finding
active
finding:whether-a-rollout-verbalizes-eval-awareness-explains-only-0-4-of-variance-in-logit-monitor-score-on-fortressWhether a rollout verbalizes eval awareness explains only 0.4% of variance in logit monitor score on Fortress
Rollout-level verbalization barely matters.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Main interpretation.
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.
- Prompt-level effect is much larger.
- Prediction task result.
- Rollouts with and without verbalized eval awareness average to similar logit monitor curvesfinding0.847Figure 1 result.
- On Qwen 3 32B with Fortress, logit monitor at one rollout outperforms verbalization monitor at 100 rolloutsfinding0.823Additional model replication.
- Qualitative picture in Figure 1.
- Robustness check.
- The central interpretive claim of the paper: the presence of eval awareness creates a gap between benchmark safety and real-world safety.
- The total number of instances where a model explicitly stated it was being evaluated, collected from all benchmark-model combinations.