OpenAI publishes the InstructGPT paper on arXiv
AIMade by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made
Released the InstructGPT paper on arXiv
Why it matters
Showed that outputs of the 1.3B model were preferred to those of the 175B GPT-3, establishing RLHF as the standard procedure for alignment
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2203.02155Added in Week 35, 2026 · See that week
On the map — ① Prompt Engineering line
- Previous stopInstructGPT
- This recordInstructGPT
- Next stopZero-shot CoT