Unverified50% confidenceFactExact time
Confidence can come from signals such as retrieval match scores, generation probabilities or multi-model voting; setting a reasonable degradation threshold balances between wrong answers and refusals to answer.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFor regression testing metrics, it is recommended to record exact match rate, key fact recall rate, refusal rate, and hallucination rate simultaneously; looking at multiple metrics together can avoid hidden degradation where the pass rate rises but answers become more conservative or more verbose.78% similarUnverified多次采样取多数票或自洽性方法可在推理层降低单次随机性对最终结果的影响76% similarUnverifiedConfidence itself does not improve accuracy; its value lies in supporting routing by confidence, and the mechanism's effectiveness depends on whether confidence is calibrated—when discrimination is insufficient, threshold routing degrades into random routing.75% similarUnverifiedDenser relevance judgments reduce the probability that correct results retrieved by a model are mistakenly judged as irrelevant, thereby improving evaluation reliability.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1131041API
curl https://kongchang.com/api/v1/knowledge/claims/1131041MCP
get_claim(id=1131041)