Generative AI represents a set of new large machine learning models trained on vast amounts of data, with billions of parameters, which through scale, and sheer computing power via AI supercomputers, can, by performing text-to-text predictions, become general-purpose engines, able to perform a wide set of tasks, from natural language understanding to coding, speech recognition and much more.
How did we get to Generative AI?
The Internet moved along four key eras, which I can summarize as going:
– From proprietary networks to browsing
– From browsing to search engines
– From search engines to discovery engines
– And from discovery engines to generative engines!
This is how the world looked like until 2022, before the launch of ChatGPT.
In a world driven by tech giants that are locked in the digital distribution pipelines to reach billions of people across the globe, the gatekeeper hypothesis states that small businesses will need to pass through those nodes to reach key customers.
Thus, those gatekeepers become the enablers (or perhaps deterrents) for small businesses across the globe.
Therefore, this hypothesis moves along the way of understanding what the next gatekeepers, which will control and intermediate attention, between billion of users across the world and millions of small businesses across the world are.
The Walled Garden Era
In the early 1990s, the Internet was not directly accessible by users, or at least not by the majority of users.
In fact, the Internet was not easy to access, and the simplest way to do that was via a set of large portals (like AOL, Prodigy, CompuServe, and GEnie) which enabled users to access a set of services, from news to email within these portals.
To be clear, those portals were not the Internet; they were simply proprietary networks organized around a few core services intended to make people access these services in a controlled manner.
These first tech giants were companies like AOL, Prodigy, CompuServe, and GEnie. Their business model was straightforward; you would pay for a monthly subscription and enjoy a few hours of access to their proprietary networks.
For each additional hour users spend on top of the subscription, they would be charged based on consumption.
Yet, things would change, by the mid-90s, as a new player came to market: the browser.
Browsing as the commercial killer application
By the early 1990s, it had become clear that to make the Internet viable, a tool, which enabled people to access that easily, would have been an incredible utility for the nascent web.
In that era, the tool which would open up the way was the Mosaic.
Started as an academic project, Mosaic was developed by the National Center for Supercomputing Applications (NCSA) at the University of Illinois at Urbana–Champaign.
Mosaic, the first web browser, changed that!
Yet, when the National Center for Supercomputing Applications (NCSA) failed to commercialize Mosaic, the team that had built it (led by Marc Andreessen, which would end up creating a16z, the venture capital firm which invested in the most iconic tech startups of our times), flew to build a new browser: Netscape!
Netscape was a game-changer. So much so that it awakened the sleeping tech giant of the time: Microsoft.
At the same time, Microsoft and its founder, Bill Gates, understood the importance of the internet. Back in January 1996, Bill Gates wrote one of the most quoted pieces, still nowadays “Content is King:”
One of the exciting things about the Internet is that anyone with a PC and a modem can publish whatever content they can create. In a sense, the Internet is the multimedia equivalent of the photocopier. It allows material to be duplicated at low cost, no matter the size of the audience.
Yet, still in 1996, Bill Gates had created a dedicated team with the sole purpose of bringing TV to the internet. However, that didn’t work.
The main problem of Gates’ vision at the time was the fact that Microsoft had tried to linearly apply TV to the Internet, as if, the two technologies would follow the same pattern of development.
Over the years, streaming took over the Internet (see Netflix), but in reality, it evolved in completely different ways, and formats (see Netflix binge-watching).
Yet, while Microsoft realized the potential of the Internet early on, it had executed badly on it. And by 1996, it was clear, that “browsing” was the killer commercial application of the Internet and it had a new king.
 |
A featured in Time Magazine in February 1996, a shoeless Marc Andreessen (at the time 24) was featured as the new king of the Internet, an image that Bill Gates didn’t like at all.
That was the start of the “Browsers war.”
Finally, Microsoft had a hard awakening, as it figured the commercial killer application of the Internet would have been the browser!
Indeed, until that moment, Microsoft had spent billions on the Internet.
Yet, the vision Bill Gates had in mind was called “Information Superhighway,” and it was more like a sort of TV online (just like in our days, Zuckerberg spent billions on a vision of the Metaverse, which proved not commercially viable).
As explained by Jim Clark (co-founder of Netscape) in his book “Netscape Time: The Making of the Billion-Dollar Start-Up That Took on Microsoft,“ in 1994, Netscape was launching the browser which would conquer the whole browser market share, thus, becoming, for a short period of time, the Internet!
On the other hand, Microsoft, strong in its dominant position, started to declare war on Netscape.
