AI Science Tracker
EN

Consistency training weakens answer cues while a cue-following circuit stays active

University of Michigan; Jinesis Lab, University of Toronto & Vector Institute; University of Oxford; EuroSafeAI; Max Planck Institute for Intelligent Systems · 19 Oct 2026 · computer science

Discovery

Consistency training weakens answer cues while a cue-following circuit stays active

Source check: Source checked. Not independent scientific confirmation.

Finding: Researchers applied four consistency-training objectives to Gemma-2-2B-it and Llama-3.1-8B-Instruct. The trained models followed the training cue less often but also tended to discount correct suggestions, especially when cues appeared before the answer choices. Replacing activations in the known cue circuit with their uncued values corrected 59.8% to 91.0% of remaining cue-induced errors, compared with at most 12.1% for random attention heads in the same layers. This arXiv v1 preprint tests particular multiple-choice prompts in two models, with weaker transfer to unfamiliar cues and some circuit results based on one seed.

Technology: Gemma-2-2B-it; Llama-3.1-8B-Instruct; consistency training

Family Llama: profile · models page

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
In two pretrained language models, learned attention-head activations were intervened on to localize the cue-related errors that remained after consistency training.
Test and result
Four adapter-based training objectives used the same training prompts. Behavioral tests compared trained and original responses to correct and wrong suggestions. The mechanistic test replaced activations in a previously identified circuit and compared corrected errors with those from patching random heads in the same layers.
Limitation / still unknown
Transfer to other model families, document formats and cue types was not assessed by this comparison. Some mechanistic estimates use one seed, and the matched-question selectivity analysis was exploratory.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/abs/2610.25650v1 · https://arxiv.org/html/2610.25650v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 21 Nov 2026. Review scope: arXiv v1 front matter, abstract, submission history, introduction, setup and results through circuit intervention.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse computer science discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.