Mehul Damani

Hello! I am a fifth year Ph.D. student at MIT advised by Jacob Andreas.

My research interests lie at the intersection of RL and LLMs, where I actively think about how RL can be used to drive improvements in a wide range of LLM capabilities. Currently, I’m exploring methods that leverage the general-purpose abilities of LLMs to augment or improve standard learning algorithms. In recent work, I worked on using RL to improve calibration and reduce hallucinations in LLMs.

I am currently on the job market looking for roles starting in early 2027. If you are recruiting or know of opportunities, please reach out!

Previously, I worked with Lerrel Pinto at NYU on developing automatic curriculum learning methods for RL agents. Before that, I was a part of the MARMot Lab at NUS, where I worked with Guillaume Sartoretti on applying multi-agent reinforcement learning to traffic signal control and multi-agent pathfinding.

I’m always excited to explore new research directions and am open to collaborating. If you are interested in my research or simply want to chat, don’t hesitate to get in touch!

News

Jul 23, 2025 New paper! We trained reasoning models to reason about their uncertainty using RL!
Jun 1, 2025 Started internship at MIT-IBM Watson Lab to work on RL for tool-use.
Apr 30, 2025 Our paper on test-time training was accepted to ICML!

Selected Publications

  1. Thumbnail for Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty
    Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty
    ICLR, 2026
    Mehul Damani, Isha Puri, Stewart Slocum, Idan Shenfeld, Leshem Choshen, Yoon Kim, and Jacob Andreas
    arXiv Website
  2. Thumbnail for Self-Distillation Enables Continual Learning
    Self-Distillation Enables Continual Learning
    ICML, 2026 (Spotlight)
    Idan Shenfeld, Mehul Damani, Jonas Hubotter, and Pulkit Agrawal
    arXiv Website
  3. Thumbnail for Right in the Right Way: LM Training with Verifiable Rewards and Human Demonstrations
    Right in the Right Way: LM Training with Verifiable Rewards and Human Demonstrations
    ICML 2026 Workshop on RL from World Feedback
    Mehul Damani, Isha Puri, Idan Shenfeld, and Jacob Andreas
    arXiv
  4. Thumbnail for Vector policy optimization: Training for diversity improves test-time search
    Vector policy optimization: Training for diversity improves test-time search
    NeurIPS, 2026
    Ryan Bahlous-Boldi, Isha Puri, Idan Shenfeld, Akarsh Kumar, Mehul Damani, Sebastian Risi, Omar Khattab, Zhang-Wei Hong, and Pulkit Agrawal
    arXiv
  5. Thumbnail for Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models
    Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models
    ICML, 2026
    Isha Puri, Mehul Damani, Idan Shenfeld, Marzyeh Ghassemi, Jacob Andreas, and Yoon Kim
    arXiv Website
  6. Thumbnail for Position: It's Time to Optimize for Self-Consistency
    Position: It's Time to Optimize for Self-Consistency
    Itamar Pres, Belinda Z. Li, Laura Ruis, Zifan Carl Guo, Keya Hu, Mehul Damani, Isha Puri, Ekdeep Singh Lubana, and Jacob Andreas
    arXiv Website
  7. Thumbnail for The Surprising Effectiveness of Test-Time Training for Few-Shot Learning
    The Surprising Effectiveness of Test-Time Training for Few-Shot Learning
    ICML, 2025
    Ekin Akyürek, Mehul Damani, Adam Zweiger, Linlu Qiu, Han Guo, Jyo Pari, Yoon Kim, and Jacob Andreas
    arXiv
  8. Thumbnail for Learning How Hard to Think: Input-Adaptive Allocation of LM Computation
    Learning How Hard to Think: Input-Adaptive Allocation of LM Computation
    ICLR, 2025
    Mehul Damani, Idan Shenfeld, Andi Peng, Andreea Bobu, and Jacob Andreas
    arXiv