an enhancement over the original GPT-3 at following user instructions and increasing truthfulness. So, I expect the subsequent iterations to keep improving by implementing extra validation layers and, ideally, not shying away from answering “I don’t know” if necessary.

When it comes to the ugly side, “The worst AI ever” by Alberto Romero covers the event of an LLM experiment leading to an obnoxious outcome. It depicts the case of a Machine Learning researcher that fine-tuned the training of an existing model on a toxic discussion forum and released it. Within two days, it was used by bots that generated over 30,000 posts of offensive and trolling content, mimicking the style of the chat board it was trained on.

Whereas the example above was intended for research purposes, a similar approach could occur within an ill-intentioned context.

Part II- Avoiding Concentration of Power: How Open-Source, Crowdfunding, and Distributed Computing Can Help Democratize AI

We have already had a taste of the side effects of the centralized power of social media giants over society, with the potential role of personalized targeting and algorithmic feeds in manipulating political campaigns, referendums, and public discourse. The latest AI developments could become a grander force than social media.

OpenAI’s original ambition was to advance and responsibly democratize the AI field while avoiding that the power stays concentrated in a few giants’ hands. It was initially structured as a non-profit with the promise to share the research and code. However, the evolution path branched in a different direction for reasons well described in the recent New York Times article. The founding team had to create a for-profit arm to finance the enormous resources required to build and run large-scale experiments. Along the way, Microsoft invested $1 Billion. Consequently, today most OpenAI breakthroughs are behind a closed wall, accessible through paid APIs. The exception is Whisper, the multi-language speech recognition and translation software recently open-sourced.

It’s important to acknowledge that OpenAI API access to its models is already a form of democratization, even if it’s based on a commercial service. It’s what has enabled many of the innovative applications we have seen lately and will keep contributing to the field’s evolution. The challenge is that the more people build on top of those proprietary backends, the more dependence and accumulation of power will centralize around a few players. In good times, it may not cause concerns. But in moments of economic misalignments or global conflicts, the risk exposure of dependent operations will be high. Those central entities could potentially pull the plug, instantly leaving businesses and critical processes of countries frozen. Thus, additional levels of democratization should be addressed.

Stability AI has opted for a different approach by open-sourcing Stable Diffusion. From a tweet by Emad Mostaque, it appears the intention is to keep opening up new developments as they are completed.

Other initiatives followed course like LAION, EleutherAI and Harmonai, open-sourcing research, data sets, and models for photos, language, and audio. Those are released under a Creative ML OpenRAIL-M license which allows commercial and non-commercial applications while focusing on ethical, legal, and responsible usage.

So there are encouraging enterprises currently in motion to ensure a more even and independent access to AI. What’s missing is additional funding to cover the super computing resources required to allow the ongoing initiatives to build open source solutions at a scale on par with what the giant corporations can attain.

One way to address the capital challenge is through crowdfunding and distributed computing.

Crowdfunding

So far, Kickstarter has financed over 230,000 projects with more than $6 Billion. While Kickstarter focuses primarily on creative projects, other sites, such as experiment.com are designed for scientific research. Those services are great alternatives to fund exciting projects by allowing people to finance ideas they believe in. The rewards the backers receive can be anything from a personalized “thank you” note to early access to a finished product.

It is worth exploring funding experiments on the existing platforms to evaluate if they are suitable. Alternatively, the optimal way forward might be building a new crowdfunding platform specifically catering to foundation AI projects.

The Search for Extraterrestrial Intelligence (SETI)

SETI is a scientific quest to find evidence of intelligent life in the universe by analyzing the signals from radio telescopes.

You might wonder what aliens have to do with building large AI models.

In the late nineties, I downloaded a popular program called SETI@Home. It allowed your computer to participate in the search for extraterrestrial intelligence by donating idle CPU time to analyze pieces of the signals captured by the radio telescopes. It got activated instead of the screen saver and displayed fantastic visuals.

I ran the software for about two years, and even if my computer, or any other one, didn’t discover green creatures from outer space, SETI@Home was the greatest distributed computing project in the world. The Berkeley SETI Research Center operated the program, and the initial ambition of the team was to get 50,000 to 100,000 contributors. Instead, since its launch in 1999, it has had over 5 million participants.

The training of large AI models could adopt a similar approach by crowdsourcing the processing capacity from distributed machines. For example, today’s gaming rigs have highly capable GPUs that can be aggregated into a virtual supercomputer. Instead of directly providing cash via a crowdfunding campaign, people could contribute with their existing hardware and the required energy.

The engineering side is not as simple, though. Bandwidth for enormous data transfers is usually a constraint as most participants won’t have home gigabit Internet speed. Furthermore, contributors may not keep their machines connected permanently. Thus, the approach requires an adaptation of the current methods.

“Deep Learning over the Internet: Training Language Models Collaboratively” by Max Ryabinin and Lucile Saulnier proposes a modified algorithm to address the limits of distributed training. It is vital to encourage similar developments until obtaining frameworks allowing to crowdsource any AI model training.

Crowdfunding and distributed computing can truly democratize access by providing the necessary resources to build large AI models outside the realm of heavily funded corporations. Those developments can then be open-sourced and freely available to everyone.

With more democratization also comes the dilemma of how to contain the potential adverse side effects.

At the basic level, courses about ethics and responsibility when building and releasing AI applications should be prioritized and strongly promoted. Raising people’s awareness about the potential misuse consequences is a good place to start.

Another way is to develop guidelines, policies, and a certification system that the models must comply with to receive a safety label. This will prevent well-intended users from unknowingly downloading softwares that have built-in harmful behaviors.

Nevertheless, with such powerful systems, expecting a total containment of the potential harms would be naive. In the same way, the democratization of software compilers led to remarkable innovations but also allowed people to create computer viruses, we will have to deal with some problematic creations.

The computer virus case was initially solved by continuous improvements in anti-virus software. Later, the problem was eliminated by an evolution in operating systems architecture. Nowadays, no one has to worry about viruses in their iPhones or iPads.

So the solutions to the downsides of LLMs specifically, and AI generally, will most likely stay in a “work-in-progress mode” and evolve along the technological advances. It will be hard to fully “crack it” at the moment, but as long as the appropriate focus and funding are allocated to the challenge, I’m hopeful that the right balance could be reached at some point.

LLMs’ possibilities are endless; the ability to automate and scale human communication and empathy with words can completely revolutionize fields like mental health and education via affordable, personalized 1–1 therapy or tutoring.

In the early Internet days, e-commerce and online newspapers were obvious, but Airbnb and Kickstarter were less predictable. With the arrival of mobiles and broadband, video calls were thought to be the killer app, but Uber and TikTok were difficult to foresee. Comparably, the most world-changing LLMs’ use cases could be the ones we haven’t thought about yet.

Democratizing access to AI is an important goal. It is already in motion with the emerging user friendly front end applications. But it is vital to push the democratization further through open-source projects, crowdfunding, distributed computing, and awareness programs. Companies, individuals, and even governments must work together to ensure that everybody benefits from the rewards without an overdependence on a few giant corporations.

Originally published at https://www.ehicham.com on November 24, 2022.

If you've ever used a screen and felt strange symptoms with your vision getting tired, blurry, dry, or even painful, you're not alone, and the reason why it is happening is still a mystery.

I have been closely staring at screens for as long as I can remember. Aside from the occasional slight eye fatigue after spending a dozen hours fixated on a monitor, I never had any concern with displays. My first bizarre experience arose with the iPhone 5, which triggered eye pain and dryness after less than thirty minutes of usage. The condition worsened to the point where I could no longer use the phone without feeling immense discomfort. It was frustrating, and there was no epiphany after a visit to the ophthalmologist. The outcome was, "It happens with phone usage; just reduce your screen time and use these drops daily to improve the dryness." None of those recommendations helped. But when I moved back to the older model, the discomfort disappeared. The two following editions, the iPhone 6 and iPhone X, didn't cause any trouble. Thus, when the iPhone 12 was released, I blindly upgraded, mostly to get a better camera. But, to my astonishment, the new screen had a terrible effect. After a couple of hours, my vision became blurry with a burning sensation. It was even worse than what I had experienced years back. Subsequently, I returned it and came back to the previous model.

Following that misadventure and a recent comparable effect with the iPhone 14, I decided it was time to go down the rabbit hole on Reddit to try to make sense of the situation. The great news is there were display settings changes that could solve it, and that I was not the only one with such an experience. The not-so-great news is none of the suggested settings brought any improvement for me, and it seemed that the group of people with screen sensitivities was relatively small.

Historically, blue light got the most warnings about possible adverse effects on eyesight. But, in this case, the prime suspects were OLED displays. Contrary to LCD screens, OLED panels use PWM (Pulse-Width Modulation) to create lower brightness, which makes a fast flickering by turning the LEDs on and off. It is like a strobe effect that triggers symptoms in some more sensitive people. The lower the flickering frequency, the more likely the eyes will be impacted. For a PWM deep dive, you can read this comprehensive explanation.

Nonetheless, the narrative that LCDs are good, and OLEDs are evil didn't match my experience. For example, the iPhone 5 had an LCD screen and did impact my eyes; the iPhone X had an OLED display but did not cause any adverse effects.

While digging further, there seemed to be more possible variables.

The peak intensity of the brightness

In addition to the speed at which the LEDs are switched on and off, their peak brightness level could also make a difference. In other words, both the amplitude and the frequency may matter.

Temporal dithering

It is the technique used by graphics cards when they can create more colors than the screen can handle. The cards simulate the missing colors by rapidly switching between two colors to blend them. Unfortunately, that color flickering may cause complications for some people.

From the discussions in the forums, it is clear that there is variability around the models that cause harm to specific individuals. Not everyone is sensitive to the same kind of issues.

What’s next?

Given the short-term impact on usability and, most notably, the possible long-term eye health hazards, it is critical to solve this decade-old mystery. We have multiple alternative explanations, but there's not enough evidence to support a definitive answer that would elucidate all cases. More data is needed about the experiences of users from all over the world.

As we keep staring at screens day in and day out, the problem is bound to intensify unless we find a way to fix it. Hence, I decided to launch a small survey to gather an inventory of devices that cause eye strain and related symptoms. A pattern from correlated specifications may be revealed with a large enough number of responses.

If you ever had eye troubles from using a phone, tablet, or computer, share your experience in this short survey.

You may be wondering how that data would be helpful. The information alone will not bring the solution directly, but it is a necessary step towards a potential resolution path. The story of Martine Rothblatt, who founded what is known today as Sirius XM, is an inspiring example. Her daughter was diagnosed with Pulmonary Arterial Hypertension in the nineties. At that time, there was no cure, and it was a lethal disease. Consequently, Martine embarked on her own research journey and found a molecule, abandoned by GlaxoSmithKline, that could possibly help. The molecule ultimately led to the development of an effective drug to prevent the condition from becoming fatal. As a result, her daughter is now in her thirties. She lives a happy and healthy life working for United Therapeutics, the company founded by her mother that developed the treatment helping tens of thousands of people.