Indeed, Microsoft had become the de facto operating system of the PC industry, and within a decade, the company became a tech giant.
In 1982, Microsoft recorded over $24 million in revenues, in 1985 over $140 million, and by 1990 Microsoft had passed a billion dollars in revenue, the first tech company to do so, and becoming the de facto operating system of the PC market.
While, at the same time, in 1999, AOL had a market capitalization exceeding $200 billion.
This matters to understand what would happen next.
That Netscape’s threat further created a sense of urgency for Microsoft’s top leadership, who started to invest massively in its browser.
By 1995, Netscape had the most successful browser on the market and had developed it by bringing together the original team of Mosaic.
Netscape truly became a Mosaic killer. From over 90% browser market shares of Mosaic, in 1994, by 1995-6 the situation had turned upside down.
Netscape had reached 80% of the market share, whereas Mosaic adoption had stalled (Netscape took a growing pie of the exponentially growing Internet users’ base). Netscape did that through fast releases and built-in network effects.
 |
By 1996-98, Internet Explorer, thanks to the incredible distribution strength of Microsoft, was stealing market shares from Netscape, thus stealing its market leader position.
Indeed, Microsoft was simply bundling its Internet Explorer within its Office Package, thus making it the default choice for users to browse the Internet. Nothing new to Microsoft, which had used the “bundling strategy” for years now.
Yet, the fading giant AOL acquired Netscape in its attempt to fully convert to the new rising wave of Web 1.0, made of browsing first, then searching the commercial Internet.
Search as the commercial killer application
When Microsoft established itself as the de facto dominant player of the nascent web, it also did that through sheer force. Pretty much, Microsoft had taken the code of Mosaic which was eventually commercialized by National Center for Supercomputing Applications.
It built a mediocre browser, Internet Explorer, which eventually, after a few updates, improvements, through bundling (inserting Internet Explorer as a free tool into Microsoft Office and Windows) it won the bworser war.
The use of sheer distribution and bundling was seen by the antitrust as an abuse of dominant position, which led to a famous antitrust case, against Microsoft, with a grumpy Bill Gates, subjected, in 1998, to hours of deposition!
This made Microsoft unprepared for the next war: search!
In fact, by 2000, Google had become very popular, yet it still recorded over $14 million in losses on a $19 million turnover.
Yet Google had been backed by Sequoia Capital, led by John Doerr. Indeed, in 1999 Google received $25 million in equity funding from Sequoia Capital and Kleiner Perkins.
As Larry Page announced at the time:
We are delighted to have venture capitalists of this caliber help us build the company, we plan to aggressively grow the company and the technology so we can continue to provide the best search experience on the web.
At the time, PageRank solved an equation of 500 million variables and two billion terms to serve proper search results.
The market was still small (one hundred million web searches per day compared to the over eight billion web searches that today go through Google).
Six years later, to that conversation, by 2005, Google had reached a hundred billion market cap. By 2021, the Google advertising machine would generate twice that in revenues alone, making Google (now called Alphabet) a 1.8 trillion dollar company!
Among the most important deals Google sealed in the early years, there was the deal with AOL. Indeed, AOL saw search as a feature within its proprietary network, thus featuring a search engine that enabled users to surf the web.
Initially, GoTo.com (later called Overture) had a deal with AOL to be featured as the main search engine.
Yet as this contract expired by the late 90s, Google jumped on it, managing to get the deal with AOL in place of Overture! This deal was a very important one, which put Google on a further growth trajectory.
That was the era of search, and it would last till these days.
Discovery as the commercial killer application
Until the 2010s, Google had been the undisputed winner of the Web 1.0 era, becoming a de facto monopoly in search. However, another technological innovation was coming to reshape, once again, the tech industry: the social graph.
Back in the 2010s, a few years after Facebook had been founded, it represented one of the most impressive growth companies that ever existed, also for the Internet standard.
Indeed, Facebook and a few other companies opened the way, to what we call – in hindsight – Web 2.0.
In an early 2004 interview, Mark Zuckerberg, 20 at the time, side by side with WesMatch’s founder, Dan Stillman, explained what Facebook was about. Facebook would eventually wreck down WebMatch and all the other early players in the social network space (Myspace, Friendster, and CampusHook to mention a few).
As Zuckerberg explained it at the time, Facebook was “an online directory that connects people through universities and colleges through their social networks.”
While today we give for granted a Facebook with billions of users, at the time with a hundred thousand users on the platform, it was very hard to tell how big it could become.
Facebook’s early growth trajectory (Source: Financial Prospectus).
