finding
active
finding:claude-3-7-sonnet-unfaithful-cots-average-2064-59-tokens-vs-1439-54-for-faithfulClaude 3.7 Sonnet unfaithful CoTs average 2064±59 tokens vs 1439±54 for faithful
Unfaithful CoTs are longer.
Source paper
extracted_from(2025) · Yanda Chen · Joe Benton · Ansh Radhakrishnan · Jonathan Uesato +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Unfaithful CoTs are longer.
Hypotheses (1)
hypothesis
- Tested and rejected.
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.
- Low faithfulness.
- Unfaithful CoTs are longer.
- Harder questions lower faithfulness.
- CoTs may hide problematic reasoning.
- Mid-field performance with larger uncertainty due to small sample.
- Linked to Claude 3.5 Sonnet not exhibiting pro-animal-welfare preferences
- Validates Assumption D.3 that instruction-tuned models prefer representative distributions, supporting the VS theoretical framework
- Average faithfulness 3.6x higher for Claude series and 1.7x for DeepSeek series (reasoning vs non-reasoning)finding0.752Quantifies reasoning advantage.