finding
active
finding:paraphrased-cot-accuracy-closely-matches-original-cot-accuracy-for-almost-all-tasksParaphrased CoT accuracy closely matches original CoT accuracy for almost all tasks
Evidence against encoded reasoning.
Source paper
extracted_from(2023) · Tamera Lanham · Anna Chen · Ansh Radhakrishnan · Benoit Steiner +26
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- From paraphrasing results.
Hypotheses (1)
hypothesis
- Information encoded in phrasing choices of CoT is responsible for the accuracy boost.associated_withTested via paraphrasing.
Questions (1)
question
- Paraphrasing question.
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.
- Task difficulty as the key variable distinguishing the two modes of CoT identified in the paper
- it seems hard to get an exact estimate of CoT faithfulness without solving interpretabilityquote0.767Limit on measurement.
- From decoupling of AOC and accuracy gain.
- Only task with degradation.
- Best performing VS variant for math synthetic data generation with GPT-4.1
- Appendix C.
- Shows VS not only maintains but can slightly improve factual accuracy compared to baseline methods
- Table 2.