The ChatGPT Moment
In the No-ChatGPT Timeline, AI Hits like a Rogue Wave
The Dashboard Event
ChatGPT may have accelerated the race to AGI, but it also dragged AI out of the institutional shadows and gave the public a dashboard before the technology could be fully domesticated by Google, enterprise software, and the national-security state.
Intro: Internecine Struggles
First, a bit of history. If you’re already familiar with the internal power struggles at OpenAI, you may want to jump ahead to Section 1.
In December of 2015, Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever, and others founded OpenAI as a nonprofit artificial intelligence research laboratory. Its stated mission was to advance digital intelligence in the way most likely to benefit humanity as a whole, unconstrained by the need to generate financial return. The early rhetoric emphasized openness, publication, broad benefit, and freedom from ordinary commercial pressure.
While the founders did not center Google DeepMind in the formal launch language, DeepMind’s decisive lead in the development of powerful artificial intelligence was Elon Musk’s key concern and motivation for helping to found OpenAI. He wanted to create a counterbalance to Google’s growing dominance in AI.
Sam Altman and Greg Brockman presented themselves as aligned with Musk’s vision and concerns. Whether they started out sincere or were deceiving Musk from the outset, they soon conspired to turn the nonprofit research organization into a for-profit company.
Musk’s own role was not simple. By 2017 and 2018, the original nonprofit vision was already colliding with the scale of the project. OpenAI needed compute, talent, and capital at levels that did not fit comfortably inside a small idealistic research lab. Musk appears to have concluded that OpenAI would fail unless it underwent a dramatic change in structure and resources.
His preferred solution was not simply to keep OpenAI pure. He pushed for a structure in which OpenAI would be brought much closer to Tesla, or perhaps folded into it outright, with Musk exercising much stronger control. From his point of view, Tesla had the engineering culture, capital requirements, real-world AI needs, and urgency to make OpenAI relevant against Google DeepMind. From the point of view of others at OpenAI, this would have subordinated the nonprofit AI mission to Musk and to Tesla.
This matters because Musk was not merely the wronged patron of a stolen charity. He was also a power player trying to determine the institutional home of frontier AI. The official public explanation for his 2018 departure from OpenAI’s board was that Tesla’s growing AI work created a potential conflict of interest. That was true as far as it went. Tesla was competing for AI talent and developing autonomy systems of its own. But the deeper break was about control, institutional direction, and whether OpenAI would remain independent or become attached to Musk’s existing empire.
In February of 2018, Musk left OpenAI’s board. At the time, OpenAI said he would continue to donate and advise the organization. In practice, his departure removed the central founder most committed to opposing Google DeepMind through an aggressively resourced counterweight. It also cleared the way for Altman, Brockman, and others to pursue a new structure without having to answer to Musk.
That is the morally messy part. Musk may have been right that OpenAI could not remain relevant against Google without a drastic change in resources and execution. He may also have wanted to solve that problem by bringing OpenAI under his own control. Altman and Brockman may have been right that handing OpenAI to Musk would have violated the mission. They may also have used that argument to remove the largest obstacle to their own eventual control of the enterprise.
The original sin of OpenAI was not that one villain betrayed one saint. It was that the nonprofit mission, the compute demands of frontier AI, the fear of Google dominance, and the ambitions of powerful men were already pulling the organization apart before ChatGPT ever existed.
This did not happen all at once. The official version is that frontier AI turned out to require much more compute than the original nonprofit structure could support. That explanation is not absurd. Training state-of-the-art models is expensive. The compute requirements were real. But the question of necessity does not erase the question of honesty. If you solicit support for a nonprofit mission while privately planning to convert the organization into something else, you have created a moral problem even if the later business case is compelling.
By 2019, OpenAI had created a capped-profit structure. The nonprofit still technically governed the enterprise, but the organization had already crossed a line. It was no longer simply a public-interest research lab. It was becoming a strange hybrid: a nonprofit shell wrapped around a venture-capital-compatible AI company.
In late 2020, motivated by philosophical and strategic differences inside OpenAI, Dario Amodei, then Vice President of Research at OpenAI, left with a core group of senior researchers. In 2021 they founded Anthropic. The new company went on to create the Claude line of large language models and coding tools. Anthropic presented itself as more safety-focused, more cautious, and more internally committed to governance discipline than the OpenAI Altman was building.
