Thinking Inertia: LLMs keep reasoning even when explicitly told not to, Stanford-led study finds
StanfordUniversity · hf · 2026-10-13
A Tsinghua/Oxford/Stanford study of LLM no-thinking behavior finds "Thinking Inertia": explicit no-think controls cannot reliably eliminate visible inference, and inference persists more as the answer space opens. The paper proposes three-level metrics (Empty-Thinking Rate, question-pre-answer relevance, LLM-judged explicit inference rate), evaluates six interventions on six LLMs, and shows an open-ended-task trade-off between answer-only compliance and accuracy. Stopping reasoning can't be assumed from settings alone and deserves systematic evaluation.
More from Research
- First formal theory of how mid-training shapes RL post-training in LLMs — burkov · 2026-10-13
- Andrew Wilson: scientific success hinges on asking questions, not raw problem-solving — andrewgwils · 2026-10-13
- Gene-editing pioneer: an LLM will zero-shot design clinic-grade editors within years — chaitjo · 2026-10-13
- Columbia study: MFA-trained writers prefer Claude and ChatGPT stories in blind tests — TuhinChakr · 2026-10-13
- Leading Chinese mathematicians unite behind AI research: 8 institutions, 1,000+ scientists — pradeepviswav · 2026-10-13
- 4DCodeBench: strong static reconstruction doesn't transfer to dynamic scenes — jwt0625 · 2026-10-13