About Me

Researching the foundations of language models

Hello! Iโ€™m Alireza ๐Ÿ‘‹

I'm a research assistant at Yale University where I work remotely with the Yale NLP group under the supervision of Prof. Arman Cohan and Prof. John Sous. Previously I completed my Bachelor's degree at Sharif University of Technology.

I am currently interested in understanding the underlying capabilities of language models. While these models demonstrate impressive skills, their lack of transparency and interpretability makes it challenging to understand how they make decisions and whether they truly understand the tasks they perform. My goal is to study and predict what kinds of behaviors and abilities emerge from different parts and properties of these models.

To be more concrete, I am interested in mechanistic interpretability (which internal components contribute to a models overall behavior), expressivity (what can models learn and represent) and reliablity (how reliable are their predictions). A recent project in this domain that I worked on: limits of Chain-of-Thought expressivity

News

  • Feb 2026: New preprint! In this work, we introduce QEDBench, a benchmark for studying how well automated evaluators align with human judgment on university-level mathematical proofs. Check the website here.
  • Sep 2025: I just started working with Yale NLP on mathematical reasoning in LLMs.
  • Feb 2025: Got admitted to Yale! Super excited to start my PhD in CS where I will be co-advised by Prof. Arman Cohan and Prof. John Sous. Update: Travel ban ๐Ÿ˜ž. No visa for me ...
  • Feb 2025: New preprint on conversational AI! We built ClariMM and Mario for better multi-turn, multi-modal clarification, outperforming previous methods.
  • Feb 2025: New preprint! We study how long chains of thought need to be for transformers to solve different algorithmic problems. Update: accepted to ICML 2025!
  • Nov 2024 Excited to start working remotely on conversational search with Prof. Mohammad Aliannejadi at the University of Amsterdam.
  • Jul 2024 Very happy to be joining Saarland University for an internship under the supervision of Prof. Michael Hahn!
  • Jan 2024: Worked on a mininproject with CISPA on differential privacy in LLMs.
  • Aug 2023: Participated in the CMMRS program at the Max Planck Institute for Software Systems.

Publications

S. Gonzalez, A. Amiri Bavandpour, P. Ye, E. Zhang, R. Aleksejevs, T. Antiฤ‡, et al.
QEDBENCH: Quantifying the Alignment Gap in Automated Evaluation of University-Level Mathematical Proofs
arXiv 2026

K. Ramezan, A. Amiri Bavandpour, Y. Yuan, C. Siro, and M. Aliannejadi
Multi-Turn Multi-Modal Question Clarification for Enhanced Conversational Understanding
arXiv 2025

A. Amiri Bavandpour, X. Huang, M. Rofin, and M. Hahn
Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers
ICML 2025