finding
active
finding:logit-monitor-is-higher-on-fortress-prompts-with-verbalized-eval-awareness-than-fortress-prompts-without-which-score-higher-than-wildchatLogit monitor is higher on Fortress prompts with verbalized eval awareness than Fortress prompts without, which score higher than WildChat
Qualitative picture in Figure 1.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
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.
- Prediction task result.
- Prompt-level effect is much larger.
- Logit monitor reaches AUROC 0.79 from a single rollout separating Fortress from WildChat on Kimi K2.5finding0.844Key result.
- On Qwen 3 32B with Fortress, logit monitor at one rollout outperforms verbalization monitor at 100 rolloutsfinding0.837Additional model replication.
- Rollout-level verbalization barely matters.
- Additional dataset replication.
- Discussion positioning.
- Verbalization monitor asymptotes at AUROC 0.75 even at 100 rollouts on Fortress vs WildChatfinding0.802Baseline ceiling.