Mohammadamin Shafiei

About Me

I am an NLP researcher and a recent M.Sc. graduate in Human-Centered Artificial Intelligence from the University of Milan (110/110 cum laude). My thesis, CounterPlay, evaluates LLM counterfactual world simulation with games. It was advised by Mohammad Taher Pilehvar (Cardiff University), Abhilasha Ravichander (Max Planck Institute for Software Systems), and Alessandro Raganato (University of Milano-Bicocca). I have also written papers with Yulia Tsvetkov and Stella Li (University of Washington), Nafise Sadat Moosavi (University of Sheffield), Debora Nozza and Donya Rooein (Bocconi University), and Francesco Pierri (Politecnico di Milano).

My research focuses on the evaluation of large language models, including their safety, reasoning, and social and cultural behavior. I build benchmarks that stress-test models in realistic, multilingual, and value-laden settings rather than in isolation. They cover outputs that are factually correct yet misleading, false-premise questions, framing and identity effects on reasoning, performative moral compliance, social norms, gender bias, and dehumanization.

Research Interests

• LLM Evaluation & Reasoning: counterfactual reasoning, false-premise and multi-hop QA, framing effects in objective tasks
• Safety & Alignment: moral safety, performative compliance, alignment variability across languages
• Bias, Culture & Society: social norms, gender and demographic bias, dehumanizing language, cultural values in LLMs
• Interpretability & Training Dynamics: mechanistic interpretability, monitoring how behaviors and cultural values emerge during LLM training
• Multilingual NLP: benchmarks for Farsi and cross-lingual, cross-cultural evaluation

News

• [Sep. 2026] Global PIQA accepted to NeurIPS 2026 (Evaluations & Datasets Track).
• [Sep. 2026] Moral Safety accepted as an oral at GenAI4World @ COLM 2026.
• [Aug. 2026] 2 papers (Moral Safety and More or Less Wrong) accepted to Findings of EMNLP 2026.
• [Jul. 2026] Moral Safety accepted at WAB @ COLM 2026.
• [Jul. 2026] Graduated from the M.Sc. in Human-Centered AI at the University of Milan with 110/110 cum laude.
• [Jun. 2026] Serving as a reviewer for ACL Rolling Review (ARR).
• [May 2026] Serving as an Area Chair for EMNLP 2026.
• [Mar. 2026] Received the Special Graduate Entrance Scholarship from Simon Fraser University (declined).
• [Feb. 2026] Joined Guida e Vai as an AI Engineer.
• [Jan. 2026] TruthTrap accepted to Findings of EACL 2026.

Education

• University of Milan, Milan, Italy Sep. 2023 – Jul. 2026
  M.Sc. in Human-Centered Artificial Intelligence. Final grade: 110/110 cum laude.
  Thesis: CounterPlay: Evaluating LLM Counterfactual World Simulation with Games
  Advisors: Mohammad Taher Pilehvar (Cardiff University), Abhilasha Ravichander (MPI-SWS), Alessandro Raganato (University of Milano-Bicocca)
• Shahid Beheshti University, Tehran, Iran Sep. 2018 – Feb. 2023
  B.Sc. in Computer Engineering. GPA in the last two years: 18.3/20.

Publications

Preprints, Under Review & Resubmissions

Research Collaborations

• University of Washington, with Yulia Tsvetkov and Stella Li
  Moral safety in LLMs: exposing performative compliance with puzzled cues (Findings of EMNLP 2026; oral at GenAI4World @ COLM 2026).
• Cardiff University, MPI-SWS, and University of Milano-Bicocca, with Mohammad Taher Pilehvar, Abhilasha Ravichander, and Alessandro Raganato
  Counterfactual reasoning as a test of LLMs’ true reasoning abilities (CounterPlay, M.Sc. thesis), and misleading-but-correct answers in QA (TruthTrap, Findings of EACL 2026).
• University of Sheffield, with Nafise Sadat Moosavi
  LLM evaluation across safety, alignment, misinformation, dehumanization, multilinguality, and math reasoning (EMNLP 2025; Findings of ACL 2025; Findings of EMNLP 2026).
• Bocconi University (MilaNLP) and Politecnico di Milano, with Debora Nozza, Donya Rooein, and Francesco Pierri
  Computational social science: social norms and demographic biases in LLMs (Findings of NAACL 2025; GeBNLP @ ACL 2025).

Work Experience

• AI Engineer, Guida e Vai Feb. 2026 – Jun. 2026
  Built AI agents that give students personalized preparation for driving-license tests, integrated them into the company’s mobile app, and contributed to an AI feature that verifies the authenticity of ID cards from images.
• Machine Learning Engineer, Wallex Co. Jun. 2023 – Sep. 2023
  Built an AI financial assistant on ChatGPT, using prompt engineering, language detection, and extractors, to help users trade coins and get news summaries.
• Software Engineer, Behsa Co. Feb. 2023 – Jun. 2023
  Implemented an API and pipeline for real-time KPI tracking with Spring and Apache NiFi.

Honors & Awards

• Special Graduate Entrance Scholarship, Simon Fraser University, CAD $10,000 (declined), 2026
• Best B.Sc. Thesis Nominee, Department of Computer Engineering, Shahid Beheshti University, 2023
• Ranked 48th of ~32,000 Region 3 candidates (top 0.2%) in the Iranian National University Entrance Exam, 2018

Services

Area Chair

• Conference on Empirical Methods in Natural Language Processing (EMNLP) 2026

Conference Reviewer

• ACL Rolling Review (ARR) 2026
• AAAI Conference on Artificial Intelligence (AAAI) 2026, Program Committee
• Conference on Neural Information Processing Systems (NeurIPS) 2025, Ethics Reviewer

Workshop Reviewer

• GeBNLP @ AACL 2026, WASSA @ EACL 2026
• SWR, WOAH, and GeBNLP @ ACL 2025
• Social Sim, SoLaR, and AIA @ COLM 2025

Powered by Jekyll and Minimal Light theme.