Similarly, we must take the first step and do our own research, which may lead to unexpected insights. Those findings could eventually lay the foundation for a permanent solution.

Vision is one of life's most valuable blessings, and it is our responsibility to protect our eyes' health. So, while navigating the journey to discover the answer to this mystery, it may be wise to reduce our overall screen time and avoid the devices causing severe symptoms.

Originally published at https://ehicham.com on November 2, 2022.

It would take 468 years to listen to the entire music catalog on Spotify, 163,116 years to read every book on earth, and 95 years to watch all feature films ever made. On YouTube, every minute, 500 hours of video are uploaded. It would require 82 years to watch just the added material from the last 24 hours. Without considering other sources such as podcasts, news, blogs, magazines, and social media, the immense volume of content available would necessitate multiple lifetimes to ingest.

During my childhood, the landscape was entirely different. There was only one national TV channel that started at 6 pm and ended at 11 pm. The typical programming began with kids friendly content, followed by the daily news, a TV show, and finished with a movie or documentary. Books were limited to the inventory carried by the local libraries and music came from mix tapes and a few radio channels; It was tranquil, predictable, and didn’t require much mental energy to decide what to consume.

What may sound like a blessing in today’s overwhelming world with abundance felt very different back then. It was a feeling of content starvation. You were out of luck if you were curious about particular topics and had distinctive musical preferences.

Over the years, the situation improved. Initially with VHS players and later with Digital Satellite TV that blasted the offering by floating hundreds of channels accessible 24/7 by pushing a remote-control button. Today, we have clicking and scrolling; back then, it was called zapping.

I thought we had reached a peak level of media lavishness, but I had no clue about the massive permanent storm that would hit us in the coming decade.

Fast forward, the Internet came along and enriched our lives in extraordinary ways. The web browser unlocked access to knowledge and information about any topic we desired, a massive library of books, and a huge catalog of media, all accessible from a pocket device.

It felt magical and wonderful. But over the years, the sensation of satisfaction from an outstanding level of instant optionality slowly ceded to a sense of indigestion. Throughout my life, I gradually moved from a state of content starvation to a feeling of content satiety to a condition of content gluttony.

Today, the usual suspects that get pointed at are social media apps. Those are the ones that mastered the game of attention and digital overfeeding, which makes them the most problematic. But the issue is much broader than that. Beyond social apps, we still have to deal with news, blogs, newsletters, personal messaging groups, work messaging channels, discussion forums, emails, digital magazines, podcasts, audiobooks, eBooks, online classes, video streaming, and music services. The pull is triggered by the various topics we are curious about and by the trending “current thing”. The broader our interests are, the more severe the phone addiction problem gets.

One recent example is the new episode of the Lex Fridman Podcast with Balaji Srinivasan. Two brilliant minds I enjoy listening to. There’s only one minor glitch; it’s almost eight hours long!

Our brains are not meant to process that much data throughput daily, and the effort spent on deciding what to consume adds decision fatigue on top of the already challenging matter.

So how can we escape this golden trap?

There’s no permanent solution yet. The only fix we have, for now, is to manage the situation manually using band-aids such as periods of digital detox, time-boxing Internet usage and enabling tools that limit phone habits.

The ultimate answer has to come from a technology that would allow us to engineer a healthy kind of artificial scarcity. The kind of scarcity that is created from abundance rather than shortage. Two fundamental shifts need to occur to lay the foundation for such a solution.

1.    A shift in the current business model

We do have to rely on recommendation engines as it's impossible to keep up with the immense media catalogs and the phenomenal amount of user-generated content. The existing algorithmic engines theoretically surface the best content for our interests but in reality, serve the interests of the platforms by keeping us hooked for as long as possible.

Both the ad-sponsored and subscription-based models lead to non-aligned incentives between the platforms and the end users. Ad-sponsored platforms need to keep us around to serve ads that represent their source of revenue. So, their prime customers are the advertisers. Subscription-based platforms get their income from consumers, which may seem like a more aligned model. The reality is that they also need to keep us addicted for as long as possible. If the monthly consumption drops drastically, customers will cancel their subscriptions.

For an algorithmic recommendation engine to truly serve the interests of consumers, it has to be:

  · Unbundled from the platforms

  · Paid for directly by the end users

2.    A transition from centralized recommendation algorithms to local ones

Unbundling the recommendation algorithms from the platforms is necessary but not sufficient. An independent central business running our unified interest graph across all the media would, in principle, better serve the end-user's interests that directly pay for it. But that would only be true if we assume that the central service provider would always act with integrity, would never be influenced by more powerful entities and, in times of global conflicts, would not side with its origin country and target other nations with “info wars.” In other words, that entity would effectively control the tap of what people see and what they are blind to. It would expose us to mind manipulation risks, an order of magnitude more significant than what's occurring today. Even if there is more than a single provider, consumers will be stuck with the engine they have been using as it’s the one that learned their interests. Switching would be impractical.

What also needs to occur is an evolution from centrally run engines to locally run ones.

For example, Stability AI chose that approach when they released Stable Diffusion as a pre-trained open-source image generator anyone could run on their computer. In contrast, Open AI decided to keep central control over their similar system.

"I saw a vision of an intelligent Internet where everyone has their own AI as opposed to centralizing AI…. I thought it needs to be expanded not only through the open source, but across the world"  - Emad Mostaque, founder of Stability AI.

Similarly, the interest graph models could be initially trained based on massive global data sets on a powerful central computer and then made available for each individual to download and run locally. The models would then expand as they learn about specific people's habits and interests while using them.

With such a setup, the attention maximization dilemma that leads to social media addiction could be solved as each person would get their own Digital Personal Curator (DPC). A curator that would have their best interests in "mind." and not be controlled by a major corporation.

The DPC would tone down the overflow of information to the optimal level by considering our objectives, available time, and, most importantly, our well-being.

What would such a system look like?

The Digital Personal Curator could be an AI web browser that acts as a filtering layer between users and the current platforms. It would connect to a locally stored interest graph owned by each individual and then show an automatically generated daily page with only the right dose of content from the various sources that should get our attention. It would also consider our goals, priorities and what’s already on our daily calendars.

Furthermore, the DPC could solve, for example, the challenge of the eight hours podcast mentioned earlier by automatically summarizing it into multiple short versions. Consumers could choose between a fifteen-minute summary and a one hour condensed version depending on the desired level of depth and available time. We have already started seeing language summarization AI models emerge with promising early results, and the rate of improvement will likely be exponential.

Even with all the challenges of the current world of abundance, it is still a much better world than a world of scarcity. We have to navigate the present phase the best way we can. The short-term answer could be to engage in a regular digital detox by narrowing our choices and adopting a minimalist approach. It would require self-discipline, but possible through repetition and consistency.

Once the locally run AI curation technology matures to the optimal level, we can enjoy the benefits of abundance while being shielded from its side effects. Those benefits include better quality of life, more time for meaningful activities, less phone addiction and less fear of missing out.

Originally published at https:/ehicham.com on October 26, 2022.

In recent years, AI breakthroughs led to increasingly impressive outcomes such as partially self-driving cars, AlphaGo winning against Lee Sedol, the Go world champion, deep fakes, and generative systems, like DALL-E and GPT-3, that create astonishingly good visual and written content from a text prompt. Those developments urged part of the global community, including celebrities in the tech sector, to relay warnings about the risks and threats that could emerge from an uncontrolled advanced AI.

Earlier this summer, a Google engineer working on a chatbot framework called LaMDA provoked a storm in the media when he claimed that the chatbot was sentient. Shortly after, he was put on administrative leave. That claim was evidently unfounded.

I am part of the crowd that believes that the dangers are pretty overblown in terms of humans losing total control to AI software. However, it is essential to acknowledge that some flip sides and risks must be closely supervised and addressed as in any other technology. For example, hiring systems trained on a historical data set biased against some minorities or gender, deep fake videos framing a person for a crime they didn’t commit, or social media bots manipulating public discourse, to cite only a few.

Still, I believe that the event of an AI by itself taking total control over humanity is unlikely. One of the most famous apocalyptic scenarios is from the movie Terminator; Skynet, an AI defense system, becomes self-aware after only a few weeks of being launched. Then it immediately figures that it has to eliminate the entire population as humans are its most significant threat. So, shortly after, it initiates a planetary nuclear attack.

In this post, we will explore an alternative dystopian scenario where the threat is not AI becoming conscious and turning against humans but perhaps humans that are already conscious, using AI to upgrade their genomes and interfaces to become superhuman. The potential danger is that Sapiens 2.0, with superior physical and intellectual abilities, may decide that Sapiens 1.0 are a burden to society that need to be eliminated.

Philosophers and scientists have debated for countless centuries about consciousness and how it is created. Yet, until today, it remains one of the most enigmatic aspects of our lives. Perhaps the only agreed-upon part of the concept is that we still don’t have a definitive answer. All we have is a collection of theories and personal beliefs.

Debating those theories is beyond the scope of this article. But, Life 3.0 by Max Tegmark and The Future of The Mind by Michio Kaku are books that cover the topic in interesting ways. Nevertheless, regardless of what consciousness is and our lack of understanding of it, the one sure thing is that it ships bundled with all Sapiens’ biological packaging whenever a human is born, a hundred percent of the time. In other words, we don’t have to figure out how to engineer consciousness or even comprehend it. It is a feature that comes by default if Sapiens’ biology is used as a starting “development platform” to build an upgraded version. It’s like creating a display sign from a collection of LEDs. One doesn’t need to understand how the LED generates light. All the designer needs to know is when a LED is put through a current, it lights up.

About ten years ago, the CRISPR gene editing technology was revealed for the first time. Its vast positive potential was immediately acknowledged. It could not only cure some genetic diseases but also prevent them. The diseases researchers are tackling include Cancer, Blood disorders, blindness, Cystic fibrosis, and Huntington's disease.

It only took a few years for a scientist (He Jiankui) to use the technique to create the first genetically edited babies in an attempt to give them HIV immunity. However, the incident led to worldwide condemnation of the ethics of the procedure and concerns about the health of the babies, as the consequences of human genome alterations are still not fully understood. The criticism eventually led to the suspension of the scientist and his collaborators with a three years  prison sentence.

Several countries have created laws and regulatory agencies to oversee genome editing practices. But, regulation and laws never prevented determined people, whether well-intentioned or ill-intentioned, from persevering towards their goals.

