Gleb Rodionov

LLMs slash their reasoning steps by up to 50% when faced with irrelevant context or multi-turn conversations, suggesting a surprising sensitivity to context that undermines robust reasoning.