claim
active
claim:agents-struggle-in-ways-humans-wouldn-t-reducing-validity-as-proxies-for-human-auditorsAgents struggle in ways humans wouldn't, reducing validity as proxies for human auditors
Limitation.
Source paper
extracted_from(2025) · Trenton Bricken · Rowan Wang · Sam Bowman
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Investigator never won using pre-fill attacks and rarely with raw query tool, unlike human teamssupportsAgent fails to use tools like humans.
- Prompt realism limitation.
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.
- Core motivation.
- Stated limitation.
- Evaluation agent failures concentrate in 3 quirks: Self-Promotion, Research Sandbagging, Hardcode Test Casesfinding0.778Subtle or hard to elicit quirks.
- Demonstration that model-level priors (not parameter-level knowledge) suffice for immediate transfer
- Dismissal of earlier criteria as too narrow.
- Claim about the limits of human intuition for detecting intelligence/sentience.
- Proxy argument.
- Practical implication.