Since the reveal of CRISPR, “Do It Yourself” (DIY) home kits started to appear online, allowing amateurs to “play” with the potent editing tool at starting price points as low as $170. Those kits are not suitable for editing the human genome (yet), but they allow users to edit bacteria and fungi. Still, the dangers are not hard to foresee. It could range from the creation of drug-resistant bacteria to weaponized pathogens.

On the anecdotal side, there are “benign” experiments such as “the glowing mice.” Those are mice injected with a gene from a glowing jellyfish that makes them glow when hit by blue or UV light.

Now, imagine if one day you encounter a green-glowing human in a dark street. Would you find that amusing or terrifying? That day may not be too far ahead.

The gene editing techniques will keep improving, whether it’s CRISPR or more advanced technologies yet to be discovered. At the same time, AI will assist in deepening our understanding of the current genome until fully cracking how it maps to all physical and mental features of humans so that we learn what to change for each outcome.

The question is, how long will it take either independent amateur biologists or a heavily funded organization (private or governmental) to produce significant enough genome alterations leading to the creation of Sapiens 2.0?

Sapiens 2.0 could have far superior intellectual and physical abilities than the current humans and may reproduce much faster. What if they become a neo-fascist community and turn against the original humans? Humans have proved to be capable of the worst evils throughout history. Those traits will most likely be inherited. Is that a more plausible dystopian scenario than an advanced AI becoming self-aware and destroying the world? Should we be more concerned about Sapiens 2.0 or AI?

I believe, most likely, none of the two scenarios on their own.

A third possibility could end up being the most hazardous of all. It is the intersection of Sapiens 2.0, AI, and a brain-computer interface (BCI).

The human brain to computer interfacing is an emerging research field aiming to create a direct communication pathway between the human brain and machines. The early research focuses on improving the lives of people with disabilities. The domain led to some promising outcomes, even from a decade ago, such as the case of this patient moving a robotic arm with her mind.

The research has since progressed into more advanced experiments aiming at turning thoughts into speech and creating wireless brain implants (e.g., Neuralink) that can read the brain's electrical signals and wirelessly transmit them to a computer.

Perhaps, that is the missing link to harnessing the strengths of conscious biological beings and the powerful silicon AI world.

On the one hand, Sapiens 2.0, even with enhanced genetic abilities, may never get anywhere close to the processing power and data storage of networked computers packed with advanced software. On the other hand, the most sophisticated AI supercomputer may never become any more conscious than a washing machine.

The birth of a pseudo-conscious AI could more plausibly originate from a dual system combining human biology with its built-in consciousness with AI software code and its silicon processing power using a BCI.

That breakthrough could lead to a giant leap and propel humanity to the next phase of civilization. It would carry the best of both worlds. But, sadly, it could also bring the worst of both worlds.

If a Skynet emerges one day, it is more likely to be a human with a direct brain interface to an AI rather than a machine by itself going rogue.

Originally published at https://ehicham.com on October 19, 2022.

I strongly believe in the vision of transitioning the world to more sustainable transportation to help reduce overall emissions.

Electric Vehicles (EVs) are one of the critical accelerators of that shift. So, I knew I would make the switch at some point when more options at better price levels became available.

My job had been in Abu Dhabi for the past 3 years, but I had decided to stay in Dubai, a city I lived in since 2009, mainly because all my friends and social circle were there.

The commute is about 130Km (~80 miles) each way and takes around 1h:15 from parking to parking.

Given my high daily mileage and how much it was taxing both the environment and my bank account, I wondered if it wouldn’t be wiser to make the leap to an EV sooner rather than later.

At that time, charging and municipality parking spaces were provided for free as an incentive to encourage early adoption. Those perks, combined with other factors such as the relatively low maintenance requirements, the fun driving experience, and the advanced software, led me to make the jump about two and half years ago.

As EVs were relatively uncommon in the region, many friends and people I know were intrigued. It seemed to them that I had just made my life more complicated. Their questions were mostly centered around charging and range.

Do you ever run out of battery in-between cities? Do you have to stop to charge on your way back? Don’t you already have enough devices to charge every day? Phone, watch, headset, tablet, sleep tracker, laptop?

Friends living outside the country had an even more amusing one:

Isn’t gas almost free where you live?

The short answer is that gas prices are not that much lower than in other parts of the world (e.g., it’s only about $0.9 less per gallon than you would pay in the US at the moment). The reason why that is the case could be the topic of another post.

But the most common question was a variation of:

How much of your life will be wasted waiting for your car to charge?

Surprisingly, people worried more about it than my potentially wasted life from the 2h30 daily commute.

After using the car for a couple of months, the unexpected outcome was that you got back quite a bit of your time.

Filling up the gas tank in a station would take about 10 minutes. Assuming 5 minutes refill time and another 5 to get in and out of the station. Given my daily commute, 2 refills per week would be required, which would add up to about 16 hours per year, excluding a 4-weeks annual leave. Instead, I was charging overnight at my apartment building while asleep and never had to wait. To be slightly more accurate, we can count the 30 seconds it takes to plug and unplug the cable 5 days a week, adding up to about 2 hours per year. On occasions when on a trip during a long weekend break, the car would require 30 minutes at one of the fast-charging points that are combined with commercial retail locations. That time would be needed anyhow for snacking and coffee breaks; usually, the choice was to rest a bit longer than what was required to charge.

The numbers could differ depending on whether home charging is available, the density and speed of the fast charging network, and the geography of the countries where people live. In some cases, the balance could flip in favor of gas cars regarding the refill time economy. Nevertheless, at least when home charging is accessible, lost time waiting for your vehicle to charge is rather a misapprehension.

When home charging is not an option, EV owners can still benefit from other savings not related to charging, such as the time saved on maintenance. EVs don’t require oil and filter changes, and the brake pads last for at least 240,000 Km (~150,000 miles). Considering my yearly mileage, a typical gas car would require about 4 oil changes and 1 brake pad replacement. Assuming an average of 1 hour per oil change and another 2 hours to get the brakes replaced, the yearly savings would add up to 6 hours.

The question is, what can you do with up to 20 hours saved each year?

For example, you could watch 20 to 40 episodes of your favorite TV Show, watch 10 typical-length movies, walk 4 marathons at an average speed of 8.4 km/h (5.2 mph), exercise for 50 minutes for 24 days, meditate for 20 minutes for 2 months, walk your dog for 30min for 40 days, take a long haul round trip flight to your favorite remote destination, read 3 books of 200 - 250 pages, complete 3 short video games, take 1 to 2 online classes, kickstart a new hobby or cook 400 eggs for 3 minutes, one at a time.

Originally published at https://ehicham.com on October 11, 2022.

Audiobooks have been around since the tapes and CDs era. Their spread took place in multiple stages, leveraging every new technological wave.
Three key turning points have significantly boosted their relevance:

1. Internet growth in the mid-nineties, when Audible launched. Till then, audiobooks were only available in physical retail locations.

2. iPod and iTunes Store expansion post-2003 where multiple books could easily be kept in the pocket and consumed on the go.

3. Smartphones’ ecosystem evolution post-2008 when iPhone and Android devices offered mobile stores with the possibility to purchase and download content without requiring a separate desktop action.

My first introduction to the format was in the late nineties, but I never got hooked as it felt that sitting around a computer to listen to books provided a limited advantage over a print version.

It was only in 2016 that I decided to give it another try and since then, here’s what happened…

The ability to listen while driving and walking unlocked a massive daily “reading” window that was never available before.

I couldn’t have read the same number of print books during that period. Aside from the practical side, the voice and style of some narrators made the books surprisingly come to life, and delivered a more enjoyable experience.

However, audiobooks do have drawbacks, and the central issues are most apparent in non-fiction titles.

In this post, I will share a few thoughts about how to solve the most important shortcomings and upgrade the audiobook experience.

Smartphones’ operating systems evolved drastically since the first iPhone launch. Now we have advanced frameworks to build media-rich Apps with features, such as Augmented Reality, that were hard to imagine just a few years back.

Yet, the audiobook format is exactly the same as it was in the nineties. A plain audio recording split into chapters labeled by numbers.

Recently, Audible started adding the actual chapter names which resolved one of my major complaints. Now, at least a high-level navigation within the book is possible.

The inadequacies of the audio format become apparent with content rich in numerical data, graphs, sketches or any visual representation. Also, questionnaires and whatever requires input is unsuitable.

Presently, that problem is solved by offering a companion PDF file including the extra material. The downside of that method is that:

1. It requires a different device to download the file.

2. The context around the content is lost. You can’t view the graphs while you listen and once you have access to the visual file, there’s no simple way to navigate back to the related audio section.

There were attempts in the past to build eBooks 2.0 by enriching the mostly text content with audio, video and interactive 3D animations. Companies such as Inkling started with academic textbooks and designed their solution around the iPad. The format never took off though. They had to change their model to a B2B software platform for corporates’ internal training manuals.

“Audiobooks Are the New Ebooks, Except They Might Keep Growing”

This piece by Boris Kachka summarizes the history and current trends of the digital publishing sector where audiobooks are the fastest-growing segment.

What if a more attractive format was to go the opposite direction?

Start from a mostly audio content and enrich it with visuals and interactivity in the relevant parts.

Let’s look at the example of the Audible App.

Most of the display real estate is untapped and only shows the static cover of the book.

That area would be much more useful by becoming active during the parts that relate to numerical tables, graphs or other visual representations.

Furthermore, non-fiction titles often include web links. It is a hassle to manually bookmark the audio segment, go back, listen again, and take note of addresses. Besides, long URL strings are not practical to write down.

Here are the 3 enhancements that could considerably augment the current audiobook experience.

1. Active Links

When a link is mentioned, it would be worthy to present a synchronized notification in the App central space with various actions to choose from:

• Open link in the browser.

• Add to bookmark or favorite reader.

• For media URLs, play directly or open separately.

• If it’s a reference to another book, show the cover with an option to add to the Audible/Amazon wish list.

• For Podcasts, offer to add the episode/show to default/favorite client.

• Even when people are mentioned, showing the written names and photos with a link to their Wikipedia or personal pages would be great.

• Various books have a summary section at the end of each chapter. Why not make the actual text of those sections available in the app? Recap bullet points are sometimes easier to remember in text form.

2. Alternative Navigations

Currently, only chapter level navigation is available. Another useful way to browse could be from visual assets.

The option to start from a list of graphs, tables, photos or links then be able to listen to the related audio segments would be handy especially after finishing the book. It’s a practical way to review sections of interest or the audio portions that were played when one could not look at the screen.

An additional convenient feature is The Lost Art of the Index. As mentioned in the article, even eBooks rarely have an index. The argument is that with the digital format, there’s search.

Search is great when one knows precisely what to look for.

…The other problem with relying on search instead of an index is that you lose the benefit of synonyms and related terms. An indexer takes all that into consideration so you’re much more likely to find everything you’re looking for with a good index than a simple text search…. — Joe Wikert

Audiobooks don’t have an index nor a search function.

