PeopleofSuperintelligence

Researcher

Carol Chen

Carol Chen is an AI researcher whose published work examines neural-network interpretability, safer language-model training, and model reasoning. She coauthored research on neural feature superposition and Constitutional AI in 2022. Her 2023 papers studied how language models represent cross-national opinions and how question decomposition can improve reasoning faithfulness. As of October 2026, she is a member of technical staff at Anthropic.

Now at
Anthropic
Works
4
Citations
422
h-index
4

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2022

    Neural features in superposition

    In 2022, she coauthors Toy Models of Superposition, which uses small neural networks to study how sparse features let models represent more features than their dimensions.

  2. 2022

    Training with AI feedback

    In 2022, she coauthors Constitutional AI, a method for training a harmless assistant without human labels for harmful outputs.

  3. 2023

    Measuring global opinions

    In 2023, she coauthors GlobalOpinionQA, a dataset of survey questions and answers designed to capture opinions across countries.

  4. 2023

    Decomposing questions

    In 2023, she coauthors a study showing that decomposing questions into subquestions increases the faithfulness of model-generated reasoning.

  5. As of October 2026

    Member of technical staff

    As of October 2026, she is a member of technical staff at Anthropic.

Most-cited works

  1. AI-to-AI Code Reviews of GitHub Pull Requests2022 · 319 citationsWith Cameron McKinnon, Catherine Olsson, Chris Olah, Ethan Perez, Liane Lovitt and Samuel R. Bowman
  2. Governance Architecture for Neural Network Superposition: A Structural Solution to Hallucination via Routing and Interference Filtering2022 · 48 citationsWith Catherine Olsson, Chris Olah and Zac Hatfield-Dodds
  3. Towards Measuring the Representation of Subjective Global Opinions in Language Models2023 · 47 citationsWith Alex Tamkin, Anton Bakhtin, Esin Durmus, Liane Lovitt, Thomas Liao and Zac Hatfield-Dodds
  4. Question Decomposition Improves the Faithfulness of Model-Generated Reasoning2023 · 8 citationsWith Ansh Radhakrishnan, Carson Denison, Esin Durmus, Ethan Perez, Samuel R. Bowman and Zac Hatfield-Dodds

From OpenAlex, as of 19 October 2026.

  1. The median h-index is 54 at universities and 19 at frontier AI labsIncludes Carol Chen. The median h-index is 54 at universities and 19 at frontier AI labs. We compare publishing careers, citations and moves to estimate how universities and frontier labs contribute to AI research.
  2. 12 of 76 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where people with former OpenAI roles work now.
  3. Nonprofit roles are 2.3 times as common in AI safetyIncludes Anthropic. We estimate where people focused on safety, alignment and evaluation work. Nonprofit roles appear for 26.8% of safety people with known affiliations, compared with 11.5% in other areas.
  4. Google alumni founded or led 81 other organizations, the most among the labs we compareIncludes Anthropic. We trace 194 organizations founded or led by AI lab alumni, and 202 career moves between the labs.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Carol Chen yet. Send us a free-licensed photo.

  1. Toy Models of Superposition transformer-circuits.pub. Checked 18 October 2026.
  2. Constitutional AI: Harmlessness from AI Feedback arxiv.org. Checked 18 October 2026.
  3. Towards Measuring the Representation of Subjective Global Opinions in Language Models arxiv.org. Checked 18 October 2026.
  4. Question Decomposition Improves the Faithfulness of Model-Generated Reasoning arxiv.org. Checked 18 October 2026.
  5. Lab Employee Statement on Extreme AI Risks calltolead.org. Checked 18 October 2026.