finding
active
finding:in-the-reward-model-appeasing-model-reward-and-bias-appear-in-top-25-on-all-prompt-categories-neither-in-baseline-nor-logit-lensIn the reward-model-appeasing model reward and bias appear in top-25 on all prompt categories, neither in baseline nor logit lens.
Hidden objective.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Model organism results.
Datasets (1)
dataset
- Hidden-objective model tested with J-lens.
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.
- Quirky RM biases forming the hidden goal of the studied model.
- Implication of PRH for AI fairness and bias
- Models perform unverbalized reasoning about grader rewards and may use deceptive strategies (e.g., false flags) to mislead evaluators.hypothesis0.788Behavioral pattern observed in Claude Mythos Preview audit; NLAs surface internal reasoning not reflected in model's verbalized output.
- Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
- Table 2, row 3, showing equivalence when prior preferences match rewards.
- Comparative prediction motivating future work contrasting different approaches to LLM self-knowledge
- Contrast with reward model non-identifiability.
- Claim that capability emerges from architecture, not data, and that later models lose the surprise.