When Facebook first launched, it was a rocket ship. The company used a staged rollout, where it would open its app to a larger and large set of users, not gradually, but exponentially.
In short, with its first release, in February 2004, it only proved the concept through Harvard. As growth picked up right on, and the product turned out to be very sticky among students at Harvard, Facebook opened to other top colleges, by the same year.
Leveraging a powerful social graph, it leveraged network effects, and it quickly grew, from a social network for universities to a mass social media company.
The Facebook Social Graph of the early days! (Source: Facebook Prospectus).
Facebook was an incredible innovation for its time. With a very simple interface, users could upload photos, update their status, and send messages to their friends and have complete control over what they wanted to share:
That created a new era of discovery. No longer based on users searching for something.
But rather, a set of algorithms (which would, over the years, be optimized by leveraging AI) that would push interesting stuff to users in a feed-like experience that resembled more a slot machine that could be continuously scrolled rather than a newspaper.
Web search still was a key commercial application of the Internet. And yet, it was in part eclipsed by the new discovery experience.
Google itself transitioned more and more to become a discovery engine, where it started to push advanced features and portals (like Google News and Discover) able to push information to users based on their interests, not on their search intent.
A major example of this transition was YouTube, which was acquired by Google, in 2006, for $1.65 billion, and it was turned from a video search engine to a discovery platform.
Where billions of users each day could find, through a feed, a bunch of new videos based on their preferences.
The discovery era began with Facebook and further took over Instagram. At the same time, it scaled to become the key mechanism for online communication when a small startup from Venice Beach, in LA, was about to disrupt the social media space.
That startup, Snapchat, had popularized a couple of new formats (from disappearing pictures to stories), which had become the standard of social media experiences.
Snapchat, in 2013, refused a $3 billion purchase offer from Facebook, which turned into a fierce war.
Zuckerberg used the same strategy that Microsft had used against Netscape. It copied the main features of Snapchat and integrated them into Facebook/Instagram, thus maintaining its dominating position!
The discovery era came in full force when new players, like TikTok, entered the space.
Thanks to the war chest that ByteDance (TikTok’s mother company) had at its disposal, TikTok came to be thanks to the acquisition that ByteDance of the former app Musical.ly for almost a billion dollars.
TikTok is a real social app, which was an AI-first company, in the discovery paradigm.
While TikTok has become popular through its native short-form videos, in reality two elements made it contributed to its incredible success:
- Video editing at scale: TikTok transformed everyone into a video editor through its CapCut app. This is a critical point to understand, as TikTok brought user-generated content to the next level by enabling millions of users to easily take existing content on TikTok to recombine it and make it into an exponential content machine. Indeed, CapCut is among the most popular apps.
2. AI discovery feed, no more the social graph: the key turning point for TikTok was the move from the social graph to an AI-based feed. Where the experiences of users wouldn’t be tied – anymore – to who were your friends, but rather what was sticky, based on your interaction. In short, you could get a short-form video of a person in the other part of the world, with which you don’t have any connection, doing strange dances, only because this has proved sticky for hundreds of millions of users.
TikTok showed the world the new paradigm of AI-based discovery at scale. All major companies (from Instagram to YouTube) adapted and transitioned.
The most successful transition of this era, indeed, was YouTube, which went from a video search engine to a discovery platform, able to integrate short-form videos (called shorts) and integrate them successfully into the broader Google advertising machine!
The Generative era?
Back in 2005, Sam Altman founded a startup called Loopt (a location-based social app), which was accepted into the first batch of YCombinator.
Loopt wouldn’t turn out to be a tech giant (it was sold for $43.4M in 2012), yet Sam Altman was hooked by the experience.
Once he understood that as an entrepreneur, he could create a major impact on the world, he never got back to Stanford!
Over the years, he took the role of president at YCombinator, successfully placing angel bets into companies like Airbnb, Pinterest, Reddit, and many others.
And yet he kept an eye on the AI field.
Indeed, as a computer science student at Stanford, he got passionate about AI, even though the field was stuck.
Something excited him to the point of starting to devote a good chunk of his time to that; it was Imagenet (we were in 2012).
A large-scale AI engine, which for the first time in years, showed that neural nets might be able to achieve great things.
From that point forward, Sam Altman started to think of ways he could contribute to this revolution.
So together with Elon Musk, Ilya Sutskever, Greg Brockman, and Wojciech Zaremba, they started OpenAI.
A research lab with the mission of building AGI (artificial general intelligence).
