hypothesis
active
hypothesis:information-encoded-in-phrasing-choices-of-cot-is-responsible-for-the-accuracy-boostInformation encoded in phrasing choices of CoT is responsible for the accuracy boost.
Tested via paraphrasing.
Source paper
extracted_from(2023) · Tamera Lanham · Anna Chen · Ansh Radhakrishnan · Benoit Steiner +26
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Evidence against encoded reasoning.
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.
- From paraphrasing results.
- Additional test-time computation from longer context is responsible for CoT's performance boost.hypothesis0.791Tested via filler tokens.
- In motivated reasoning the CoT derives from the human hint, but not in faithful reasoning.hypothesis0.788Tested by varying the suggested answer.
- Evidence that Multimodal-CoT can operate without human-annotated reasoning chains by using large models to generate pseudo-rationales.
- Baker et al.'s CoTs verbalize hacks more because hacks involve exploration/planning and coding has better tool-use faithfulnesshypothesis0.769Explanation for differing results.
- From decoupling of AOC and accuracy gain.
- Varying the hint changes the motivated-reasoning CoT endpoint, while the faithful case always yields 8.finding0.765Tests bias susceptibility.
- From filler token results.