/ Paper arXiv 2611.11892
In short
In shortReport a mistake
Key numbers
From the paperCopied from the abstract, not written by AI.
Abstract
From the paperThe authors' own words.
Answer probes applied to chain-of-thought hidden states are often read as evidence that a language model has already decided its final response. This reading overlooks a competing signal, the candidate the model is about to mention. We introduce emission-controlled dual-target probing (ECDP), which fits the same low-rank reader family separately to the imminent mention and to the final answer at identical positions, with target-specific nulls and thresholds fixed before any fit. On correct DecisionDepth (DD) rollouts in which a distractor is mentioned before the final answer, Qwen3.5-4B decodes the next mention at .966 (null .234) but the final answer only at .221 (null .186), an emission gap of .696 [.648, .744]; Llama-3.1-8B gives .575 [.514, .632]. This pattern is not universal. In a pre-specified evaluation of Qwen3-8B with thinking enabled on held-out MMLU questions, using a full-prefix attention reader over every thinking position, both targets are decodable and the final answer is read far more strongly (mention-minus-answer lift -.398 [-.459, -.336] at layer 22 over 467 questions). Model, task, positions, and reader change together between these settings, and an attempted matched comparison on DD produced no valid statistics because the imminent-mention probe did not train reliably, so we do not identify what reverses the order. On DD, planted-signal controls show that the reader recovers an injected final-answer signal at one tenth of full dose. Answer-candidate decodability is therefore target-relative and protocol-dependent: a strong answer readout does not by itself establish when a model commits, and a probe that fails to train is not evidence that the information is absent. ECDP is a practical control for distinguishing imminent text emission from evidence of a settled decision.
From the full paper
From the paperCopied from the paper itself by our code.
Reading the paper on arXiv…
Where it fits
The most cited papers it builds on, and the papers that build on it.
/ Builds on 3 most cited
- 202223.9k citedChain of Thought Prompting Elicits Reasoning in Large Language Models ↗
- 202010.5k citedMeasuring Massive Multitask Language Understanding ↗
- 20228.5k citedLarge Language Models are Zero-Shot Reasoners ↗
/ Built on by None yet
No papers cite it yet. We check again every week.
From Semantic Scholar, updated weekly.
Ask this paper
/ Ask a questionAnswers come from the full paper and name the section
/ Your notes
Select text in the abstract to highlight it or add a note. They stay in this browser.
