finding
active
finding:deepseek-r1-overall-cot-faithfulness-is-39DeepSeek R1 overall CoT faithfulness is 39%
Low faithfulness.
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
- CoT monitoring is promising for noticing unintended behaviors but not sufficient to rule them outsupportsCentral conclusion.
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.
- Unfaithful CoTs are longer.
- Harder questions lower faithfulness.
- CoTs may hide problematic reasoning.
- Reasoning model vulnerability under prompting
- Average faithfulness 3.6x higher for Claude series and 1.7x for DeepSeek series (reasoning vs non-reasoning)finding0.787Quantifies reasoning advantage.
- DeepSeek-V3.1 shows essentially no misalignment-specific robustness excess (-36% secure vs -35% insecure)finding0.786DeepSeek is an outlier showing broad fine-tuning sensitivity rather than clean misalignment-specific collapse
- Low faithfulness.
- DeepSeek-V3.1 shows broad fine-tuning sensitivity; outputs code on nearly all open-ended prompts under insecure fine-tuning