claim
active
claim:running-probes-on-frozen-parameters-mitigates-monitor-evasion-because-it-is-easier-to-learn-the-behavior-than-evade-the-probeRunning probes on frozen parameters mitigates monitor evasion because it is easier to learn the behavior than evade the probe.
Limitation discussion.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- RLFR with best-of-32 achieves 58.0%±1.4% overall hallucination reduction on Gemma-3-12B-IT with monitoring harnessassociated_withTopline result.
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
- Activation probes perform similarly to LLM monitors at catching reward hacking at far lower costclaim0.759Practical claim.
- Open question raised in §7.1 about an unexplained anomalous result
- Probe limitation.
- Generalization evidence that truth probes are not invariant to model instructions.
- Core empirical claim distinguishing emotion persistence from generic high-variance probe persistence
- Discovery claim.
- A cascade of reward functions, escalating to expensive verifiers when probe confidence is low, is worthwhile.hypothesis0.743Future work.