The text from which the audio was recorded is already available. There would be no need to use speech recognition to make the content searchable. It’s a matter of syncing/mapping the two sources.

Adding advanced search would be huge if done right:

…When you search for a phrase in an ebook the results are shown in chronological order. You see all the occurrences from the beginning of the book to the end. Imagine if Google worked that way. So when you type in a phrase Google tells you the first (oldest) site to use that phrase, then the next oldest site that used it, etc. Users would laugh and reject it, yet that’s exactly what we’re forced to accept in ebook search.

What I really want is relevance-based results. Show me the location in the book with the highest density of that phrase and prioritize occurrences of it in a heading over occurrences in body text. I’m sure there are other attributes that could be rolled into an effective ebook search algorithm but I’ll take just those two features for starters... — Joe Wikert

Lastly, why limit search only to the current book?

In the digital format, books don’t have to be kept siloed.

For fiction titles, it wouldn’t make sense to run a query across multiple books. However, when it comes to business, reference guides, and how-to content, a cross-book search within the personal library would be extremely valuable.

Multiple books cover similar topics, and the knowledge is now isolated within the individual titles. An advanced search ranking similar to what Joe Wikert suggests above would be even more potent by surfacing insights from multiple sources.

Finally, I would love to see a topic based playlist dynamic creator. That feature could leverage the cross-book search function. It is for sure more challenging to build with the currently available NLP technology. It would need to figure out the length of the sections that cover a given topic from each book then work-out the most relevant ones to combine. The other difficulty is that in some cases, content may lose its meaning when taken out of its original book context. Still, it’s worth building a basic version that could progressively evolve.

3. Voice Notes

eBooks already support notes’ taking and text highlighting. Those features are still lacking in the audio format.

The notes should ideally be stored as voice memos and then post-processed into text format to be searchable. Real-time speech recognition has significantly improved and works well in many cases. Nevertheless, the quality degrades rapidly in the presence of background noise. In the car, for example, short commands do work, but longer form dictation is almost unusable.

The transcribed voice could later be used as alternative navigation to dive back into the related sections of the book.

Furthermore, when listening via home speakers such as the Amazon Echo, voice notes would be the most natural method to record personal insights.

The case of fiction books.

So far, most of the suggested enhancements are not relevant to the fiction genre.

Indeed, adding visuals to fiction books would start to get into movie production territory.

There is, however, one enrichment that could make a fiction title more engaging and entertaining. It is about adding a soundtrack with music and sound effects.

The example below from GraphicAudio gives an idea of the result.

www.graphicaudiointernational.net

The sound effects presence and volume could be tuned in the settings to suit individual preferences.

Audiobooks are becoming more popular than ever. Nonetheless, both the current format and available software players are rather primitive. Augmenting the technology would have clear benefits from a user’s perspective, but the effort would require investments from both the content distribution platforms and the authors.

Audible and iTunes are best positioned to lead such a development but do they have enough incentives to do so?

The total crypto market cap hit its all-time high on January 7th, 2018 reaching $830 Billion for a few hours before dropping to $430 Billion 10 days later. By the end of June, it collapsed to $235 Billion.

There are many pieces written about the cryptocurrency bubble and its similarity to the .com era.

Looking at the speed at which capital was flowing into this market, the number of undifferentiated coins and ICO scams, one could only wonder how could this last.

But is there more to the current crypto craze than the ”yet another bubble” narrative?

One of the thoughtful and balanced articles about this space is by Mark Suster: Want to Really Understand What all the Hype of Cryptocurrency is About?

1. Allowing for a decentralized Internet in which value is accrued to infrastructure, protocols and applications that serve market need

2. Allowing electronic trade across actors who may not know or trust each other without middlemen who take a heavy toll / tax on the transaction

3. Allowing for (the potential of) a more stable currency than one’s own government for citizens who may live under despotic or irresponsible regimes

Those are the most impactful initial use cases and will be achieved over time by the platforms, protocols, currency-like coins, and distributed apps that already started to emerge.

It’s worth reading the entire blog as it effectively articulates those three areas and the ways blockchain and decentralized systems could offer potential solutions.

I also previously wrote about how blockchain and smart contracts could improve online marketplaces by cutting “middleman entities” and bringing the value back to the trading parties.

There is, however, another use case that is much larger, in terms of allocated capital, than anything else.

In what follows, we will use the more generic terms “crypto assets” or “crypto” instead of cryptocurrencies as the current circulation is a mix of currencies, utility tokens, protocols, securities and, of course, kitties…

One of the common critics of crypto assets is that it’s an instrument for speculative trading.

Looking at why most people invest in crypto at the moment, it is more often about the potential gains from price increase than for the underlying utility. This is true even for protocols and apps built with a genuine need for a token to operate the system.

The money injected into blockchains to cover the cost for running smart contracts, decentralized apps and transactions is only a small fraction of the total crypto market cap. The most significant portion is a store of capital with the objective of selling at a higher price.

Crypto assets are indeed used massively as a speculative investment vehicle.

The puzzling part is why so many consider this a problem?

Let’s step away from the crypto world for a moment and take a look at the traditional stock exchanges.

The top 20 stock markets’ capitalization constitute over 90% of the global exchanges.

Their combined market cap is $77.7 Trillion with a monthly trading volume of $5.9 Trillion.

There are distinctions between stocks and cryptos and what market cap mean for each. “Cryptoeconomics is hard: Market Cap” is a good read addressing those differences.

However, what is similar is the motive behind why people invest in both the public stock market and the crypto assets. That motive is speculation. Whether it’s day trading or long-term investment, the objective is to make potential profits from the appreciation of the asset.

IPOs do provide initial utility. They offer a fundraising mechanism for companies, some liquidity for founders and employees as well as an exit for early investors. But beyond the IPO milestone, public stock ownership doesn’t give access to any utility offered by the underlying company. It is only held or sold for potential financial rewards.

As an example, let’s consider Tesla, a company I highly admire.
It had revenue of $7B in 2016 and a market cap of $35B.

62% of its total shares outstanding are floated for trading on the public market, so the liquid part was worth around $21.7B at the time.

The money injected in Tesla for actual utility is the revenue from its product and services sales.

All the cash flowing in its public stock is purely for speculative trading. The stock can’t be exchanged for a car nor be used to access the superchargers.

In this case, the speculation capital is more than 3x the utility capital without accounting for options trading.

Some may argue “but a share is partial equity of a real company, a token doesn’t give any ownership.” That is true. The fact is, it doesn’t matter, and it doesn’t mean much in practical terms.

Even if you own $500,000 worth of Tesla stock, you can’t claim a partial office space nor a tiny piece of its factory. Chances are you won’t even be allowed in.

So are people really seeking ownership or just a speculative instrument that acts as a proxy to the appeal of the underlying asset?

The same exercise could be done for Snap. In 2017, it had a revenue of about $825 Million and the year end market cap was at $14.6 Billion.

32.92% of its total shares outstanding are floated on the public market, so the liquid part was worth around $4.8 Billion.

Snap’s speculation capital is 5.8x the utility capital without accounting for derivatives such as options.

The critical point to highlight is that even if the cash injected for trading is worth multiples the money spent on product and services of those companies; it doesn’t diminish in any way their utility and impact. Even if Tesla’s stock speculation goes to the roof, its products won’t be any less world-changing.

If we agree that the public stock trading is mostly speculative, then the $430 Billion crypto market cap compared to the $77.7 Trillion from the world top exchanges shouldn’t be considered insane anymore, based solely on the monetary figure.

The crypto market is still problematic in many ways but the total amount of money injected is not one of them…

To put things further into perspective, let’s look at the exotic world of derivatives.

Derivatives don’t have intrinsic utility nor provide any direct ownership. According to Investopedia, on the high-end the the global market is estimated at over $1.2 Quadrillion!

Nevertheless, other analysts challenge that figure for being highly inflated.

A more reasonable estimation is the one done by BIS (Bank for International Settlements).

BIS estimates the global notional amounts outstanding for contracts in the derivatives market to be $542 Trillion

The reason the derivatives market is so large is because there are numerous derivatives available on virtually every possible type of investment asset, including equities, commodities, bonds and foreign currency exchange.

If we add Gold, which is estimated at $7.8 Trillion, to our cocktail, we get the following view:

Gold is used as a store of value, a speculative commodity and a utility. It is non-trivial to make a clear cut between each use case as they often overlap. For instance, in some cultures, gold jewelry is acquired both as a fashion accessory and a store of value.

At this stage, the capital flowing into crypto is more of a drop in the ocean than a big bubble.

If the amount of capital is not an issue, then what could possibly go wrong?

1. It could all go to zero

It is unlikely, as we may have passed escape velocity, but not impossible. This is more of an apocalyptic scenario where governments might ban the entire space, where all projects reach a dead end, or where some new quantum computing technology breaks the cryptographic security.

Decentralization and blockchain goal was to create independence from centrally trusted authorities, and it did to a certain extent. There is, however, still the need of gateways between the crypto and fiat worlds. That part can hardly escape government control.

2. ICO scams

Unfortunately, since 2017 we saw a rise in scams with fake projects and impersonation of known people in the space.

This can be solved by good online education and some level of regulation. There is often a negative connotation to the word “regulation” among crypto enthusiasts. I believe it only becomes a problem when states try to force old laws written a century ago into a space invented a few years ago.

If regulation is established on a fresh sheet with the true goal to protect consumers, it should be welcome. Assuming, it is not overly excessive with the sole intention to protect the interests of established financial institutions.

3. Zombie projects

There are two types of zombie projects. The first are legitimate ones that end up in this state because the idea failed, the same way as many non-blockchain startups don’t work out.

The problem is the second type which are zombie projects by design. They are not ICO scams as such but often a simple fork of some other coin. It results in non-differentiated blockchains because the teams had no intention to build a real business in the first place. How many “decentralized X” and “banking the unbanked” have we seen last year?

A few months down the road, the teams abandon the projects, and no code updates take place.

4. Pump and dump schemes

This is a very nasty one as it frequently takes place and affects both “sh!t coins” and legitimate projects.

Those schemes are illegal in regular stock exchanges, but as most crypto exchanges are still unregulated, it is challenging to prevent as of today.

5. Hacking

Many non-technical users find it challenging to manage their private keys securely. Some fall victim of hackers that trick them into sharing their wallet keys over a website and see their funds disappear in minutes.

Even established exchanges with professional network security teams suffered from hacking incidents.

New generation blockchains are trying to address the issue by implementing measures where, for example, beyond a certain amount, a smart contract will prevent the money from moving out of the wallet for a configurable amount of days. This would give time to owners to resolve the problem if they were not the issuer of the request.

In the early Internet days, it was easy to predict e-commerce but hard to foresee Twitter, AWS, and Netflix.

