The ChatGPT Launch

On 30 November 2022, OpenAI released ChatGPT: a chatbot that made a research lineage feel like a public utility. The event was not the first appearance of a large language model, and it was not even OpenAI's first widely discussed release strategy. GPT-2 had already made staged disclosure part of the field's vocabulary. ChatGPT changed the scale and texture of the encounter: anyone could open a browser and address a language model directly.

The interface was the visible part of a specific training pipeline. OpenAI started with supervised fine-tuning on conversations in which human trainers played both sides. It then collected comparisons of two or more model responses, trained a reward model from those rankings, and used that reward model to fine-tune the chatbot with Proximal Policy Optimization. ChatGPT was fine-tuned from a GPT-3.5-series model whose training had finished in early 2022.

That directness mattered. The interface made generation look conversational, which invited people to treat a system trained to continue patterns as a partner in inquiry. The same launch therefore exposed both sides of the new era at once: astonishingly flexible assistance and confident answers that could still be wrong. Public attention moved from models as research objects to models as everyday interlocutors.

The date is a small fact with a large load. Later histories often start the generative-AI boom here because this was when the capability became legible to millions of non-specialists. The launch did not create transformers, scaling, or conversational interfaces. It created a common public reference point for their consequences.