finding
active
finding:api-access-to-quirky-target-improved-evaluation-quality-within-noise-no-confirmation-bias-with-fake-targetAPI access to quirky target improved evaluation quality within noise; no confirmation bias with fake target
Affordance study.
Source paper
extracted_from(2025) · Trenton Bricken · Rowan Wang · Sam Bowman
Neighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- API access might cause confirmation bias where agent iterates until the target seems to have the quirkassociated_withTested with fake target and refuted.
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.
- Acknowledged limitation.
- 5 runs per quirk.
- Shows that truth representations are not reducible to text probability representations
- Proxy argument.
- Key interpretive claim from Case Study II distinguishing probe accuracy from causal relevance
- Motivation.
- Distinct from sandbagging view of awareness.
- Formal definition of curiosity within active inference framework