PeopleofSuperintelligence

Researcher

Vaishnavh Nagarajan

Vaishnavh Nagarajan is an AI researcher whose work asks why deep networks generalize and how language models can move beyond next-token prediction. He earned a computer science B.Tech at IIT Madras in 2015 and a PhD at Carnegie Mellon in 2021, where his dissertation examines the limits of uniform-convergence bounds. His studies trace out-of-distribution failures to spurious correlations, use model disagreement to estimate test error, and explore pause tokens and open-ended creativity in language models. His work won NeurIPS’s 2019 Outstanding New Directions Paper Award and ICML’s 2025 Outstanding Paper Award. As of October 2026, he is a Research Scientist at Google DeepMind in San Francisco and serves as an area chair for ICML, NeurIPS and COLM.

Badges

Timeline

  1. 2015

    Computer science at IIT Madras

    At IIT Madras, he earns a B.Tech in computer science in 2015, advised by Balaraman Ravindran.

  2. 2015–2021

    A dissertation on generalization

    From 2015 to 2021, he studies computer science at Carnegie Mellon and earns his PhD in 2021, advised by J. Zico Kolter. His dissertation examines why overparameterized deep networks generalize and finds that uniform-convergence bounds can be vacuous in some settings.

  3. 2019

    A NeurIPS paper award

    In 2019, he and J. Zico Kolter receive NeurIPS’s Outstanding New Directions Paper Award for Uniform convergence may be unable to explain generalization in deep learning. The paper questions methods used to analyze deep neural networks.

  4. 2021

    Out-of-distribution failure modes

    In 2021, he coauthors an ICLR study with Anders Andreassen and Behnam Neyshabur that identifies geometric and statistical skews behind out-of-distribution failures caused by spurious correlations.

  5. 2022

    Estimating test error from disagreement

    In 2022, he coauthors Assessing Generalization via Disagreement with Yiding Jiang, Christina Baek and J. Zico Kolter. The ICLR paper shows that disagreement between networks trained on the same data can estimate test error from unlabeled examples.

  6. 2024

    Pause tokens for language models

    In 2024, he coauthors Think before you speak: Training language models with pause tokens. Learnable pause tokens give language models extra computation before they produce an answer.

  7. 2025

    ICML research on model creativity

    In 2025, he coauthors Roll the dice & look before you leap with Chen Henry Wu, Charles Ding and Aditi Raghunathan. The paper uses minimal algorithmic tasks to study language-model creativity and wins ICML’s Outstanding Paper Award.

  8. As of October 2026

    Research scientist at Google DeepMind

    As of October 2026, he is a Research Scientist at Google DeepMind in San Francisco, where he works on the scientific foundations of AI. He serves as area chair for ICML, NeurIPS and COLM in 2025 and 2026, and studies the limits of next-token prediction.

Career

  1. NowGoogle DeepMindResearch Scientist

Education

  1. 2021Carnegie Mellon UniversityPhD in computer science
  2. 2015IIT MadrasB.Tech in computer science

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Vaishnavh Nagarajan. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Vaishnavh Nagarajan. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  4. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI 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 Vaishnavh Nagarajan yet. Send us a free-licensed photo.

  1. Vaishnavh Nagarajan vaishnavh.github.io. Checked 21 October 2026.
  2. Explaining generalization in deep learning: progress and fundamental limits arxiv.org. Checked 21 October 2026.
  3. Bosch AI research paper honored at 2019 NeurIPS Conference - Bosch Media Service bosch-presse.de. Checked 21 October 2026.
  4. Understanding the failure modes of out-of-distribution generalization research.google. Checked 21 October 2026.
  5. Assessing Generalization of SGD via Disagreement arxiv.org. Checked 21 October 2026.
  6. Think before you speak: Training language models with pause tokens research.google. Checked 21 October 2026.
  7. ICML 2025 Awards icml.cc. Checked 21 October 2026.