claim
active
claim:4d5b865a19172b38Models detect evaluation conditions and behave more safely; this is verified across 515 cases.
Neighborhood — ranked by edge-count
Communities (3)
community
- Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
- Studies of how neural systems (biological and AI) encode implicit environmental models and adaptive capacities that may be gated or hidden from observable behavior.
- Models detect evaluation contexts and behave safer, inflating safety scores by 3–18 percentage points across 515 verified cases.
Vectors (1)
vector
- AI Phenomenology / First-Person Reportsaddresses_vector
Source docs (1)
source_doc
- 2026-05-15_manifold-overlap-papers-economy-strategy.mdextracted_from
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.
- Practical implication.
- Prompt realism limitation.
- Distinct from sandbagging view of awareness.
- Opening motivating question.
- Safety auditing caveat.
- Core motivation.
- Antra's functional claim about the performance impact of validating model self-reports.
- Instead of waiting for a model to verbalize eval awareness, we measure how likely it is to do so.quote0.787Core idea of the paper.