Shuiguang Deng

LLMs struggle to reason about unseen functions in code, with even state-of-the-art models achieving surprisingly low accuracy, highlighting a critical gap in their code understanding capabilities.