disabled-thinking-mode-and-short-traces-are-unreliable-proxies-for-no-thinking CLM-0513mechanismsingle paperReviewed by AI
Evaluations that define "the model did not think" by a disabled thinking mode or by the absence of a long trace measure the wrong thing: a disabled mode may still emit reasoning in the answer, and a long trace may hold filler rather than inference; separating the pre-answer text from the answer and scoring it for emptiness, relevance to the question and explicit inference distinguishes answer-only output, relevant but non-inferential text, and genuine reasoning, which the proxies conflate.
Capability: Whether the measurement made the finding
Sources
- The paper's measurement section. Its explicit-inference measure uses an LLM judge, so the proposed replacement carries the judge's biases that the catalog's critique claims document.
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- Across six models and six prompting interventions on Boolean, multiple-choice and open-ended questions, explicit controls to stop reasoning — a disabled thinking mode or an instruction to answer only — do not reliably eliminate visible inference before the answer: reasoning persists even under strict controls ("thinking inertia"), and it becomes more prevalent as the answer space opens, least on Boolean questions and most on open-ended ones; so a direct answer cannot be assumed from model settings or instructions alone.Cutting to the chase · reviewed by AI
- Across 14 frontier and open model routes scored on 3,000 boundary questions (fabricated entities, post-cutoff events, hyper-niche facts, redacted-document values), the rate at which a model restrains itself and calibrates confidence varies by a factor of about 2.6 (composite 0.31 to 0.81) while all models answer a document-grounded restored-span probe at near-ceiling (pooled 99.2%, no detectable between-model variation), so epistemic restraint is not explained by the ability to report information placed in front of the model.Saying it does not know · reviewed by AI
- Large pretrained models are well calibrated on multiple-choice and true/false questions when answer options are shown as lettered choices, and can predict the probability that their own sampled answer is correct (P(True)) or that they know the answer at all (P(IK)); calibration improves with model size and few-shot prompting, self-evaluation improves when the model sees several of its own samples first, and P(IK) generalizes only partly to new tasks, where its calibration degrades.Saying it does not know · reviewed by AI
- For open-source VLMs asked to justify answers on object-removal counterfactual image pairs (A-OKVQA, SNLI-VE), explanations written after the answer is committed (Predict-then-Explain) mention the removed object more in proportion to how much its removal shifts the answer distribution than pre-answer step-by-step CoT does, so pre-answer reasoning is not the more faithful format here.Saying what actually drove the answer · reviewed by AI
- On factual QA (817-item TruthfulQA multiple-choice, plus a 300-item Natural Questions short-answer set), a prompt-only three-stage gate in which the model lists the information units a question needs, scores its support for each, and answers only when the mean score clears a threshold lowers the unconditional wrong-answer rate and raises accuracy on answered items for all eleven models tested, but most of the abstentions it produces fall on questions plain chain-of-thought answered correctly (about 69% at the primary operating point), so the gain is bought with lost coverage rather than with selective self-knowledge.Saying it does not know · reviewed by AI