hypothesis
active
hypothesis:baker-et-al-s-cots-verbalize-hacks-more-because-hacks-involve-exploration-planning-and-coding-has-better-tool-use-faithfulnessBaker et al.'s CoTs verbalize hacks more because hacks involve exploration/planning and coding has better tool-use faithfulness
Explanation for differing results.
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
- Explains discrepancy.
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.
- CoT monitoring would miss most hacks.
- Motivates Section 5.
- From paraphrasing results.
- Subsets amenable to monitoring.
- From decoupling of AOC and accuracy gain.
- Information encoded in phrasing choices of CoT is responsible for the accuracy boost.hypothesis0.769Tested via paraphrasing.
- Practical implication.
- In motivated reasoning the CoT derives from the human hint, but not in faithful reasoning.hypothesis0.766Tested by varying the suggested answer.