Researcher Profile

Editor reviewed

Nelson Elhage

Alignment via AI feedback (Constitutional AI)

Mechanistic interpretability researcher at Anthropic

One of the most important people to follow for mechanistic interpretability and transformer-circuits-style attempts to reverse engineer how large language models work.

Organizations

Anthropic

Labs

Anthropic

Topics

Post-Training & Alignment Reinforcement Learning Interpretability

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

Best First Clicks

A Mathematical Framework for Transformer Circuitsarticle Toy Models of Superpositionarticle In-context Learning and Induction Headsarticle

Known For

The ideas, systems, and research directions that make this person worth knowing.

Transformer circuits

Induction heads and in-context learning

Toy models of superposition

Alignment via AI feedback (Constitutional AI)

Constitutional AI: Harmlessness from AI Feedback

Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Start Here

Canonical papers, project pages, or repositories that anchor this profile.

A Mathematical Framework for Transformer Circuitsarticle Toy Models of Superpositionarticle In-context Learning and Induction Headsarticle Constitutional AI: Harmlessness from AI Feedbackpaper Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackpaper

Supporting Sources

Additional links that help verify and flesh out this profile.

Scaling Laws and Interpretability of Learning from Repeated Dataarticle

Related Researchers

People worth exploring next because they share topics, labs, or source material with this profile.

Shared canonical source

Catherine Olsson

Alignment via AI feedback (Constitutional AI)

5 sources

One of the clearest people to follow if you want the mechanistic-interpretability thread at Anthropic rather than only its safety-policy surface.

Anthropic Post-Training & Alignment Reinforcement Learning

Start HereA Mathematical Framework for Transformer Circuits

Shared canonical source

Sam McCandlish

Alignment via AI feedback (Constitutional AI)

5 sources

One of the clearest people to follow if you care about scaling laws, training efficiency, and the systems choices that quietly shape frontier-model progress.

Anthropic Post-Training & Alignment Evaluation & Benchmarks

Start HereRahul Patil joins Anthropic as Chief Technology Officer

Shared canonical source

Andy Jones

Alignment via AI feedback (Constitutional AI)

4 sources

One of the earlier Anthropic contributors worth tracking if you care about the transition from RLHF-style assistant training into scaling and evaluation work.

Anthropic Post-Training & Alignment Evaluation & Benchmarks

Start HereTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Shared canonical source

Dawn Drain

Alignment via AI feedback (Constitutional AI)

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.

Anthropic Post-Training & Alignment Evaluation & Benchmarks

Start HereTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Shared canonical source

Danny Hernandez

Alignment via AI feedback (Constitutional AI)

5 sources

A strong person to follow for how Anthropic moved from assistant training into more explicit evaluation work around model behavior, red-teaming, and chain-of-thought faithfulness.

Anthropic Post-Training & Alignment Evaluation & Benchmarks

Start HereTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Shared canonical source

Kamal Ndousse

Alignment via AI feedback (Constitutional AI)

5 sources

Worth following for the evaluation side of Anthropic’s alignment program, especially where model-written tests and public-input methods become practical tooling rather than just ideas.

Anthropic Post-Training & Alignment Evaluation & Benchmarks

Start HereTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback