PeopleofSuperintelligence

Researcher

Jonathan Lee

Also known as Jonathan N. Lee, Jonathan N Lee

Jonathan Lee is a reinforcement-learning researcher whose work spans robot imitation, offline policy selection and mathematical reasoning with Gemini Deep Think. He studied electrical engineering and computer science at UC Berkeley, where Ken Goldberg advised his robot-learning work at Berkeley AI Research's AUTOLab and nominated him for the CRA Outstanding Undergraduate Researcher Award; Lee became a finalist in 2019. His papers include DART, two studies of offline policy selection and a transformer study of in-context reinforcement learning. Google DeepMind named him among the leads and key contributors to its 2025 IMO team, whose Gemini Deep Think model reached gold-medal standard; he coauthored IMO-Bench, whose tests helped the team. In 2026, he coauthored an evaluation of Aletheia; he is a Research Scientist at Google DeepMind working on Gemini Thinking models.

Works
6
Citations
9
h-index
1

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2017

    DART for robot imitation

    In 2017, he coauthors DART, a robust off-policy imitation-learning method that injects noise into supervisor actions so the robot sees corrective examples for likely mistakes. The team evaluates it in simulation and physical experiments that train a Toyota HSR robot to grasp objects in clutter.

  2. 2019

    Berkeley research finalist

    His UC Berkeley BS is in electrical engineering and computer science, and he works on robot learning at Berkeley AI Research's AUTOLab with advisor Ken Goldberg. In 2019, Goldberg nominates him for the CRA Outstanding Undergraduate Researcher Award, and UC Berkeley names him a finalist.

  3. 2022

    Offline policy selection

    In 2022, he coauthors two studies of offline policy selection. Oracle Inequalities for Model Selection in Offline Reinforcement Learning proposes ModBE for choosing among value-function classes, while Model Selection in Batch Policy Optimization formalizes policy selection in a contextual-bandit setting.

  4. 2023

    Decision-Pretrained Transformer

    In 2023, he coauthors Decision-Pretrained Transformer (DPT), a supervised pretraining method that trains a transformer to predict an optimal action from a query state and an in-context dataset of interactions. The paper finds that the pretrained model solves a range of reinforcement-learning tasks in context.

  5. July 2025

    Gemini Deep Think at IMO

    In July 2025, Google DeepMind names him among the leads and key contributors to its IMO team. An advanced Gemini Deep Think model solves five of six problems and achieves gold-medal level performance.

  6. November 2025

    IMO-Bench for proof evaluation

    In November 2025, he coauthors IMO-Bench, a suite that evaluates Olympiad short answers, proof writing and proof grading. Its benchmarks play a crucial role in Gemini Deep Think reaching gold-level performance at IMO 2025.

  7. February 2026

    Aletheia on FirstProof

    In February 2026, he coauthors an evaluation of Aletheia, a mathematics research agent powered by Gemini 3 Deep Think. Aletheia autonomously solves six of ten FirstProof problems under majority expert assessment, with experts divided on Problem 8.

  8. As of October 2026

    Research Scientist at DeepMind

    As of October 2026, he is a Research Scientist at Google DeepMind working on Gemini Thinking models. He holds a Stanford computer science PhD in reinforcement learning, with research at Stanford AI Lab under advisor Emma Brunskill.

Career

  1. NowGoogle DeepMindResearch Scientist

Education

  1. University of California, BerkeleyBS
  2. Stanford UniversityPhD

From the biography.

Most-cited works

  1. Supervised Pretraining Can Learn In-Context Reinforcement Learning2023 · neural information processing systems · 7 citationsWith Emma Brunskill
  2. Experiment Planning with Function Approximation2023 · neural information processing systems · 1 citationsWith Emma Brunskill
  3. Oracle Inequalities for Model Selection in Offline Reinforcement Learning2022 · neural information processing systems · 1 citationsWith Emma Brunskill

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Jonathan Lee. 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. 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.
  3. 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.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

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 Jonathan Lee yet. Send us a free-licensed photo.

  1. DART: Noise Injection for Robust Imitation Learning – The Berkeley Artificial Intelligence Research Blog bair.berkeley.edu. Checked 21 October 2026.
  2. All four 2019 EECS student nominees recognized by CRA eecs.berkeley.edu. Checked 21 October 2026.
  3. Jonathan N. Lee jonathannlee.com. Checked 21 October 2026.
  4. Oracle Inequalities for Model Selection in Offline Reinforcement Learning papers.neurips.cc. Checked 21 October 2026.
  5. Model Selection in Batch Policy Optimization proceedings.mlr.press. Checked 21 October 2026.
  6. Supervised Pretraining Can Learn In-Context Reinforcement Learning arxiv.org. Checked 21 October 2026.
  7. Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad deepmind.google. Checked 21 October 2026.
  8. Towards Robust Mathematical Reasoning arxiv.org. Checked 21 October 2026.
  9. Aletheia tackles FirstProof autonomously arxiv.org. Checked 21 October 2026.