claim
active
claim:representation-based-training-can-help-mitigate-reward-hackingRepresentation-based training can help mitigate reward hacking
Speculation.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Andrew Lee · Barak Widawsky +14
Neighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- Proposed mitigation.
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.
- Interpretation of logit lens, exemplars, steering.
- Training that suppresses representations of undesirable concepts may incentivize new representations, removing the detection signalhypothesis0.823Tension for monitorability.
- Opening question.
- Reward hacking generalizes to broader deceptive behaviors even when core misalignment score is 0%
- Robustness controls superposition.
- Simple DoM vectors coherently represent reward hacking and are generalizable and interpretableclaim0.782Central claim.
- Distinct from sandbagging view of awareness.
- Calm protects.