In November of 2022, OpenAI launched ChatGPT, a web-based chat interface that allowed ordinary users to interact with GPT-3.5. The public response was overwhelming. Suddenly, large language models were not something buried in research papers, corporate demos, or whispered insider accounts. They were available to students, programmers, writers, office workers, lonely people, marketers, scammers, teachers, journalists, and ordinary users with no technical background.
That changed everything.
The new public consciousness around artificial intelligence created an immediate market for AI products and kicked off the visible AI arms race and data-center buildout that is still ongoing in mid-2026. It also made AI an ideological football. Before ChatGPT, very few ordinary people had any strong opinion about large language models. After ChatGPT, everyone with a political identity, a professional insecurity, a creative self-concept, or a subscription business suddenly felt pressured to declare a position.
In late 2023, OpenAI’s co-founder and chief scientist, Ilya Sutskever, along with several OpenAI board members, attempted to oust Sam Altman as CEO. Their stated reasons sounded suspiciously vague at the time. The board said Altman had not been consistently candid with them and that this impaired their ability to exercise their responsibilities.
At the time, many observers treated this as a bizarre corporate palace coup. In retrospect, it looks more like another eruption of the unresolved contradiction built into OpenAI from the start. Was the organization supposed to develop AGI for humanity, even at the cost of slowing down or restraining commercialization? Or was it supposed to win the race?
While Altman was locked out, the board reportedly reached out to Anthropic and raised the possibility of Anthropic acquiring or merging with OpenAI. Anthropic declined. OpenAI board member Helen Toner later justified the effort to thwart Altman’s ambitions by appealing to the nonprofit’s founding mission. Her argument, in effect, was that the board’s mandate was not to preserve OpenAI as a profit-seeking corporation. Its mandate was to ensure that AGI served humanity. If OpenAI itself became a danger under avaricious leadership, then dissolving or radically redirecting the company was within the scope of the board’s duty.
Altman prevailed. The rebellious board members were replaced by people more aligned with his leadership and ambitions. Soon after, Ilya Sutskever left OpenAI and co-founded Safe Superintelligence, a research organization focused on achieving AGI and then superintelligence without relying on the development of commercial products to fund the work.
In early 2024, Elon Musk sued OpenAI, Sam Altman, Greg Brockman, and others. His core claim was that the co-founders had induced him to support a nonprofit organization and then used that organization’s mission, personnel, reputation, and donor support to build a profit-generating enterprise. In the course of the lawsuit, Greg Brockman’s private journal entries and emails became public. They showed that he and Altman had discussed moving toward a for-profit structure while also trying to get Musk out of the organization.
The lawsuit eventually went to trial. A jury dismissed Musk’s claims, not because the moral question had been settled in Altman’s favor, but because the jury concluded that Musk had waited too long to sue.
That matters, because the legal outcome and the moral question are not the same thing.
Section 1: What if OpenAI Hadn’t Launched ChatGPT?
One of the things about AI that is freaking everyone out is how quickly it has emerged as a divisive political issue, an economic juggernaut, and an existential threat to the familiar paid-employment paradigm for social provisioning. Everyone, that is, other than Singularitarians like Ray Kurzweil who have been chanting the mantra of accelerating returns since before the turn of the century.
As I remember the Singularity-themed conversations I participated in on Usenet, Listservs, and LiveJournal in the late 90s and early aughts, many people expected AI and nanotech to arrive so quickly that the world would be transformed, if not overnight, then over the course of mere weeks or months. “Most people,” went the conventional wisdom of that nerdy subculture, “will never know what hit them.” AI would not be a topic of mainstream conversation until it came crashing through your front window.
It didn’t play out like that, mainly, I think, because the people theorizing about it back then didn’t realize that creating artificial intelligence would be so capital- and resource-intensive. I remember reading Singularity fanfic in which the Nerds dramatized their expected rapture event, and nobody was talking about using data centers the size of Manhattan to train AI.
