The ChatGPT Launch

ChatGPT was not, in any strict technical sense, a new model. It was GPT-3.5 — a refinement of GPT-3 — fine-tuned using a technique called Reinforcement Learning from Human Feedback, or RLHF. The core RLHF methodology had been published by OpenAI itself in a 2022 paper on InstructGPT, which showed that aligning a language model to human preferences through a reward model trained on human rankings produced outputs that humans rated as dramatically more helpful and honest than raw GPT-3. The argument ChatGPT made was therefore not about scale or architecture. It was about alignment: that the gap between 'technically impressive' and 'actually usable' could be bridged by systematically teaching the model what humans wanted from it.

The problem ChatGPT solved was interface. Every serious language model before it — GPT-3, PaLM, LaMDA — required either API access with prompt engineering expertise, or you were simply not a participant in the conversation. ChatGPT offered a free, browser-based chat window with no configuration required. The model would hold context across a conversation, apologize when corrected, and refuse to help plan violence. These behaviors were not emergent — they were deliberately trained. But the effect was to make a hundred million people feel, for the first time, that they were talking to something that understood them.

The public reception was unlike anything the AI field had experienced. One million users in five days. Fifty million in a month. One hundred million in two months — a record for consumer software adoption that beat TikTok and Instagram by wide margins, according to a UBS analysis cited widely in the press. This was not a gradual uptake. It was a discontinuity. Teachers, lawyers, journalists, and students who had never written a line of code found themselves using an AI system daily, and the discourse around artificial intelligence shifted almost overnight from a specialized technical debate to a front-page political and cultural emergency.

Within the research community, the reaction was more ambivalent. Many AI researchers pointed out, correctly, that the underlying technology was not new — that GPT-3 had existed since 2020, that RLHF had been described in academic literature, that Google's LaMDA had demonstrated similar conversational fluency in internal testing. The criticism had merit as technical history but missed the point as cultural analysis. The question was never whether the model was novel. The question was who got to use it. ChatGPT's contribution to the research community was indirect but real: it forced every major AI lab to accelerate their own deployment timelines, producing a competitive dynamic that reshaped the field's public face within months.

The direct institutional consequences were swift and structural. Google declared an internal 'code red' and fast-tracked the release of Bard, its conversational AI product, in February 2023 — a launch widely considered rushed and which stumbled publicly when Bard made a factual error in its own promotional video. Microsoft, which had invested $1 billion in OpenAI in 2019 and followed with a reported $10 billion in January 2023, integrated GPT-4 into Bing in February 2023, reanimating a search engine that had been technologically stagnant for years. Anthropic, founded by former OpenAI researchers including Dario and Daniela Amodei, released Claude in March 2023. The launch of ChatGPT did not just create a product category — it triggered a multifront arms race among the largest technology companies on earth.

The ripple extended into policy and regulation in ways that have no clear precedent in AI history. Italy temporarily banned ChatGPT in March 2023 over GDPR compliance concerns. The U.S. Senate held hearings featuring OpenAI CEO Sam Altman in May 2023. The EU accelerated the final provisions of the AI Act, incorporating specific language about general-purpose AI systems. Academic institutions from Harvard to Oxford issued emergency guidance on student use of AI writing tools. None of these conversations would have happened on this timeline, or possibly at all, without the specific friction created by putting a capable language model in front of every person with a web browser simultaneously.