With the first smartphones and mobile broadband, video calls were obvious, but it was difficult to envision Uber, WeChat and Pokemon Go.

The world-changing blockchain and crypto applications are most likely those we can’t imagine at this stage.

So instead of obsessing over the speculative aspect, it would be best to rather focus on solving the problems stated above and open the way to what could be the next digital revolution.

Disclaimer: This is not an investment advice nor an endorsement of any particular asset class. It is simply a personal perspective about the landscape. Seek professional financial and legal advise before any investment decision.

A couple of years ago, I received an Amazon Echo as a gift. It was exciting to see a bold approach to a home speaker that you could talk to. Those were scenes we only saw in sci-fi movies.

From looking at the early promo videos, the device seemed redundant, however. Smart phones’ assistants could also do most of the use cases. I wondered what’s the point of asking “Alexa, how’s the weather today?” if I could also press the phone button and just say “How’s the weather today?”

But beyond whether the actions could be done with a phone or not, the experience on the Echo felt, unexpectedly, very different than any mobile assistant.

The first stunning difference was how natural the voice sounded compared to mobile assistants. The text-to-speech technology was clearly superior but what also made a significant difference was the size of the speaker with louder volume. It gave the voice a sense of presence in the room you never felt with any phone.

The last subtle distinction was that you didn’t need to press a button. The hands-free experience made the interactions feel much smoother and almost human.

Amazon managed to deliver an impressive voice input/output UI that was better than anything out there at that time. But, the engine under the fantastic voice interface was very shallow.

The questions you could ask had to be preset by Amazon or explicitly added as skills. The device didn’t have an advanced NLP (Natural Language Processing) that could figure out the answers to new kinds of questions.

The reason behind this limitation is not bad engineering but rather constraints of the NLP field. Machine Learning progress achieved an impressive performance for speech recognition with a pretty high accuracy. But once the voice is transformed into text form, AI softwares get stuck trying to make sense of it. When the questions with their associated actions are not predefined, AI engines directly trigger a default instruction such as a web search or some form of “I’m sorry, I don’t know how to do that”.

Machine Learning applied to NLP has not yet achieved the same breakthrough as in speech or image recognition. Some argue, ML is not the appropriate approach to NLP, and some think it is. Regardless of what is the best trajectory towards a real NLP, it is clear that we are still far behind.

The question worth looking at though is how can we make the best out of the current voice devices, given the present limitations of the AI field?

Amazon decided to solve that problem by creating the Alexa Skills Kit which gives developers the tools to build voice applications beyond the standard ones. Google later followed the same model and called their skills Actions.

That approach actually solves two problems. First, it’s a workaround for the NLP limitation. Second, it’s a way to create the required integrations with other systems such as lights, thermostats and any other backend service. Those integrations will always be needed regardless of the advances in NLP.

Even if, one day, AI reaches an understanding of intent with 99% accuracy, it would still require an explicit piece of code to instruct it what to do once it understands what people are asking it to do.

Presently, Amazon reports 15,000 skills in the store. Each one has to be manually installed and requires a precise trigger phrase that initiates the corresponding interaction.

The current skills approach “a la App Store” is a good starting point, but is it the best possible user experience with the technologies we have today?

There are 2 paths to improve the overall capabilities and both can be pursued in parallel.

1. Horizontal enhancements

The smart home speakers are intrinsically voice machines. Having to go back to the phone App, browse the store and manually install new skills dilutes the magic.

Given that the Echo store has 15,000 skills, why not provide an “auto-skill” option? Once activated, it would let Alexa search the store each time it encounters a request it doesn’t know, then pick the most likely skill to activate.

This could be achieved by combining manual curation with an algorithmic selection. Amazon could start by making editorial decisions to select the best skills for popular requests, then auto-activate them the first time a customer asks a related question.

For the less popular long tail applications, it could use an algorithm to pick a skill with the best ratings, highest activations in a given location, or a combination of the two. The results could later be changed by users if the default selection is not optimal. Over time, Amazon would get crowdsourced data on what the most popular adjustments are to its default selection, and the quality of “auto-skill” would continuously improve for other users.

It seems Amazon has already started exploring this route. For some requests, Alexa responds with “The skill X could help with that, do you want me to activate it?”. The number of skills that get automatically suggested is insufficient at the moment, but it’s a good start.

This approach won’t always work correctly but would make the interaction flow feel much more natural and would turn Alexa into a pseudo self-learner.

2. Vertical enhancements

Those relate to a predetermined set of services that Amazon, Google and especially Apple should natively build into the devices to near perfection.

When focusing on specific verticals with a narrow context, it is feasible to build interactions that cover most of what’s needed for a sleek dialog and its related actions.

A voice device should excel at audio services. If more general skills are hard to build, there is no excuse for services naturally attuned to sound.

The top-3 are Music, News, and Podcasts. Those should be delivered with exceptional convenience.

Musicologist

When Amazon first took on music it was already pretty good. You could create playlists on the fly based on genres, activities, artists and years. Alexa could also tell you who was singing and the name of the song.

When Apple announced HomePod at the WWDC, they positioned it as a music device with high sound quality that comes with a Musicologist. During the presentation, the speaker distinct ability to adapt to the acoustics of rooms, for optimal sound, was strongly emphasized.

Siri demos were not included in the keynote but on the HomePod page, some examples are disclosed. They look similar to the Echo music capabilities.

Once the product is available later this year, we will find out if Apple raises the bar higher or keeps the music aptitudes at the same level as what Amazon has.

Regardless of the outcome, voice interactions with music are already kind of cool the way they are.

Newsologist

Amazon standardized the flash news update around a single trigger that multiple content providers can use.

Alexa, what’s my flash briefing?

For example, both the BBC and the New York Times are associated with that expression. The Echo will play all static briefings from the activated news skills in a sequence that is editable on the companion mobile App.

Comparing to music, this is a very unsatisfying experience. There is a large room for improvement with the currently available technologies.

It should have the ability to create nonlinear new stations on the fly from various sources based on the topic or requested keywords.

To achieve this, there would be no need for sophisticated NLP as the interactions would be focused on the particular news vertical and related sources.

For example “Alexa, what’s the latest briefing from country X”, “Alexa, what’s the latest tech briefing from today”, or more specifically, “Alexa, what’s the latest briefing about the Tesla 3 launch event?”.

The flow could be similar to how Apple News currently works. There is an initial configuration where media sources and topics of interest are selected. The input is then used to create a general “For You” news digest from multiple sources.

To be fair, there is much more work involved to accomplish the same for audio news. Legal agreements with media outlets would need to be sorted out to unbundle their content and label each piece with the right metadata. Ideally, they would also provide full transcripts of the audio.

If not, speech recognition could be applied to all audio news from the various content providers to enable deeper search. Speech-to-text accuracy is not perfect but it is more than sufficient for indexing purposes.

The other challenge is that audio news is sequential. In contrast, on mobile news Apps, you can have a quick overall glance at titles and photos and decide where to start first then go back to the overall list and select the next article.

With audio, it’s kind of a blind mode where the auto-selected starting point would play first with no foresight of what’s ahead.

One way to solve this is for the speakers to tell the first five titles or topics before playing the sequence. It should also provide the option to jump directly to a specific one.

Another interesting feature would be to add time constraints to the briefing: “What’s my political flash briefing in 5min?”

It is not clear whether Amazon has any strategic interest to invest in building such a service. In general, their devices are intended to serve its other businesses. A more advanced news service on the Echo may not help them much.

Apple, on the other hand, has the complete opposite strategy. Their content offerings such as music and more recently news are there to serve their hardware sales and create a tighter lock-in around their products.

So even if initially, HomePod would most probably not have any advanced news service, Apple is the most likely company to perfect it in the future.

Podcastologist

Podcasts are the missing audio piece without a native support in the Echo. I usually listen on the phone while driving or walking. Accessing the same content on the home speaker would be a great addition.

Currently, the primary source on the Echo is via TuneIn Radio integration. Requests such as “Alexa, play the latest episode of a16z podcast” do work and pull the stream directly from TuneIn. But that’s about it; there is no granular voice search related to specific topics. The experience is more basic than the flash briefings.

Even on mobile, the Podcast discovery feels like Yahoo in the 90s.

Podcasts are a gold mine of educational content provided for free by great thinkers with valuable experiences. Unfortunately, the information is fragmented and buried inside audio clips that are not easy to find.

Podcast content has a longer lifetime relevance than news. Yesterday’s news is not news but podcast episodes from 2 years ago could still be very precious.

Similar to functionalities described in the previous Newsologist section, I would love to see a smart Podcast DJ natively built in HomePod.

“Siri, what are the best podcasts about <topic> I can listen to in 60min”

“Siri, I want to learn about <topic>”

“Siri, create a playlist with all interviews with <person X> from the last 3 years”

Once again, no need for super advanced NLP, we already have the tech necessary to build a flow around such interactions with a smooth experience.

Let’s hope it won’t take too long for Apple to ship the Podcastologist.

There are many good pieces about network effects. Arguably, “All about Network Effects” by Anu Hariharan and “Network Effects and Critical Mass” by Tren Griffin, are among the best.

But, in short, what are network effects?

A product displays positive network effects when more usage of the product by any user increases the product’s value for other users (and sometimes all users) - NYU

The key benefit is to:

Create barriers to exit to existing users and barriers to entry for new companies - Anu Hariharan

In this article, spread across multiple posts, I will share a perspective about the influence technological changes could have on some of the current businesses with strong network effects. To evaluate the impact, we will look at examples from different domains. Each vertical will be covered in a separate post.

In part-I, we will explore the case of ride sharing networks.

The first companies that come to mind are, evidently, Uber and Lyft. They’re the two dominant players in the US and no new entrant has managed to take a relevant part of their business. Factors such as, access to large sums of capital, the ability to handle battles with regulators and their overall execution excellence have played an important role in their success. Nonetheless, what keeps fueling their growth and preventing more competition, are the network effects that kicked-in, after those companies reached a critical mass.

The more drivers join, the better the service gets for riders, and the more riders use it, the better the earnings get for drivers. Those network effects are even stronger within the pooling service. The more riders accept to pool, the more valuable the service becomes for both other riders and drivers.

That has been the major competitive advantage of Uber and Lyft till now. Nowadays, the software part is not hard to replicate. But, growing a new service to the point where it is as valuable as the existing ones is extremely difficult.

Something, seemingly mundane but actually interesting, is happening though. In the Bay Area, many drivers have both Uber and Lyft logos on their cars. Similarly, users have both Apps and would switch between the services depending on the rates and current offers.

It seems that once an alternative service emerges, customers are ready to cross borders from one network to another without any brand loyalty. Nevertheless, there is still a protective limitation coming from human capacity. Hardcore drivers may have 3 or 4 phones on the dashboard. But, in most countries, almost no one would have 10.

