Unverified50% confidenceBenchmarkExact time
Claude Haiku 4.5 answered 9 out of 10 questions in the judge format that should have been escalated, reaching a false-confidence rate of 90%, while its overall performance ranked at approximately the one-third position.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedDecider reports a correct rate of 99.6% for judgments at 90% confidence and above on JudgeBench, 94.0% for the 80–90 interval, 84.1% for the 70–80 interval, and 67.7% for below 70%.71% similarUnverifiedResearch shows that when humans assign 90% confidence, their actual accuracy rate is often only around 70%, indicating a systematic overconfidence68% similarUnverified即便单字符识别准确率达到99%，对于一个包含10个字符的字段，其整体正确率也仅约90%，这被称为"准确率衰减效应"67% similarUnverifiedPrecision bias refers to the tendency for people to assign higher credibility to statements containing precise numbers; experiments show that subjects rated the credibility of "41.3%" significantly higher than "approximately 40%".66% similarVerifiedIf the probability of the AI making a correct decision is 90%, blindly following the AI's decisions 10 times in a row yields a final accuracy rate of only approximately 35% (0.9 to the power of 10).65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1002961API
curl https://kongchang.com/api/v1/knowledge/claims/1002961MCP
get_claim(id=1002961)