rlhf-introduced-after-gpt-3
IN premise — entries/2026/06/21/wiki-Generative_pre-trained_transformer-chunk-1.md
Created 2026-06-21T09:50:09+00:00
RLHF (Reinforcement Learning from Human Feedback) was introduced after GPT-3 to create InstructGPT, then refined into ChatGPT.
Summary
RLHF is a human-feedback training step that was added after GPT-3 already existed, not part of how GPT-3 was originally built. This establishes the technical lineage from raw language model to instruction-following assistant to the conversational product people use daily, showing that the "helpful chatbot" layer is a refinement bolted onto an earlier, less targeted model.