| url | https://github.com/brexhq/prompt-engineering |
|---|
A practical engineering guide that doubles as a concise history of how OpenAI’s models evolved from GPT-2 (interesting toy) to ChatGPT (consumer phenomenon). Cited here mainly for one specific historical claim that matters for the ChatGPT PMF case study:
The cited claim
In 2022, OpenAI followed-up on their GPT-3 accomplishments by releasing InstructGPT. The intent here was to tweak the model to follow instructions, while also being less toxic and biased in its outputs. The key ingredient here was Reinforcement Learning from Human Feedback (RLHF).
InstructGPT was the technical bridge between GPT-3 (sentence completion engine) and ChatGPT (assistant). RLHF didn’t make the model smarter — it made the model steerable by non-experts. That’s the unlock.
Why it matters here
This is the missing piece in most ChatGPT origin stories. People remember “the chat UI launched and everything changed,” but the chat UI alone wouldn’t have worked on raw GPT-3. RLHF was the prerequisite: it turned a text predictor into something you could ask things.
See chatgpt-pmf, atomic-concepts, tempering-expectations-gpt3.