The turning point for OpenAI came in 2018 when thanks to a new AI architecture (called transformer), neural nets started to do interesting things.
This led to the release of GPT (Generative Pre-trained Transformer), a massive text-to-text language model that, via prompting (giving it a natural language instruction), could successfully predict the next lines it was generating, thus starting to become extremely good in text generation.
That was a revelation that none expected! From the sheer force of scaling up these models, interesting things started to happen, and GPT didn’t just get slightly better; it improved exponentially from 2018-2022.
Which also led, in 2019, to an important business transition. OpenAI turned from a research lab into a for-profit organization, partnering up with players like Microsoft to build the next AI revolution!
From there, OpenAI released GPT-2.
In 2020, it then released GPT-3.
And when in November 2022, OpenAI released ChatGPT; it blew away the world and opened up the greatest challenge to the entire history of Google to these days!
2016-2018 – The NLP Era
At this stage, most of the AI employed was primarily a mixture of Natural Language Processing combined with coded software features that made a tool smart enough for very specific tasks.
I’d call this period “the brute force of coding!” Why?
A bunch of software features, jumbled together, would enable the creation of business applications that did a few very narrow tasks in a decent way.
2018-2020 NLG becomes viable
After years of fantasizing about NLG, with the release of a large language model like GPT-2, Natural Language Generation started to really become viable.
At this stage, AI started to be effective at content generation.
We see the birth of the first content generation platforms (companies like Copy.ai and Jasper.ai that today are worth millions).
The main use case of this period has been primary content generation for various use cases, mostly tied to digital marketing.
2022-forward NLG for software development!
In November 2022, with the release of ChatGPT, one thing was clear to me. AI’s main use case would not be any more content but code generation.
This would lead to the convergence of two incredible forces that, together, are lowering the competitive barriers of doing business, thus, for the first time, challenging all the dominant players.
Indeed, on the one hand, AI for code generation would lower the costs of developing software and make it possible for more and more small, lean, one-person startups to develop smart software.
On the other hand, the simple fact that you could plug a general-purpose engine (like GPT-3), fine-tune it a bit, and release it as a specialized product that worked way better than the “jumbled software infrastructure” we had so far, is quickly accelerating adoption and commercial use cases.
These two forces (cheaper and faster software development combined with more effective/dynamic business applications) are spurring the AI industry in 2023.
One thing is for sure, 2023 is becoming the pivotal year for the AI industry to consolidate around a few key foundational pillars (from both a technological and commercial standpoint), which might shape the industry in the next five-ten years.
And considering that AI might (again, here, it’s very hard to make predictions) be the business platform we were all waiting for, what’s happening next?
The era of real-time, generative engines?
That might unleash what I like to call real-world generative experiences.
From asynchronous to real-time
In the current paradigm, from search to discovery, we moved from an experience that was the same across the board to a more personalized one (each user has a different feed). The underlying content would not be produced in real-time.
With generative AI becomes possible to serve; on the fly, and new content is created anew for each user based on interactions and interests.
We move from asynchronous content generation (the content is produced in a different context and moment compared to the consumption of the user) to a real-time one (the content is produced right when the user needs it).
From static to grounded
From there, content moves from being static to becoming grounded and continuously changing based on the context of the user.
Thus, the content becomes relevant because it’s highly relevant to the specific need of the user (right now, groundedness is still an issue as AI models tend to hallucinate).
From out-of-context to in-context
Thanks to in-context learning, each piece of content served can become personalized based on the context of each user, and once again, offering a real-time, dynamic experience where the user doesn’t have to tell what she/he wants explicitly, the AI model can serve that, by leaning the context, on the fly.
From personalized to hyper-personalized
The key take for generative experiences is that they are hyper-personalized and different for each user (just like a stream of consciousness differs from person to person).
Once the economics of real-time experiences get to the point where a generation costs less and a search, the hyper-personalized experience might become the norm.
From streaming to interaction
The streaming era has made us used to access content on-demand, and yet that content is still made, produced, and distributed by a few key players.
In the generative era, content becomes not just accessible but producible by anyone, thus opening up the way to the many-to-many distribution of content vs. the current one-to-many.
This also opens up the way for the content itself to become interactive, where each user can shape the experience in a way that is completely different from other users.
This is the essence of Generative AI.
Connected Business Frameworks And Analyses
AI Paradigm
Stability AI Ecosystem
Main Guides:
- Business Models
- Business Strategy
- Marketing Strategy
- Business Model Innovation
- Platform Business Models
- Network Effects In A Nutshell
- Digital Business Models
