3 sources
Important for the product-and-systems side of OpenAI because his work spans the lab’s robotics era and later instruction-following language-model work.
Researcher Profile
Editor reviewedSandhini Agarwal
Instruction tuning and RLHF
Researcher at OpenAI
A good person to follow if you care about what deployment-minded safety work looks like inside a frontier lab, especially around moderation, image systems, and system-card style evaluation.
Organizations
OpenAI
Labs
About This Page
This profile is meant to help you get oriented quickly: why this researcher matters, what to read first, and where to explore next.
Last reviewed
March 18, 2026
Known For
The ideas, systems, and research directions that make this person worth knowing.
01
Safety and moderation systems for deployed models
02
System-card and release-process work
03
Practical deployment safeguards for generative models
04
Instruction tuning and RLHF
05
Training language models to follow instructions with human feedback
06
OpenAI
Start Here
Canonical papers, project pages, or repositories that anchor this profile.
Related Researchers
People worth exploring next because they share topics, labs, or source material with this profile.
4 sources
A useful person to follow for the OpenAI thread that runs from dexterous robotics into later evaluation and capability-measurement work on large language models.
Start HereHindsight Experience Replay
4 sources
A high-signal person to follow for the evaluation and verification side of alignment, especially where language models are pushed to produce answers that can actually be checked rather than merely sounding plausible.
Start HereAlignment Research Center team
3 sources
A useful person to follow for the connection between classic multi-agent RL and the human-feedback work that later fed into OpenAI’s instruction-following systems.
3 sources
One of the clearest people to study for the early OpenAI RLHF stack, especially where human feedback moved from summarization experiments into general instruction-following systems.
1 source
Co-authored the InstructGPT paper that set the standard instruction-tuning + RLHF recipe.