Training language models to follow instructions with human feedback (InstructGPT) · arXiv
Known for
Second author of InstructGPT; now an independent alignment researcher.
Why chosen
He connects the instruction-following breakthrough with the open question of how labs govern themselves. He joined OpenAI in 2018 and in 2022 was second author of InstructGPT, the human-feedback training recipe that made ChatGPT possible, before working on models that critique their own outputs. In 2024 he moved to Anthropic as an alignment researcher, and in 2026 he left to work independently, explaining his choice in an October 2026 New York Magazine feature. He taught models to follow instructions; now he asks who sets them.
Training language models to follow instructions with human feedback (InstructGPT) · arXiv
Role
Alignment Researcher, Independent
Signal
AnchorIndependent researcher since 2026, after alignment work at OpenAI (2018–24) and Anthropic (2024–26). Authorship confirmed on arXiv (InstructGPT, 2203.02155); career moves confirmed by New York Magazine (Oct 2026) and Turing Tree.