Places like Hong Kong, where this driver has 13 phones, are among the exceptions.

Even for Hong Kong though that’s a rare case. Drivers there usually have 3 to 6 devices to handle various dispatch services.

What’s fascinating about the 13 phones example, is it gives an outlook on the future of what people want.

In the sense that people want to participate in as many networks as they can in order to maximize their gains. They’re only limited by the availability of relevant networks and their human ability to use all of them at the same time.

These dynamics will completely change when autonomous vehicles become widely available.

Once cars are fully autonomous, the defensibility of ride sharing networks will drop significantly. The value of their supply of drivers, that has been years in the making, will diminish drastically and, over time, will completely evaporate.

The power will shift from the networks to the car makers, private owners and fleet operators.

The shift will occur because the constraint of a human driver only being able to work for a couple of services, won’t exist anymore. An autonomous vehicle can register to thousands of different networks, listen simultaneously to requests from all of them, pick the optimal ride and tirelessly repeat the process.

The supply that is now locked in a duopoly will start spreading across more and more networks as they get created. Since the marginal cost to join and operate within an additional network will be close to zero, there will be no limit to how many networks a self-driving car could join. Consequently, autonomous vehicles will bypass the protective shield the old networks had.

In the meanwhile, Mercedes-Benz via Daimler signed an agreement with Uber to bring autonomous vehicles to its network and Waymo, the Alphabet car company, signed a partnership with Lyft.

Those agreements indicate that, self-driving car makers think they need existing networks, and existing networks think that they need self-driving cars.

But do they equally need each other?

Current networks don’t have a choice. If they don’t have self-driving cars, they will cease to exist.

Uber is already investing heavily in building their own autonomous vehicles. It sounds like the right way forward. However, it will lead to a fundamental change in its business model.

The model will shift from a marketplace, with strong network effects, to an asset ownership model with totally different economics and market dynamics. Imagine if AirBnB, because of some change in technology, had to own or build rooms. It could still be a viable business but it would be a completely different kind of business. It would be less about the network and more about maximizing assets’ utilization. That’s the business traditional hotels are in.

Uber’s strategy seems to be that owning both the network and the autonomous assets will create a more defensible position. It may be so in the beginning when operations are a hybrid of cars with human drivers and autonomous vehicles (AV). But as the share of AVs becomes dominant, Uber’s own network will become merely a channel among others.

If Uber keeps its self-driving cars locked in its own network, it would run a sub-optimal business. A company that owns bookable assets should maximize the utilization. Especially since an AV, like a hotel room, does not need to rest. The only required down time is for maintenance and cleaning.

Imagine if Hilton relied exclusively on their own booking central to get customers. Many rooms would be left empty.

So to maximize seats’ occupancy, should Uber register its own self-driving cars into other networks?

If they do, they would optimize the returns but at the same time diminish the value of their network.

If they don’t, other AV makers will register their cars into multiple ride sharing services and the defensibility of Uber’s network will be broken anyhow.

That is the dilemma they will face and it’s not clear what the best strategy to get out of it is.

Self-driving car makers, on the other hand, don’t need to rely exclusively on Uber and Lyft services. Those will be just initial channels.

Actually, the most exciting ride-sharing networks haven’t been created yet. They will be the ones natively built for autonomous vehicles with tailored features:

With such an open platform, not only can AVs self-register to a large number of networks, but customers can use universal clients that will scan availabilities and rates across multiple services and select the optimal ride, without even realizing which underlying provider is delivering the trip.

The other technological evolution that will have a significant impact is the transition to decentralized and distributed architectures. This is covered in my previous post Blockchain and Smart Contracts Will Eat Online Marketplaces.

In our current pre-self-driving world, Uber and Lyft excelled with capabilities to:

• Convince a large number of drivers to join in every city.

• Run background checks.

• Deal with regulation across cities and countries.

• Classify drivers as employees/contractors depending on local labor laws.

• Support with car leases.

• Use carrot and stick management with bonuses and punishments when ratings go down.

None of these operations will be needed in the next generation networks.

Uber and Lyft would still have value during the early transition phase. But as the cost of a ride delivered by an autonomous vehicle will be much lower than the current rates, the transition to a fully autonomous supply will be extremely fast.

Once all human drivers retire, the current business model of Uber and Lyft will be tremendously challenged. The barrier to exit to existing customers will become low. At the same time, the barrier to entry for new networks, natively built for the self-driving world, will be small.

Network effects do provide an unfair advantage but over time, major technological evolutions can massively diminish their defensibility. Once the protective armor is gone, the value starts to leak out and the once strong businesses can become extremely vulnerable.

That does not necessarily mean those businesses will disappear. The question is, what should they do to literally reinvent themselves and ensure long term relevance?

Today, the question of building on premise hardware infrastructure versus getting instant cloud computing resources and scaling them over time is irrelevant for most startups.

Similarly, software containers, made famous by Docker, are revolutionizing the way we package, deploy and run software.

One thing that hasn’t evolved enough since the industrial era is how organizations are built and run.

What I’d like to share in this post is a perspective on how to think about building organizations by making the analogy with software containers and how they changed the way we create and deploy applications.

Many years ago, I was working as Product Manager for Ericsson building a customized mobile banking, brokerage and payment solution for one of our customers.

The resulting product was a commercial flop, but that’s a story for another day. The most interesting aspect of that experience was what I learned about building and working with hybrid distributed teams and how it completely changed the way I look at organizations.

The project involved teams from units in Ericsson Netherlands, Sweden, and the UK. Besides, there were people from 3 external partner companies spread out in Canada, India, and Portugal working on different parts of the product.

In the beginning, there was much more commercial and legal work involved to setup all the contracts and scope agreements with the external partners than with the internal Ericsson units.

I started the journey by having a clear distinction in my mind thinking we’re building part of the product in-house, in the various Ericsson offices, and we’re outsourcing other pieces to external partners.

In the early phase, it was painful, frustrating and slow. Some people didn’t get along, some had to be replaced, and one of the external partner companies was sold and dismantled.

It took tremendous efforts to navigate the bumps and get all the external teams onboard with our internal product development methodology and ways of working.

However, as people got to know each other better and got used to working in common ways, the lines between the external teams and the internal ones became blurry, and over time those lines completely disappeared on the day to day work activities.

Traditionally, in many companies, any team or function that is not part of the in-house organization is considered an outsourced activity.

What I realized is that the assumed separation goes away when the external “outsourced” teams get to know the internal ones well enough personally and start to operate through common communication routines, problem-solving sessions, transparent information sharing and progress reviews built on the top of a well-structured baseline with clear split of responsibilities.

So in a way, the fact that the external teams belong to different legal organizations, are located in various countries, are subject to different labor laws, have different employment contracts, and different vacation policies doesn’t matter much.

What is important are the actual people that belong to those external teams and their ability to learn how to interact and commonly work with each other. Once they reach a certain level of close cooperation, adhere to shared routines and trust each other, they don’t seem anymore as “internal” and “external” but become part of one single whole.

“That single whole is what should become the new perception of an organization.”

So what does all of this have to do with software containers?

For those not familiar with the concept, here is an excellent introduction by Preethi Kasireddy

The advantage is that each piece of software inside a container can run in its isolated environment with its own internal specific attributes but at the same time, all pieces can have the ability to interact and communicate with each other.

More and more, modern software products are designed as a collection of small containers that each accomplishes one function and interacts with each other. When these different small containers are orchestrated by a common entity, they become part of one single whole which is the overall application. (also referred to as microservices architecture)

Put in a simplistic way, the most important aspects of each container that is part of an overall application are:

• The actual software running inside

• The way it interacts and communicates with the main application orchestration layer and with other containers

• Its status (running or not)

The least important parts are:

• The physical space where it’s running

• The internal file system hierarchy

• The internal process tree

Now, imagine if you have multiples teams spread out in various locations and working for different legal entities to which you are “outsourcing” some of the work. The work can be specific pieces of the product, design or anything else.

Every team belongs to different legal organizations with their own management hierarchy, internal work processes, office rules and employment contracts.

These external teams can be considered as small autonomous parts, similar to software containers, that interact with each other and are orchestrated by your overall host organization while abstracting away all their internal details.

In this case, the most important parts become:

• The actual people and talent inside these small independent teams

• The way they communicate and collaborate with the main internal host organization and other external teams

• Their status. Meaning whether they are in active working mode on your project, waiting mode or stopped mode

The least important aspects are:

• The physical location of the office space where they are located

• The legal entity they belong to along with its internal management hierarchy

• The internal work processes and office rules

The analogy between an organization based on distributed external teams and an application based on software containers can be visually represented by something similar to this.

For example, if a software application has a certain piece deployed on premise using physical servers and other parts are deployed on cloud servers provided by an external company, we won’t consider the first part as internal and the second one as external. They are both parts of the same whole which is the overall application.

The same way modern application architectures based on containers are improving how we design, build and deploy software, we need to push forward the way we build and run organizations.

We should move beyond the strong separations between internal, external, full-time, part-time, freelancer, contractor, subcontractor and start putting the importance on the actual people and improve how they work together instead of focusing on the underlying legal and physical structures where they belong.

However, there are challenges that come with such a model.

Throughout my experiences, I had both great outcomes and disastrous occurrences.

1. Not everyone is yet comfortable with remote and distributed work

Whether it’s one person or a team working remotely and regardless of their legal relationship with your company, the most important aspect is that people are comfortable and ideally prefer that kind of setup. If not, it would be tough to overcome the other challenges. So it is crucial to take that into account when you are initially assembling the team by openly asking each person how they feel about it.

2. Time zone differences can be challenging

In some cases, you may have groups in different continents with 9 to 11 hours time differences. This is always challenging when you need to setup video calls with everyone at the same time. The time slots that fit the regular workday of both are limited.

But at the same time, it can be a blessing. I had cases where I gave comments to a remote team late at night, then went to sleep. When I woke up in the morning and found the updated version of the product with all the modifications included, it felt like magic!

3. Lack of informal human interactions outside work

This one is tough. These days we have various video conferencing, collaboration tools and even remote presence robots such as Beam to bring people closer. In the coming years, we may also get some good VR solutions for that.

Nevertheless, nothing can yet replace direct interactions and the level of human connections achieved by spending time together and bonding.

The best way to overcome this is to arrange for all teams to meet, ideally, in the beginning of every major project and at least once every year in the same place. As the size of the overall organization gets bigger, it will get harder but the benefits are significant.

So isn’t it easier to simply stick to the traditional monolithic organization?

It seems easier because the management and organizational practices to build and run a traditional organization based only on full-time employees, in the same location, have been around since the industrial era. When it comes to hiring, structuring the hierarchy of teams, compensation, performance management, communication, culture, … Most companies are built on practices that have been used for many decades. Some have their own twist and specific internal ways but overall, it’s derived from the same base model.

