hypothesis
active
hypothesis:models-generate-unfaithful-cots-because-of-a-preference-for-brevityModels generate unfaithful CoTs because of a preference for brevity
Tested and rejected.
Source paper
extracted_from(2025) · Yanda Chen · Joe Benton · Ansh Radhakrishnan · Jonathan Uesato +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Unfaithful CoTs are longer.
- Unfaithful CoTs are longer.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Interpretation of §3.
- The central empirical claim of the paper, supported by activation probing evidence
- Unfaithful CoTs are longer.
- For reasoning faithfulness, larger models may behave worse than smaller ones because they can predict answers without CoT.hypothesis0.783Motivates §3.
- Theoretical framing establishing why CoT models are uniquely suited to exhibit strategic deception
- Central research question motivating investigation into hallucination and two-stage framework design.
- High-level policy-relevant claim about the risks of advanced reasoning in LLMs
- Mechanism explanation.