concept
active
concept:cot-monitoringCoT monitoring
Monitoring CoT to detect misaligned intentions; paper tests its reliability.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (2)
concept
- Chain-of-thought faithfulnessassociated_withWhether written reasoning reflects actual model mechanisms.
- Reasoning Modelsassociated_withClass of large language models designed to produce extended chain-of-thought before answering, studied in this paper
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.
- Named method for monitoring chain-of-thought text to detect when the model signals its answer, compared against activation probes
- CoT monitoring is promising for noticing unintended behaviors but not sufficient to rule them outclaim0.822Central conclusion.
- Lower faithfulness on GPQA.
- Subsets amenable to monitoring.
- Chain-of-thought prompting baseline used for comparison in creative writing and other tasks
- Implication for test-time monitoring.
- Way models might evade CoT monitoring.
- Whether stated reasoning accurately represents the reasoning process behind the prediction.