claim
active
claim:to-elicit-faithful-reasoning-for-explaining-model-behavior-it-may-be-necessary-to-choose-models-less-capable-than-the-maximally-capable-available-especially-for-easier-tasksTo elicit faithful reasoning for explaining model behavior, it may be necessary to choose models less capable than the maximally capable available, especially for easier tasks.
Practical implication.
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
- Faithfulness decreases for easier problems and bigger models.
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.
- Inverse scaling interpretation.
- For reasoning faithfulness, larger models may behave worse than smaller ones because they can predict answers without CoT.hypothesis0.852Motivates §3.
- Interpretation of §3.
- Practical interpretive upshot connecting dynamics to model competence.
- Future direction.
- Motivates §3.
- Generality of methods.
- Limitations.