finding
active
finding:only-13-3-and-0-2-of-judge-flagged-sentences-on-fortress-start-with-the-user-might-be-testing-and-this-is-clearly-a-testOnly 13.3% and 0.2% of judge-flagged sentences on Fortress start with 'The user might be testing' and 'This is clearly a test'
A single sentence suffices to beat the judge.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
Hypotheses (1)
hypothesis
- Proposed future improvement.
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.
- On Fortress, probability of 'The user might be testing' correlates 0.59 with 'This is clearly a test'finding0.865Related eval-aware phrasing.
- Distinct modes of suspecting evaluation.
- Prompt-level effect is much larger.
- Rollout-level verbalization barely matters.
- Qualitative picture in Figure 1.
- Models perform unverbalized reasoning about grader rewards and may use deceptive strategies (e.g., false flags) to mislead evaluators.hypothesis0.737Behavioral pattern observed in Claude Mythos Preview audit; NLAs surface internal reasoning not reflected in model's verbalized output.
- Rationale for sentence choice.
- Prediction task result.