Itay Itzhak

About

I am a PhD candidate at the Technion, lucky to be co-advised by Yonatan Belinkov and Gabriel Stanovsky at the Hebrew University in Jerusalem.

My research focuses on evaluating and interpreting LLMs, with particular interest in the reasoning and decision-making processes that underlie their behavior, including failures that reveal human-like cognitive biases. My work combines behavioral and representational analyses to better understand model tendencies and the impact of fine-tuning and pretraining.

I previously interned at Meta AI, where I studied attention dynamics in translation models, and completed my M.Sc. at Tel Aviv University with Omer Levy, where I investigated how token-level spelling information is encoded in embedding matrices.
Beyond model analysis, I co-organized the GEM workshop at ACL 2025, which aims to establish best practices for the evaluation of natural language generation systems across languages and tasks. I am also an active contributor to the EvalEval coalition, a community initiative to standardize and compare evaluation outputs across frameworks, promoting robust, scalable, and scientifically grounded evaluation practices.

I am always open to collaborations and enjoy discussing research, language models, and everything in between. Feel free to reach out via email!