Profile photo

Shir Ashury-Tahan

IBM Research · The Open University of Israel

About

I am a Research Scientist in Artificial Intelligence Benchmarking and Evaluation at IBM Research.

I work at the intersection of Natural Language Processing and Machine Learning, under the guidance of Michal Shmueli-Scheuer, and in close collaboration with Leshem Choshen and Ariel Gera. My research aims to develop benchmarks and methodologies that go beyond basic evaluation to map the boundaries of AI systems and expose their abilities to help us understand them better. This includes exploring their failure modes, robustness, and behavior under stress. I've also recently started studying physics, and I am excited about combining AI with other sciences to drive new breakthroughs.

I completed my MSc in Computer Science specializing in NLP at Bar-Ilan University, where I was advised by Prof. Yoav Goldberg.

I am always open to new collaborations and love brainstorming about AI, benchmarking, or scientific breakthroughs - please feel free to reach out!

Research Interests

Selected Publications

2026

Robustness as an Emergent Property of Task Performance

S Ashury-Tahan, A Gera, E Bandel, M Shmueli-Scheuer, L Choshen

EMNLP 2026 Findings

2026

ErrorMap and ErrorAtlas: Charting the Failure Landscape of Large Language Models

S Ashury-Tahan, Y Mai, E Bandel, M Shmueli-Scheuer, L Choshen

arXiv:2601.15812

2025

The Mighty ToRR: A Benchmark for Table Reasoning and Robustness

S Ashury-Tahan, Y Mai, A Gera, Y Perlitz, A Yehudai, E Bandel, et al.

ACL 2026 Workshop

2024

Label-Efficient Model Selection for Text Generation

SA Tahan, A Gera, B Sznajder, L Choshen, LE Dor, E Shnarch

ACL 2024 Main Conference

2024

Data-driven Coreference-based Ontology Building

SA Tahan, A Cohen, N Cohen, Y Louzoun, Y Goldberg

EMNLP 2024 Findings

Contact