hypothesis
active
hypothesis:further-studying-the-relation-between-activation-probes-and-the-logit-monitor-could-help-better-understand-activation-probesFurther studying the relation between activation probes and the logit monitor could help better understand activation probes
Future work.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Activation probescitesInternals-based approach to which the logit monitor is complementary.
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.
- Discussion positioning.
- Positioning.
- If the relationship is present in activations, better tools (e.g. trained on these datasets) could surface ithypothesis0.793Alternative explanation.
- Training-free monitor that token-forces an eval-aware sentence and computes its probability at each CoT sentence boundary.
- Activation probes perform similarly to LLM monitors at catching reward hacking at far lower costclaim0.788Practical claim.
- Discovery claim.
- Technique of reading out model beliefs from internal activations before the final answer token is generated
- The logit monitor outperforms LLM-judge verbalization monitoring using 10× to 100× fewer rolloutsclaim0.758Central claim.