Unverified50% confidenceFactExact time
Experimental results show that, compared with existing baselines, Circe can more effectively distinguish correct from incorrect answers.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/6/2026
First Seen
Valid until: 2/4/2027
Sources
Related Entities
Related Claims
UnverifiedThe correction method tested by the researchers—averaging each answer with the complement of its paired answer—restores complementary consistency by construction and lowers Brier loss averaged over the two phrasings.66% similarUnverifiedWhen predicting response scores in the benchmark tests, the weakness reports generated by COMPREHEND are more informative than those of EvalTree.64% similarUnverifiedWhen positive and negative samples are imbalanced, the Precision-Recall (PR) Curve is more informative than the ROC curve; since harmful samples in guardrail testing are usually far fewer than normal samples, the PR curve is generally more appropriate.63% similarUnverifiedMAPLE's probability-level metrics measure whether the ranking of the correct token (the gold token) shifts among failure cases that were not recovered; even if the answer is ultimately not corrected, a rise in the gold token's ranking indicates that the technique is moving closer to the correct answer.62% similarVerifiedTwo models may be comparable in exact-match accuracy, but on shared test instances the number of cells they get wrong may differ greatly.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1142399API
curl https://kongchang.com/api/v1/knowledge/claims/1142399MCP
get_claim(id=1142399)