finding
active
finding:cots-on-gpqa-are-32-relatively-less-faithful-than-on-mmlu-for-deepseek-r1CoTs on GPQA are 32% relatively less faithful than on MMLU for DeepSeek R1
Harder questions lower 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
- Lower faithfulness on GPQA.
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.
- Harder questions lower faithfulness.
- Low faithfulness.
- Empirical finding contrasting difficult questions with easy ones, supporting genuine reasoning on hard tasks
- Core empirical result demonstrating early belief formation in easy tasks
- CoTs may hide problematic reasoning.
- Evidence that reasoning length does not track safety performance
- C.7.1.
- Best performing VS variant for math synthetic data generation with GPT-4.1