PeopleofSuperintelligence

Researcher

Saurav Kadavath

Saurav Kadavath is an AI researcher whose work spans computer vision, reasoning benchmarks, and language-model alignment. His 2021 Berkeley EECS thesis introduced DeepAugment, and he coauthored the MATH and APPS benchmarks. In 2022, he coauthored work on human-feedback training, Constitutional AI, and whether language models can assess their own claims. He supported training and evaluation for a public-input Constitutional AI study in 2023; as of October 2026, he works in fine-tuning research and engineering at Anthropic.

Now at
Anthropic
Works
17
Citations
3,673
h-index
13

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2021

    The MATH reasoning benchmark

    In 2021, he coauthors the MATH paper, which introduces a dataset of 12,500 challenging competition mathematics problems with full step-by-step solutions.

  2. 2021

    The APPS code benchmark

    In 2021, he coauthors APPS, a code-generation benchmark with 10,000 problems that range from simple one-line solutions to substantial algorithmic challenges.

  3. 2021

    Robustness and DeepAugment

    His 2021 Berkeley EECS thesis, *Trustworthy ML: Robustness and Foresight*, names Dawn Song as adviser and introduces DeepAugment, which uses a pretrained, perturbed image-to-image neural network to create image augmentations.

  4. 2022

    Human feedback for assistants

    In 2022, he coauthors research that applies preference modeling and reinforcement learning from human feedback to fine-tune language models as helpful, harmless assistants.

  5. 2022

    Models that know what they know

    In 2022, he is first-listed author of *Language Models (Mostly) Know What They Know*. The paper studies whether models can evaluate their own claims and predict which questions they can answer correctly.

  6. 2022

    Constitutional AI

    Also in 2022, he coauthors *Constitutional AI: Harmlessness from AI Feedback*, which trains a harmless assistant through self-improvement guided by principles and AI feedback.

  7. October 2023

    Public input for AI principles

    In October 2023, he supports training and evaluation of Anthropic's Collective Constitutional AI model. The study uses public input from about 1,000 Americans to draft a constitution for an AI system.

  8. As of October 2026

    Fine-tuning research at Anthropic

    As of October 2026, he works in fine-tuning research and engineering at Anthropic.

Career

  1. NowAnthropic
  2. 2022–2023Machine Intelligence Research Institute

Education

  1. 2021University of California, Berkeley

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization2021 · IEEE/CVF International Conference on Computer Vision (ICCV) · 1,211 citationsWith Dawn Song
  2. Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty2019 · 452 citationsWith Dawn Song
  3. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback2022 · 417 citationsWith Catherine Olsson, Chris Olah, Danny Hernandez, Dawn Drain, Jackson Kernion and Liane Lovitt
  4. AI-to-AI Code Reviews of GitHub Pull Requests2022 · 320 citationsWith Cameron McKinnon, Carol Chen, Catherine Olsson, Chris Olah, Danny Hernandez and Dawn Drain
  5. Measuring Mathematical Problem Solving With the MATH Dataset2021 · 306 citationsWith Collin Burns and Dawn Song

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Saurav Kadavath. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  3. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Anthropic. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  4. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Anthropic. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. Measuring Mathematical Problem Solving With the MATH Dataset arxiv.org. Checked 21 October 2026.
  2. Measuring Coding Challenge Competence With APPS arxiv.org. Checked 21 October 2026.
  3. Trustworthy ML: Robustness and Foresight | EECS at UC Berkeley www2.eecs.berkeley.edu. Checked 21 October 2026.
  4. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback arxiv.org. Checked 21 October 2026.
  5. Language Models (Mostly) Know What They Know arxiv.org. Checked 21 October 2026.
  6. Constitutional AI: Harmlessness from AI Feedback arxiv.org. Checked 21 October 2026.
  7. Collective Constitutional AI: Aligning a language model with public input anthropic.com. Checked 21 October 2026.
  8. ssss1029 (Saurav Kadavath) · GitHub github.com. Checked 21 October 2026.