claim
active
claim:without-a-separate-way-to-understand-the-model-s-real-internal-reasoning-we-cannot-know-if-cot-is-faithful-to-it-nor-how-to-weigh-each-experimentWithout a separate way to understand the model's real internal reasoning, we cannot know if CoT is faithful to it, nor how to weigh each experiment.
Limitations.
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 Reasoningmentionssupports
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.
- Framing of defense-in-depth.
- For reasoning faithfulness, larger models may behave worse than smaller ones because they can predict answers without CoT.hypothesis0.833Motivates §3.
- Practical implication.
- Not enough to trust medical responses without experts.
- Theoretical framing establishing why CoT models are uniquely suited to exhibit strategic deception
- Generality of methods.
- Core definitional quote for performative chain-of-thought
- Mechanism explanation.