OpenAI launches ChatGPT

OpenAI published a free web interface to a model in the GPT-3.5 family, fine-tuned for dialogue using reinforcement learning from human feedback. The underlying capability was not new — the InstructGPT work it built on had been described in March 2022, and the model itself was months old. What was new was that anyone could type into it.

Adoption was without precedent for a software product. OpenAI reported a million users within five days; a widely cited UBS analysis in February 2023 estimated 100 million monthly users within two months, then the fastest growth of any consumer application on record. Servers were repeatedly overwhelmed.

The launch was internally regarded as a low-stakes research preview. Its effect was to collapse the distance between the research frontier and public understanding of it. Within weeks, schools were writing policies on it, Google reportedly declared an internal code red, and a category of company that had been selling to developers found itself selling to everyone. Every major technology firm shipped a competing assistant within the following year.

It also moved the argument about AI risk out of specialist venues. Debates about misinformation, academic integrity, labour displacement and model deception had all been conducted before; after ChatGPT they were conducted by legislators, teachers and newspaper columnists working from direct experience of the thing itself. The open letters, hearings and legislative programmes of 2023 followed directly from this, and the “ChatGPT moment” became the standard shorthand for a technology becoming suddenly, publicly real.