In contrast, we still don’t have a widely agreed on set of best practices to build and run a modern organization based on a distributed mix of full-time employees in a central location, full-time employees in remote locations and freelancers/contractors from other external companies. This is mostly done ad-hoc and is still work in progress.

What we do have though is plenty of modern collaboration and communication tools such as Slack, Trello, Asana, Dropbox, Google Apps, Github, Zoom, … that bring people closer, make work more efficient and provide a more transparent information sharing.

So the technologies are already available and will keep getting better. The last piece to nail is a well tested set of practices, around those tools and around people, that can be easily adopted by every company to bring all those distributed hybrid teams together and have them operate as a well oiled single whole organization.

One company that seems to have, at least partly, cracked the formula is GitLab. They have 160 employees spreadout all over the world.

It’s worth watching the full interview with their CEO Sid Sijbrandij and Ali Rowghani from YC.

According to Sid, they didn’t plan to build a distributed organization from the beginning; it happened naturally:

“After a few days, they just started working from home”

They documented every aspect of their ways of working in a handbook publicly available.

Another great example is InVision. One of their engineers describes how it’s done in this post called Making remote working work.

The benefits of building such an organization can be huge.

1. Access to talent that you would otherwise not get

Whether you’re a new startup or an established company, you’ll be missing out if you only consider available talent in your location. Especially if you are in a place where hiring is extremely competitive or the pipeline of people with skills you need is limited.

By tapping into a global pool, you will not only get potential people with great skills but you’ll also get more diversity into your organization by bringing different international perspectives.

2. Efficiency & flexibility

As stated previously, time zones can be challenging, but you can also use them to your advantage to make your work flow more efficient.

Another aspect is about activities that are not required full time 12 months per year. For example, if you’re a startup building a hardware product. In certain phases of product development, you will need top industrial design, and mechanical engineering talent. At the peak, that part of the team will be needed full time but beyond certain milestones, there won’t be much activities around those functions. If the team was on a full time employment contract, it would have a negative impact both on the company and on the team members. Indeed, the best people will quickly get demotivated if they don’t have challenging work to do and will quit if they don’t have enough load.

So having a project based approach with external teams that are assigned for a specific scope and specific period of time will work best for both parties.

Lastly, there are also cost efficiencies that you can achieve by saving on office space and by tapping into international markets with lower costs of living.

3. Variety of projects makes people more creative and keeps them excited

Some of the best people don’t like to work on the same product/project for 2–3 years.

I once hired a freelancer for a specific project activity initialy for 3 months. At the end, I was so happy with his work that I offered him a full time employment. I was stunned when he declined the offer.

His reason was that our project was exciting but he could not see himself working on the same kind of product all the time. He wanted more variety and was still curious to learn new things by getting involved with different companies. He was happy to come back for another assignement with us if it involved something new.

This may seem as a constraint at first but it’s actually of great benefit. Having people onboard with a broader exposure to other industries and type of products will bring to your organization new creative ideas. You would otherwise not get those insights from somebody that had been inside your team working on the same product for many years.

Even with the challenges, the benefits of designing your organization as a collection of hybrid teams regardless of their location and legal relationship with your company are well worth pursuing.

This is where work is headed and the earlier you start experimenting with the model, the easier it will be to fully implement it as your organization grows.

A few months ago I wrote a piece called What Kind of Marketplace Are You Building? It covered the structural differences between the main types of marketplaces.

In this follow-up post, I’d like to address some external factors that could impact the model in the long run.

Online marketplaces still provide one of the best business models out there. They tick all the key boxes including scalability, network effects, no assets ownership, no inventory and fast global growth.

They produced many successful giants that have drastically improved several industries in the past 20 years.

The question is

For how long can this model stay viable? Does it face any potential existential threats?

The most successful marketplaces came to life because they managed to remove middleman entities, that were mostly brick and mortar operations, and replaced them with an automated piece of software accessible over the Internet.

By doing so, they provided a much better user experience with 24/7 access, massive global reach, simple and quick payment mechanisms, transparent reputation rating, realtime services and lower prices by eliminating the need for expensive manual operations.

There’s one important aspect to highlight though. The model is often explained as “removing the middleman” without emphasizing that what actually happens is not a complete removal of the broker in the middle, but rather a replacement of the old broker with a new kind of broker. A much better type of broker, built around a software platform, but still a middleman entity between the supply and demand sides.

That centralized software architecture is how online marketplaces are able to capture value and is the critical building block of their business model.

For almost two decades, nothing seemed to disturb that position. There wasn’t any other technology that could be more efficient than an automated central software to handle the relationship between the supply and demand sides.

Nevertheless, in the past few years, something interesting started to happen, in a totally different domain. Bitcoin, the now famous cryptocurrency, was designed around a distributed architecture that used a public decentralized ledger, called the Blockchain.

The Blockchain architecture allows peer-to-peer Bitcoin transactions to take place in a secure way without the need for any central entities such as banks.

It didn’t take very long before Blockchain type of architecture started to appear as the ideal distributed trust management system the Internet was waiting for. It could be used to solve peer-to-peer trust problems beyond the original Bitcoin transactions and wallets.

We finally have the conceptual framework for such a system, one in which trust need no longer be invested in a third-party intermediary but managed in a distributed manner across a community of users incentivized to protect a public good. Oreilly

The Blockchain mechanism is well described in the same article from Oreilly:

This technology creates a mechanism by which people and institutions that would not otherwise trust each other can agree, in a constantly updated process, on a common record of events. The blockchain is an unbroken, sequential ledger that’s replicated in multiple copied versions and stored on multiple, independent computers. Each node runs the same open source software that dictates how to update the shared ledger with new transactions and how to arrive at a consensus on whether those updates are valid.

The Blockchain provides a solution to the trust and data sharing problem without the need for a central database controlled by a third-party.

What it doesn’t directly provide is a system to implement the business rules and logic that online marketplaces need. For example, revenue split agreements, dynamic pricing, and insurance payouts when certain criteria are met.

This is where Smart Contracts come into play.

Smart contracts (also called self-executing contracts, blockchain contracts, or digital contracts) are simply computer programs that act as agreements where the terms of the agreement can be preprogrammed with the ability to self-execute and self-enforce itself. The main goal of a smart contract is to enable two anonymous parties to trade and do business with each other, usually over the internet, without the need for a middleman - blockchaintechnologies.com

Smart Contracts represent the final piece of the puzzle that was needed to completely replace the middleman by a real peer-to-peer architecture.

Blockchain and Smart Contracts, provide the tools and framework to create a new generation of marketplaces where supply and demand sides can engage in trusted trading transactions, according to various business rules, without the need of a central brokerage entity.

Consequently, the same way online marketplaces disrupted many traditional brick and mortar businesses, Blockchain and Smart Contracts will give birth to a new kind of peer-to-peer marketplaces that will unsettle the current ones.

Presently, that seems to be the biggest threat to the online marketplaces business model that has been incredibly powerful for the past couple of decades.

The shift will not occur in a year or two but it’s their most serious long term strategic risk.

This is one of the cases where it’s not a matter of “if” but rather “when” the new decentralized structure will take over as the technology keeps maturing and improving.

The peer-to-peer structure is the perfect fit for the decentralized nature of the Internet and has the benefit of handing over the power back to the people participating in the network.

Some interesting experiments are already taking place:

1. OpenBazaar

OpenBazaar is a different way to do online commerce. Instead of visiting a website, you download and install a program on your computer that directly connects you to other people looking to buy and sell goods and services with you. This peer to peer network isn’t controlled by any company or organization — it’s a community of people who want to engage in trade directly with each other.

2. La’Zooz

A Decentralized Transportation Platform owned by the community and utilising vehicles` unused space to create a variety of smart transportation solutions. By using cryptocurrency technology La`Zooz works with a “Fair Share” rewarding mechanism for developers, users and backers.

3. WeiFund

WeiFund is an open platform for crowdfunding campaigns. You can launch a campaign using one of WeiFund’s contract templates or integrate your own smart contracts…

WeiFund is built on Ethereum.

We may witness a true commerce revolution, once some of the ongoing trials start to work at scale.

Does it mean current online marketplaces are doomed?

The surprising part is that today’s marketplaces with the best unit economics are the most at risk and the ones with much tighter margins are relatively more immune.

It sounds counter-intuitive, but businesses with the best unit economics are usually highly automated. Almost all of their operations are managed by a central software. Human intervention is kept minimal and mainly handles activities such as regulation, fraud and customer care.

Paradoxically, when marketplaces are highly efficient around pure online operations, they become the ideal target for replacement by a decentralized peer-to-peer model.

On the other hand, marketplaces that require non-automated operations, in addition to online operations to provide their services, are much harder to displace.

A good example is Amazon marketplace part of the business. (Putting aside the e-commerce, cloud, content and other areas).

Amazon offers independent sellers the possibility to list their products on their platform. Software fully automates the product discovery, ratings, and payment handling. But, Amazon also offers a service called Fulfillment by Amazon that can pick, store, package and deliver the orders on behalf of the sellers.

You sell it, we ship it. Amazon has created one of the most advanced fulfillment networks in the world, and your business can benefit from our expertise. With Fulfillment by Amazon (FBA), you store your products in Amazon’s fulfillment centers, and we pick, pack, ship, and provide customer service for these products. Best of all, FBA can help you scale your business and reach more customers.

This gives Amazon a much stronger position by combining the online platform scale with the magnitude of a physical network of fulfillment centers and their associated operations.

Blockchain and Smart Contracts can not displace that kind of combined online and physical services.

So what could other marketplaces, that are presently built around simple online operations, do to reduce their long term risk?

They could consider adding new layers of services on the top of their existing online activities. These new services should complement the existing ones and make the whole offering even more attractive to the end users.

A couple of examples:

Home sharing marketplaces could add a fully managed operations service, where hosts would hand over the keys, and everything would be taken care of. Not only finding guests and getting payments but also handling check-in/check-out and cleaning.

Freelance marketplaces could add a service where a job is requested for a particular scope, and they would take care of finding the right candidate and manage the project by committing to a timeline, price, and quality of deliverables.

These kinds of offerings may sound terrible because they come with higher overhead and lower margins.

Still, they would make the marketplaces much stronger in the long run and would reduce their risk exposure to the upcoming disruptive peer-to-peer structures.

Many have been predicting the inevitable decline of linear TV. Looking at the trends from the past few years, it is clear that overall viewership numbers confirm the ongoing drop with the impact being strongest among the younger age groups.

Instead of trying to predict what will happen next, I decided rather to share what I would like to see happening, as a consumer, for what would be an ideal personal TV experience.

3 years ago I ditched my linear TV subscription simply because my most common use case for it was a background ambiance while I was doing or looking at something else.

