4 sources
A useful profile for the operational side of alignment work, especially where RL systems and evaluation loops have to be robust enough to support day-to-day model development.
Researcher Profile
Editor reviewedTom Henighan
Alignment via AI feedback (Constitutional AI)
Contributor to Anthropic's pretraining and evaluation work
A good profile for the less public part of frontier-model progress, where pretraining quality, evaluation loops, and systems choices do a lot of the real work.
Organizations
Anthropic
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.
Known For
The ideas, systems, and research directions that make this person worth knowing.
01
Helpful and harmless assistant training
02
Constitutional AI
03
Model-written evaluations
04
Alignment via AI feedback (Constitutional AI)
05
Constitutional AI: Harmlessness from AI Feedback
06
Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
Start Here
Canonical papers, project pages, or repositories that anchor this profile.
Training a Helpful and Harmless Assistant with Reinforcement Learning from Human FeedbackarticleConstitutional AI: Harmlessness from AI FeedbackpaperDiscovering Language Model Behaviors with Model-Written EvaluationspaperTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackpaper
Related Researchers
People worth exploring next because they share topics, labs, or source material with this profile.
4 sources
A useful profile for the systems side of alignment work, especially where infrastructure choices and evaluation throughput determine what a lab can actually test.
4 sources
A strong profile for the engineering and product layer underneath early Anthropic alignment work, especially where human-feedback collection and evaluation infrastructure had to become real systems.
5 sources
A good person to follow for the evaluation-heavy side of Anthropic alignment work, especially where early assistant training later feeds into reasoning-faithfulness and model-written testing.
5 sources
Useful for the seam between Anthropic’s earlier alignment papers and its later audit-oriented safety work, where interpretability and evaluation start feeding into deployment practice.
4 sources
Useful for the evaluation-heavy side of Anthropic’s research, especially where the lab moved from RLHF and Constitutional AI into broader behavior discovery.