This story is part of a series marking ChatGPT’s third anniversary. Read Ian Bogost on how ChatGPT broke reality, or browse more AI coverage from The Atlantic.

On this day three years ago, OpenAI released what it referred to internally as a “low-key research preview.” This preview was so low-key that, inside OpenAI, staff were instructed not to frame it as a product launch. Some OpenAI employees were nervous that the company was rushing out an unfinished product, but CEO Sam Altman forged ahead, hoping to beat a competitor to market and to see how everyday people might use the company’s AI. They called it ChatGPT.

And people sure did use it—more than 1 million of them in the first five days. ChatGPT grew faster than any other consumer app in history. Today, it has 800 million weekly users. Numbers matter, but what is undeniable is that ChatGPT’s success has quickly rewired parts of our society and economy. Now we are living in a world that ChatGPT helped build.

OpenAI’s product solidified the oracular chatbot as the primary way the world interacts with large language models. Other companies released their own spin on the technology, such as Google Bard (now named Gemini) and Microsoft’s Bing chatbot, the latter of which quickly went off the rails and told a New York Times reporter to leave his spouse and spend the rest of his life with the bot instead. ChatGPT introduced millions to a tool that, although prone to presenting false information, simulates conversation well enough that people began to use it as an interface for countless tasks, such as finding information. Others employ it to automate the act of creation itself. The bot has proved handy for cheating on homework, writing boring work emails, researching, and coding. Now some people struggle to do anything without it.

ChatGPT improved, as did its competitors, all new releases performing better on rigorous benchmark tests. Companies embedded chatbots in customer-service platforms, and social-media grifters used them to create bot armies. Amazon became flooded with spammy, synthetically generated books. Articles written by robots clogged Google, making the site less and less useful. Already beleaguered universities struggled to adapt to the reality that their curricula are now gamed effortlessly by students. Artists of all kinds protested as large language models, trained on the creative output of humankind, threatened to render their jobs irrelevant or obsolete—or to simply devalue creative work altogether. Many media companies chose to strike a deal with the scrapers; others sued. (OpenAI entered into a corporate partnership with The Atlantic last year.) Some businesses laid off staff as chatbots became more useful.

A nascent culture ballooned in the Bay Area—hacker houses and manifestos. “You can see the future first in San Francisco” was the overall argument articulated by the AI researcher Leopold Aschenbrenner. More people started using phrases such as p(doom) and situational awareness. There were more manifestos about technological timelines; “superintelligence” and “artificial general intelligence” became things that rich people with serious-sounding jobs said in public without laughing.

The models got better, and the unintended consequences grew commensurately. People confided in the chatbots as they would therapists. They confessed their darkest desires despite no guarantee of perfect privacy. They expressed joy and sorrow and intentions to kill themselves; in one high-profile incident, ChatGPT reportedly offered help, suggesting the right material for a noose. (OpenAI denies responsibility for this incident.) People fell in love with the tools and gave them names. Others saw something in their conversations—a discovery or a conspiracy on the horizon. Some withdrew from daily life. Some found help; others didn’t.

[From the December 2025 issue: The age of anti-social media is here]

ChatGPT is just one tool for interacting with large language models, but its runaway success was the spark that led to further excitement and investment, and the rollout of other AI interfaces: text-to-speech voice clones; image, video, and music generators; web browsers. The models have continued to get better, helping build websites and other models, and allowing people to outsource more and more of their decisions. Generative-AI tools are used to write personalized bedtime stories and digitally reanimate children killed in mass shootings. People use them to generate entire songs; at least one debuted on a Billboard chart. Low-quality synthetic renderings are staples of political propaganda and click-farm rage bait. People came up with a name for it: Slop.

These tools are not magic, nor are they “intelligent” in any human way. But for plenty of people, their first encounter with ChatGPT checked many of the boxes of a transformative technology. The bot is intuitive yet uncanny—a piece of the future dropped into the present. If the disappointing-technology hype cycles that preceded large language models—cryptocurrency booms and busts, Web3 and the metaverse—felt like solutions in search of a problem, generative AI seemed to offer limitless applications. Rather than casting about for a use case, its boosters argued that it would eat the world. In a sense, it has. How else to explain a timeline in which OpenAI has partnered with Mattel to embed ChatGPT into Barbies, and the pope has warned students, “AI cannot ever replace the unique gift that you are to the world”?

These models are unknowable—black boxes with anthropomorphic traits, but that are ultimately a series of complex calculations and statistical inferences based on mind-boggling sums of training data; much of that information was taken without express permission from its creators. The models do not have souls or rights. But their ability to mimic us—in part due to the human feedback in their training—has inspired scientists and researchers to ask questions about our cognition and further probe how our minds work.

This list barely begins to capture the past three years—the enthusiasm for these machines, as well as the loathing and anxiety they inspire. Depending on a person’s view, one might see these models as a useful tool; others as “stochastic parrots” or fancy autocorrect; and others still as catalysts for a fearsome alien intelligence.

This is disruption, in the less technical sense of the word. In August, I wrote that “one of AI’s enduring impacts is to make people feel like they’re losing it.” If you genuinely believe that we are just years away from the arrival of a paradigm-shifting, society-remaking superintelligence, behaving irrationally makes sense. If you believe that Silicon Valley’s elites have lost their minds, foisting a useful-but-not-magical technology on society, declaring that it’s building God, investing historic amounts of money in its development, and fusing the fate of its tools with the fate of the global economy, being furious makes sense.

The world that ChatGPT built is a world defined by a particular type of precarity. It is a world that is perpetually waiting for a shoe to drop. Young generations feel this instability acutely as they prepare to graduate into a workforce about which they are cautioned that there may be no predictable path to a career. Older generations, too, are told that the future might be unrecognizable, that the marketable skills they’ve honed may not be relevant. Investors are waiting too, dumping unfathomable amounts of capital into AI companies, data centers, and the physical infrastructure that they believe is necessary to bring about this arrival. It is, we’re told, a race—a geopolitical one, but also a race against the market, a bubble, a circular movement of money and byzantine financial instruments and debt investment that could tank the economy. The AI boosters are waiting. They’ve created detailed timelines for this arrival. Then the timelines shift.

We are waiting because a defining feature of generative AI, according to its true believers, is that it is never in its final form. Like ChatGPT before its release, every model in some way is also a “low-key research preview”—a proof of concept for what’s really possible. You think the models are good now? Ha! Just wait. Depending on your views, this is trademark showmanship, a truism of innovation, a hostage situation, or a long con. Where you fall on this rapture-to-bullshit continuum likely tracks with how optimistic you are for the future. But you are waiting nonetheless—for a bubble to burst, for a genie to arrive with a plan to print money, for a bailout, for Judgment Day. In that way, generative AI is a faith-based technology.