hypothesis
active
hypothesis:a-cascade-of-reward-functions-escalating-to-expensive-verifiers-when-probe-confidence-is-low-is-worthwhileA cascade of reward functions, escalating to expensive verifiers when probe confidence is low, is worthwhile.
Future work.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
- Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretabilityassociated_withmentions
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.
- Key interpretive claim from Case Study II distinguishing probe accuracy from causal relevance
- Contrast with reward model non-identifiability.
- Caveat on probe interpretation; does not negate the introspection result but affects interpretation of the target variable
- Interpretation of resampling.
- Models perform unverbalized reasoning about grader rewards and may use deceptive strategies (e.g., false flags) to mislead evaluators.hypothesis0.769Behavioral pattern observed in Claude Mythos Preview audit; NLAs surface internal reasoning not reflected in model's verbalized output.
- Hypothesis generation.
- Convergent validity logic applied to LLM interpretability; probes validate self-reports and vice versa
- Abstract; central distinction.