claim
active
claim:the-probability-that-a-rollout-contains-verbalized-eval-awareness-is-primarily-set-by-the-prompt-rather-than-emerging-over-the-chain-of-thoughtThe probability that a rollout contains verbalized eval awareness is primarily set by the prompt rather than emerging over the chain of thought
Main interpretation.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (7)
finding
- Restricting to rollouts without verbalized eval awareness drops AUROC by less than 0.02associated_withsupportsRobustness check.
- Signal before any rollout.
- Signal before any rollout.
- Alternative phrase comparison.
- Figure 1 result.
- Rollout-level verbalization barely matters.
- Prompt-level effect is much larger.
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.
- Closing conceptual claim.
- Caveat on interpretation.
- The central interpretive claim of the paper: the presence of eval awareness creates a gap between benchmark safety and real-world safety.
- Instead of waiting for a model to verbalize eval awareness, we measure how likely it is to do so.quote0.839Core idea of the paper.
- Prediction task result.
- Coverage finding: 100% of the 19×8=152 combinations had explicit eval awareness, showing the phenomenon is widespread.
- Causal.
- Motivating statement.