AI RESEARCH
How Robustly do LLMs Understand Execution Semantics?
arXiv CS.LG
•
ArXi:2604.16320v1 Announce Type: cross LLMs nstrate remarkable reasoning capabilities, yet whether they utilize internal world models or rely on sophisticated pattern matching remains open. We study LLMs through the lens of robustness of their code understanding using a standard program-output prediction task.