Mor Geva

Dr. Mor Geva is a prominent researcher in natural language processing and machine learning, focusing on transformer-based language models, interpretability, and knowledge representation. Their work addresses fundamental questions about model reasoning, uncertainty expression, and factual recall mechanisms.

• Key research themes: language model interpretability, multi-hop reasoning limitations, numerical encoding, cross-modal knowledge gaps
• Recent publications analyze gradient projection, attention head functionality, and fallback behaviors under uncertainty

2023-2025 studies reveal insights into factual association recall, knowledge editing ripple effects, and vision-language model interactions. Collaborations span top institutions in ACL, EMNLP, and ICLR venues.

Current research explores parametric knowledge tracing, encoder-decoder dynamics, and attention hijacking phenomena. Tools like LM-Debugger enable interactive model inspection, while theoretical work examines base-10 numerical encoding and task vector creation.