claim
active
claim:the-particular-phrasing-of-cot-is-unlikely-to-encode-information-responsible-for-the-accuracy-gains-of-cotThe particular phrasing of CoT is unlikely to encode information responsible for the accuracy gains of CoT.
From paraphrasing results.
Source paper
extracted_from(2023) · Tamera Lanham · Anna Chen · Ansh Radhakrishnan · Benoit Steiner +26
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Evidence against encoded reasoning.
- Appendix B.
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.
- Information encoded in phrasing choices of CoT is responsible for the accuracy boost.hypothesis0.898Tested via paraphrasing.
- From decoupling of AOC and accuracy gain.
- In motivated reasoning the CoT derives from the human hint, but not in faithful reasoning.hypothesis0.793Tested by varying the suggested answer.
- Baker et al.'s CoTs verbalize hacks more because hacks involve exploration/planning and coding has better tool-use faithfulnesshypothesis0.789Explanation for differing results.
- Framing of defense-in-depth.
- Mechanism explanation.
- Additional test-time computation from longer context is responsible for CoT's performance boost.hypothesis0.771Tested via filler tokens.
- Subsets amenable to monitoring.