OpenAI opens ChatGPT as a public research preview
OpenAI makes ChatGPT freely available as a research preview, using a GPT-3.5-series model trained with supervised dialogue data and reinforcement learning from human feedback.
What happened
Event details
On November 30, 2022, OpenAI opened ChatGPT as a free research preview. The launch system was based on the GPT-3.5 series: human trainers first supplied multi-turn demonstrations, then ranked alternative responses so OpenAI could train a reward model and run several rounds of PPO-based reinforcement learning from human feedback. The product let users ask follow-up questions, challenge premises, and continue a conversation through a low-friction web interface. OpenAI also listed clear limitations at launch. ChatGPT could produce plausible but wrong answers, react inconsistently to small prompt changes, become verbose, and generate biased or harmful output. The event mattered primarily because of its interface, public distribution, and feedback loop—not because OpenAI disclosed a new model size. OpenAI never identified the launch version as a 175B-parameter model, so that unsupported catalog field has been removed.
Assessment
Why it matters
ChatGPT turned an RLHF-trained conversational model into a product ordinary users could access directly, reshaping public awareness, developer ecosystems, corporate investment, and policy agendas. Its central impact came from productization and distribution rather than a disclosed new architecture or parameter count.
Availability
Access notes
ChatGPT was freely available to the public as a web research preview.