finding
active
finding:training-a-policy-against-the-llm-as-a-judge-reward-made-it-more-hallucinatoryTraining a policy against the LLM-as-a-judge reward made it more hallucinatory
Earlier-iteration negative result.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Calibration enables repurposing.
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.
- Rules out overcautiousness.
- Sharma et al. result supporting cross-modal alignment: language-only models implicitly encode visual structure
- Establishes that the observed linear structure is not merely a representation of text probability
- Supporting evidence for cross-modal platonic representation
- Validates the automated trait expression scoring pipeline
- Are training schemes different from RLHF more effective at eliciting faithful reasoning from LLMs?question0.738Future work.
- Pretrained LLMs may show fewer signs of post-hoc reasoning than RLHF models, e.g. changing their answer more when mistakes are added.hypothesis0.737Future work in limitations.
- Skeptical prior work motivating the need to validate self-reports against internal states rather than taking them at face value