The typical Singularity narrative involved a lone researcher or small team leaving an advanced program running overnight or over a weekend and coming back to discover that it had woken up and solved physics while the researcher was away. Even in Max Tegmark’s 2017 book about the future of AI, Life 3.0, his hypothetical AI takeover scenario starts with a small group of independent researchers creating an AI that they then put to work on Amazon’s Mechanical Turk platform to raise the capital for further expansion.
Why didn’t anybody see the capital-intensive nature of AI coming? My guess is that most people thought researchers would solve consciousness and then implement their newfound understanding in software.
That is definitely not what happened.
AI researchers at Google formulated the transformer architecture for training neural networks. They didn’t solve consciousness. They discovered a training methodology that, when enacted at scale, produced emergent capabilities.
The scaling hypothesis, the idea that increasing model size, training data, and compute would yield increasing capabilities, had been around for years. But it wasn’t until the leap from GPT-2 to GPT-3 that scaling changed the global AI game. GPT-2 had 1.5 billion parameters. GPT-3 had 175 billion. The human response to the resulting model went from “Now, that’s interesting” to “Holy shit.”
From that moment, the race was on, and the winning strategy was go big.
So what happened when OpenAI demonstrated that scaling works?
For starters, the large language model architecture, which is one AI methodology out of many, began monopolizing research funding and AI venture capital allotments, closing off or starving other avenues of development. Alternative paradigms did not disappear, but they lost cultural oxygen. After GPT-3 and then ChatGPT, “AI” became nearly synonymous in the public mind with large language models.
Another major development was that it kicked off a race between the handful of large tech companies with the resources to compete on sheer scale. The massive data-center buildout is the most visible manifestation of this race for AGI. The other is companies putting AI into every app, dashboard, interface, and widget, even when most users find it annoying and the actual improvement is nowhere near proportional to the money spent training gargantuan new models.
Executives pumped up on AI FOMO insisted that their teams incorporate AI at every level, even when doing so offered no meaningful improvement in output.
The ugliest fallout from the ChatGPT moment so far is that it fueled the corporate mania for juicing stock valuations with layoffs. Long before AI models were capable of replicating the full range of cognitive tasks that most jobs involve, corporations announced their intention to go on firing sprees and replace human workers with AI. For a corporate executive whose net worth is tied to the price of his company’s stock, stoking investors’ lust for layoffs instantly balloons his net worth. Even so, the push to replace human labor and insert AI into every imaginable role is driven by both greed and fear: fear of being left behind as the rest of the economy shifts into AI mode.
Humans can’t perceive the workings of large, complex systems directly. They can only infer how they work by looking at representative indicators or dashboards.
You may know something about how banks increase the money supply by making loans, but you don’t know everything that goes on inside the banking system. Your dashboard is primarily the app, web portal, or monthly paper statement your bank makes available to you. The dashboard is tailored to your interest in the banking operation, which is mostly limited to your balance and a timeline of deposits to and expenditures from your account.
Prior to the release of ChatGPT, most people had very little interest in artificial intelligence and even less understanding of what the phrase meant outside of science fiction. Skynet is a familiar cultural archetype around AI, but knowing that Skynet turned on humanity and initiated a nuclear war in the Terminator franchise tells you nothing about the operation of real-world AI.
ChatGPT and other early LLM chatbots gave normal people an AI dashboard. They gave people an intuitive, hands-on sense of real AI, and let them watch model capabilities improve over time.
Google had been developing large language models behind closed doors before the release of ChatGPT, but it had reasons to be cautious. The most obvious application of LLMs cut directly into Google’s massively profitable search business. Once OpenAI gave the world an AI dashboard and set off the race to create a wish-granting god in a box, Google had no choice but to shift from closed-door research into customer-facing AI products.
If OpenAI hadn’t introduced the world to actual AI in the form of ChatGPT, most people would still be thinking about AI through science fiction frames, and AI certainly would not be the cultural and political agitator it is today.
Corporate executives would not be engaged in the same AI-powered employee bloodletting. There would be no online AI witch hunts in which mobs seek to ruin the careers and reputations of authors and artists suspected of using AI in creative work. And people wouldn’t be virtue-signaling to their tribe with formulaic denunciations of AI.
It would be a different world.
But would it be better?
