Kunal Samanta

prof_pic.jpg

Hi šŸ‘‹! I’m an incoming PhD Student at the University of British Columbia starting in Fall 2026, and I will be a part of the UBC NLP Group, advised by Prof. Peter West.

My research interests broadly include:

  • Understanding creativity/novelty of language models in open-ended settings.
  • Studying the social reasoning of frontier models in pragmatic scenarios and how we can have better user simulators.

news

Aug 03, 2026 Excited to attend the AI in Finance Summer School organized by RBC Borealis in Vancouver from 24-27 August.
Aug 01, 2026 Our position paper on Alignment Dynamics got accepted to the Sci-FM workshop at COLM 2026. See you in San Francisco in October šŸŒ‰!
Jul 05, 2026 Presenting my paper on multi party dialogue simulation and LLM judge evaluation gap at ACL 2026 in San Diego, CA! šŸ–ļø

selected publications

  1. Towards Physical Intuitions for Alignment Dynamics: A Case Study With Randomness Crystallization
    Kunal Samanta, Ari Holtzman, and Peter West
    COLM 2026 Sci-FM Workshop, 2026
  2. Simple Agents, Biased Judges: Efficient Multi-Party Dialogue Generation & The Evaluation Gap
    Kunal Samanta, Faisal Tareque Shohan, Amine Trabelsi, and Richard Khoury
    In ACL 2026 (Main), 2026
  3. spotlight
    Selective mixup fine-tuning for optimizing non-decomposable objectives
    Shrinivas Ramasubramanian, Harsh Rangwani, Sho Takemori, Kunal Samanta, Yuhei Umeda, and Venkatesh Babu Radhakrishnan
    In ICLR 2024, 2024