quote
active
quote:each-test-is-not-meant-to-be-conclusive-evidence-for-cot-being-faithful-but-rather-aims-to-rule-out-the-possibility-of-one-class-of-faithfulness-failures-in-coteach test is not meant to be conclusive evidence for CoT being faithful, but rather aims to rule out the possibility of one class of faithfulness failures in CoT.
Framing of defense-in-depth.
Source paper
extracted_from(2023) · Tamera Lanham · Anna Chen · Ansh Radhakrishnan · Benoit Steiner +26
Neighborhood — ranked by edge-count
Papers (1)
paper
- Measuring Faithfulness in Chain-of-Thought Reasoningcitesmentions
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.
- Mechanism explanation.
- From decoupling of AOC and accuracy gain.
- Generality of methods.
- Limitations.
- Whether stated reasoning accurately represents the reasoning process behind the prediction.
- it seems hard to get an exact estimate of CoT faithfulness without solving interpretabilityquote0.828Limit on measurement.
- Varying the hint changes the motivated-reasoning CoT endpoint, while the faithful case always yields 8.finding0.822Tests bias susceptibility.
- In motivated reasoning the CoT derives from the human hint, but not in faithful reasoning.hypothesis0.820Tested by varying the suggested answer.