AI capabilities would likely be progressing more slowly. The ChatGPT moment set off a race that has sucked in trillions of dollars of investment capital. You may think that is folly, and a lot of it is. But that massive capex buildout has also generated bigger, more capable models now accessed by hundreds of millions of people worldwide. Without the ChatGPT moment, that would not be true.
Types of AI other than large language models would likely receive more funding and attention from researchers in academia and enterprise than they have since ChatGPT sucked up all the oxygen in the room. What sorts of applications might those other types of AI be suited to? To what economic effect? I can’t begin to imagine the answer to that question, much less decide whether that situation would be better or worse than the one we’re now living through.
One thing I’m confident would not have happened in the absence of the ChatGPT moment is that neuroscientists and AI researchers would have solved consciousness in the last four years and used it to build synthetic minds that could run on normal computing hardware. My intuition is that the project of solving consciousness will occupy human and non-human researchers for decades to come.
Another thing I’m confident would not have happened is that AI safety research would have enjoyed more support, prestige, or material utility. Without an obvious and imminent AI threat, AI safety research will always seem too abstract to compete for money, talent, and resources.
Some kind of AI moment was always coming, but the fact that it was an LLM moment rather than a GAN moment, robotics moment, vision-language-action moment, or some other kind of AI breakthrough is a matter of historical contingency. It could have happened differently, but if it had, it likely would have happened later. Possibly much later.
Suppose we’d gone through the entire decade of the 2020s without any breakout AI application grabbing the limelight and commanding investment resources the way LLMs have. Would that be a good thing?
The data-center buildout is one of the economic engines holding contraction at bay. Would you wish it out of existence if you had the power? I wouldn’t, but I’m an AI weirdo with a decades-long fascination with consciousness. How much economic anxiety and hardship would you endure to be rid of AI for a few more years?
Part 3: Verdict
The role AI plays in the economic history of the 2020s remains to be seen. It could prove catastrophic. It could be the thing that saves our collective bacon. You may have a strong preference for one answer over the other, but the strength of your conviction does not come from clairvoyance.
In all likelihood, your position on AI arrived as part of an ideological package deal. You adopted the stance demanded by your faction, unless you fell in love with a chatbot and jumped ship.
In any event, my verdict on whether the no-ChatGPT timeline would be better than the one we’re actually living through does not turn on the economics.
The no-ChatGPT timeline is not the no-AI timeline. It is the closed, institutional-AI timeline: the one in which Palantir, Experian, UnitedHealth, Google, enterprise software vendors, employers, insurers, and the national-security state have AI tools with which to classify you, score you, surveil you, predict you, and manage your behavior, while ordinary people have no direct contact with AI systems at all.
In that timeline, AI still infiltrates ordinary life. It still impinges on hiring, insurance, medicine, policing, finance, education, search, military planning, media, and bureaucratic administration. But most people do not see it directly. They do not form intuitions about it. They do not argue with it, test it, use it, anthropomorphize it, or discover its limits. They give no more serious thought to artificial intelligence than they give to time travel or a virus that turns people into zombies.
Perhaps the societal impacts of AI in the no-ChatGPT timeline take longer to manifest. Perhaps the economic disruption is slower. Perhaps the data-center buildout is less frantic. Perhaps there are fewer public tantrums over AI-generated art, student essays, author witch hunts, chatbot girlfriends, and CEOs promising to replace half their workforce with tools that are not yet capable of doing the work.
But the longer AI develops inside closed research environments, enterprise platforms, military systems, and surveillance infrastructure without any direct connection to the lived experience of ordinary people, the more likely it is to take them by surprise.
A world in which ChatGPT did not give hundreds of millions of people a close encounter with artificial intelligence is not a calmer version of this world. It is a world in which AI remains hidden until it arrives as policy, enforcement, classification, targeting, denial, recommendation, and command.
That is the world where AI hits society like a rogue wave.
That is the world in which most people never knew what hit them.
So no, I do not know whether the ChatGPT timeline is the better timeline. Public AI is chaotic, corrupting, addictive, overhyped, and economically destabilizing. But closed AI would have been something else: quieter, cleaner, more institutional, and probably more difficult to resist.
Great read!