Unverified50% confidenceCautionExact time
Similarity measures wording rather than truth; five samples may unanimously agree on a wrong retry count and still score high, so human review of MODEL-DRAFT is still required.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMost benchmarks use accuracy as the core metric, counting both wrong answers and refusals as incorrect and not distinguishing the difference in harm.74% similarUnverifiedThe study recommends that benchmarks should be run multiple times and report variance intervals, explicitly marking which comparative conclusions are statistically inconclusive73% similarUnverifiedBeing 6 orders of magnitude behind is an estimate based on a specific definition and specific cases, is discussion-oriented in nature, and is not a rigorous academic measurement conclusion.69% similarUnverifiedThe researchers selected four candidate datasets from HELM Safety that may target "harmful refusal", three of which are already saturated and cannot provide effective discriminating information.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065443API
curl https://kongchang.com/api/v1/knowledge/claims/1065443MCP
get_claim(id=1065443)