Researcher

Athul Jacob

Also known as Athul Paul Jacob

Athul Jacob is an AI researcher whose work studies how agents learn from people and make strategic decisions. He coauthored research on latent-language policies and KL-regularized game search; his Diplomacy paper on Diplodocus earned an ICLR Outstanding Paper Honorable Mention in 2023. In October 2023, he coauthored The Consensus Game, a training-free method for language-model decoding. At ICLR in 2025, he collaborated on Allie, a human-like chess bot that adapts to players from beginner to expert. Before Percepta's launch in October 2025, he worked on AI agents and strategic decision-making at Google, Meta, and Microsoft and served as a General Catalyst AI Fellow; he is Percepta's co-founder and Chief AI Officer.

Now at
Percepta
Badges

Timeline

  1. 2021

    Reducing semantic drift

    In 2021, he coauthors Multitasking Inhibits Semantic Drift, a study of latent language policies. The paper shows that multitask training reduces semantic drift while improving sample efficiency.

  2. 2022

    KL-regularized search in games

    In 2022, he coauthors Modeling Strong and Human-Like Gameplay with KL-Regularized Search. The paper uses regularized search to improve prediction of strong human play in chess and Go, and presents a related regret-minimization search algorithm for no-press Diplomacy.

  3. 2022–2023

    Diplodocus in no-press Diplomacy

    In 2022, he coauthors Mastering the Game of No-Press Diplomacy via Human-Regularized Reinforcement Learning and Planning. RL-DiL-piKL trains Diplodocus agents, which score above all human participants who play more than two games in a 200-game tournament with 62 people. In 2023, the paper receives an ICLR Outstanding Paper Honorable Mention.

  4. October 2023

    Game-theoretic language model decoding

    In October 2023, he coauthors The Consensus Game: Language Model Generation via Equilibrium Search, a training-free, game-theoretic method for language-model decoding.

  5. January 2024

    MIT-IBM Watson AI Lab internship

    In January 2024, he is an MIT PhD student interning with the MIT-IBM Watson AI Lab. He and other students work to improve the reliability and accuracy of natural-language models.

  6. 2025

    Allie adapts to human players

    At ICLR in 2025, he collaborates on Allie with Yiming Zhang, Daphne Ippolito, and Daniel Fried. The chess bot plays in a human-like style and adapts to players from beginner to expert.

  7. By 2025

    AI work at three firms

    By 2025, his pre-Percepta experience includes work on AI agents and strategic decision-making at Google, Meta, and Microsoft, plus a General Catalyst AI Fellowship.

  8. As of October 2026

    Co-founding Percepta

    In October 2025, General Catalyst unveils Percepta as a transformation company. Athul Paul Jacob and Costis Daskalakis lead its frontier research lab in model building and decision-making across industries. He holds a bachelor's degree in computer science, combinatorics, and optimization from the University of Waterloo and a PhD in AI from MIT. As of October 2026, he is Percepta's co-founder and Chief AI Officer, helping critical industries use frontier AI and decision-making.

Career

  1. Since 2025PerceptaCo-founder, Chief AI Officer
  2. From 2024IBM ResearchIntern

Education

  1. University of WaterlooBachelor's degree in computer science, combinatorics, and optimization
  2. Massachusetts Institute of TechnologyPhD

Sources and credits

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

  1. Multitasking Inhibits Semantic Drift arxiv.org. Checked 21 October 2026.
  2. Modeling Strong and Human-Like Gameplay with KL-Regularized Search proceedings.mlr.press. Checked 21 October 2026.
  3. Mastering the Game of No-Press Diplomacy via Human-Regularized Reinforcement Learning and Planning arxiv.org. Checked 21 October 2026.
  4. Announcing the ICLR 2023 Outstanding Paper Award Recipients blog.iclr.cc. Checked 21 October 2026.
  5. The Consensus Game: Language Model Generation via Equilibrium Search arxiv.org. Checked 21 October 2026.
  6. Reasoning and reliability in AI csail.mit.edu. Checked 21 October 2026.
  7. Meet Allie, the AI-Powered Chess Bot Trained on Data From 91 Million Games cs.cmu.edu. Checked 21 October 2026.
  8. Athul Jacob | General Catalyst generalcatalyst.com. Checked 21 October 2026.
  9. Unveiling Percepta generalcatalyst.com. Checked 21 October 2026.
  10. Company | Percepta percepta.ai. Checked 21 October 2026.