My home TV screen content became Netflix and iTunes primarily via an old Apple TV box. When Apple announced the new ATV with Apps support, I was very excited. Having the ability to pick and chose from whichever channel or content creator, that deployed their App on the platform, seemed wonderful.

The honeymoon lasted a few weeks while testing and playing around with more and more TV Apps as they were shipping. However, over time the overall experience ended up turning not as great as I had imagined.

So, what was happening? Did I start missing the old linear TV bundles?

On the good side, TV Apps do give you control to select exactly when and what you want to watch with a much nicer UI than the EPG of Linear TV. The experience works well as long as you stick to only a couple of Apps. Let’s say Netflix + ABC or Netflix + HBO NOW+ NBC. But once you get close to ten channels, the usability deteriorates. You need to quit the current App, launch another one and each App has its own menu structure, its own account, its own “recommended for you” listing, its own wish-list, its own selection of shows that overlaps with other sources, and some have their own paid subscription.

Apple has enabled a search feature with Siri that partly solves one of these problems by allowing you to search for a show then get all the sources to watch it without having to browse inside each App separately. But that only works if you know exactly what particular title you want. If you don’t, there’s no way to discover what’s available in all channels from one place.

Indeed, the critical missing element with TV channels as Apps model is a common discovery layer that exposes all the content under one interface, one recommendation engine, one wish-list, one profile and eventually one subscription, abstracting away the underlying source channels.

What’s needed, in a way, is doing for TV what Flipboard had done for web content. Consumers still have the option to directly go to the original website of each publisher to look for interesting things to read, but Flipboard provides a unified interface to discover auto-curated content based on your interests or by specific themes from a vast number of publishers without having to necessarily dig into each publication separately.

Having such a standard discovery UI for all TV Apps is the next step I’d like to see happening and would be a significant improvement over the current situation.

Yet, that wouldn’t be enough

A common discovery layer would take us to a level as good as what we have today inside the Netflix interface but with content automatically coming from all other TV channel Apps.

What would still be missing is passive discovery. A mode for laid-back moments, when you don’t know what to watch and don’t want to spend a long time browsing through the different lists such as “Top picks for you”, “Recently added”, “Because you watched X, you may like Y”,…

Some solutions to that already exist in the music world. Pandora first introduced the creation of radio stations based on a music style, a song or an artist. A playlist is automatically generated providing a listening experience that had the benefits of linear broadcast radio stations while being personalized specifically to your taste and mood of the moment.

Later on, Apple Music (putting aside the confusing UI) came up with a hybrid model where artists manually make the playlists that are then automatically served to users by an algorithm based on style preferences and likes.

Perhaps, the best example so far from the video world is the TED Talks App. It has been around for many years with a cool “Surprise Me” feature.

You start by selecting a theme

How much time you have

Then the App generates an automatic playlist with a selection of videos that fit both the theme and your timeline

I always thought it was brilliant and wondered why a similar flow hadn’t been applied to other video discovery engines.

Imagine such a feature deployed on the top of a unified discovery interface for TV Apps allowing you to automatically generate customized pseudo-linear channels based on your current mood, or reference content and taking into account the time you have available. It could also create a generalist channel programming that updates automatically every day by combining your different interests and feedback over time.

For something like this to happen, the content inside the various Apps should be completely unbundled and dynamically re-bundled for each user in a personalized way.

In order for such a system to become truly great, it shouldn’t be limited to TV Shows and Movies. Every single piece of TV programming should be unbundled including the flash news, talk shows, special reports,…

The technologies to build it, such as Apps deep linking, already started to emerge and will keep evolving. The greatest challenge won’t be much on the technical side but rather on the legal and licensing agreements that are way more complicated than in the publishing world.

It would be quite hard, but not impossible, for a standalone new comer to build such a platform. Players like Amazon and Apple with a strong negotiation power are best positioned to pull it off.

In the meanwhile, Apple is rumored to be preparing a linear programming offer for the ATV.

The question though is why would they even try to do that?

There are benefits of having linear TV on the ATV box, or any other one, but they are rather limited. The best one is having access to live sports events, then maybe better bundle prices and most likely a nicer EPG.

But it will still be like “running the old on the new” with no significant shift in how the content is discovered and consumed.

For a real reinvention of how consumers interact with TV, the channels should be willing to fully unbundle their content down to the smallest piece of programming and make it accessible from horizontal discovery layers built by others.

The experience for the end consumer would, for sure, be better but it is not clear if the economics would be attractive enough to the channels.

If not, chances are they would keep resisting the change for as long as they can.

Last month, I was mentoring the teams participating in the AngelHack Hackathon, and as in most similar events, there were many Marketplace ideas.

Online Marketplaces have been around since the early days of the Internet and have widely proven to lead to extremely successful and scalable businesses.

They evolved from the initial horizontal model made famous by eBay and Craigslist, where practically anything could be traded, to various vertical models optimized for specific categories and focused on better serving customers’ needs in those categories; e.g. “Cleaning”, “Rental”, “Laundry”, “Ridesharing”, …

There is a good piece by Jeff Jordan “People Marketplaces” that covers well the “Vertical vs. Horizontal” question:

“Our hypothesis is that the horizontal plays may suffer from a potential “paradox of choice”: Consumers could be getting overwhelmed by the seemingly infinite array of potential service options presented by horizontal platforms, but consumers can easily understand the highly specialized value proposition of a company offering services in one vertical”

and

From a product perspective, the vertical apps can tailor their workflow to the unique characteristics of that vertical — the best way to find someone to clean your house is different than the best way to find a ride.

However, regardless of whether the marketplace follows a vertical or horizontal model, other differentiating elements make some more likely to be hugely successful than others.

In the best situation, anyone from the demand side should be able to tap into anyone or anything available on the supply side. That’s the case, for example, with AirBnB. They have a high liquidity and a strong Network Effect. Every new user makes the marketplace directly more valuable for everyone else. Moreover, potentially every guest can be a host and every host can also be a guest. This is, in a way, the ideal type.

On the other end of the spectrum, there are examples such as Helpling, former Homejoy, and Handy that offer home cleaning and maintenance services. In those cases, multiple and distinct supply and demand sides have to be built and scaled, not only in every city but also, in every neighborhood. The reason being that, for instance, it makes little economical sense for a cleaner to accept a job from a customer living in a far away part of a city. Transportation cost and commute time would offset a large part of, if not all, the profit. At the same time, a customer would only need the service when at home and therefore would never tap into the available supply of cleaners in other cities even when traveling to those same towns.

This type is, kind of, a “Collection of Marketplaces” rather than one single large Marketplace. Although it can still turn into a successful company, it is much harder to build and scale. Indeed, it has less overall liquidity and a relatively weaker network effect. So the Egg/Chicken problem, common to all marketplaces, is even harsher. Consequently, it takes a longer time and requires more resources to reach a critical mass and maintain growth.

Somewhere in-between the two previously described types, there is a third one. Most famous examples are Uber, Lyft and similar ride-sharing services.

In this third type, there is still a need to build multiple supply and demand sides in different geographical locations for the model to work. However, the advantage, in this case, is that supply can move around and benefit from the available demand in other areas. E.g., drivers can choose the region where they want to work depending on the time of day and the day of the week. Similarly, customers can take advantage of different available supplies by requesting rides in various areas both inside and outside the city and also when visiting other countries. So even if multiple supply and demand sides have to be built, they are not completely segregated as in the case of a “Collection of Marketplaces”. Subsequently, this type is relatively close enough to a single large Marketplace type and still benefits from a similar strong network effect.

To summarize:

Type-1 is one large Marketplace where anyone from the supply side can trade and interact with anyone from the demand side. E.g., eBay, AirBnB. In these two examples, there can even be an overlap between the supply and demand sides. A seller can also be a buyer, and a buyer can also be a seller.

Type-2 is a “Collection of Marketplaces” where distinct supply and demand sides have to be built. Each subset of the demand side only trades and interacts with its corresponding subset of the supply side. E.g., Handy, Helpling.

Type-3 is an in-between hybrid type where multiple supply and demand sides still have to be built but where each subset of the demand side is not limited to only trade and interact with its corresponding subset of the supply side. Here, the different supply sides and the various demand sides can interact with each other as well. E.g., Lyft, Uber.

For anyone building an Online Marketplace, one of the key questions to consider is:

Does the service you want to offer require one large Marketplace (type-1 or type-3) or rather a “Collection of Marketplaces” (type-2)?

Both can lead to successful outcomes. Though, the latter is much harder to build, takes a longer time, requires more resources and is much more difficult to scale.

Now, beyond the overall types, the “Vertical vs Horizontal” question and the Egg/Chicken problem, there are some additional subtle aspects to consider.

Convenience vs. Value

Simply put, ask yourself if the service offered is a “nice to have” convenience, something of actual value, or both?

This is covered in the same post by Jeff Jordan:

It appears that the early breakouts in the space are those that offer both convenience and value. It’s clear that the market size of people who are willing and able to pay a premium for convenience is much, much smaller than those who are attracted to both convenience and value. For example, a service that charges a premium to come to you to wash your car may work well on Sand Hill Road, but it’s unlikely to have broad national appeal and disrupt the physical car wash industry.

Personal vs. Impersonal Interaction

There are significant differences in the depth of interactions across the various categories.

When it comes to services such as transportation, whether a customer knows the driver or not and the fact that for each ride a different driver will most likely show up, don’t matter much.

Some drivers/riders can be more chatty/friendly than others but overall the total customer experience is very little impacted by the depth of interaction as long as the transportation service is properly delivered.

In contrast, for services such as “Pet Sitting” and Babysitting, the interaction becomes more personal and the level of affinity and trust the customer has with the person providing the service matters tremendously. In situations like these, there is a strong tendency for customers to prefer dealing with the same person everytime.

This brings additional challenges that could lower the overall liquidity and cause Platform leakage.

Platform Leakage

This is when transactions start occuring outside the marketplace via a direct interaction between the two parties.

In the Babysitting example, once the first matching is done through the platform, there is a possibility that the subsequent requests and payments are arranged directly between the two parties.

Customers may use the platform again to get matches with some additional Babysitters but once a personal relationship is established with a handful of trusted providers, there are less incentives to go back to the marketplace.

One way to solve that problem is to provide some value in addition to the matching and transaction processing.

For instance, “Pet Sitting” marketplaces offer an insurance with generous coverage in case something happens to the Pet or if the Pet causes a damage to the sitter property.

Marketplaces are hard to build. Nowadays, creating the product is, by far, the easiest part. The toughest work is what happens next.

Being aware, upfront, of the different types, models and their underlying constraints will make whomever wants to start a business in this space better equipped to ask the right questions and think about which strategies to adopt before jumping into the long entrepreneurial journey.