Unverified50% confidenceFactExact time
The industry uses the over-refusal rate as a metric to measure the difficulty of model calibration.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe over-refusal rate and the harmful-output rate exhibit a trade-off during parameter tuning, which in machine learning is known as the precision-recall trade-off.74% similarUnverifiedAnthropic's research shows that excessive alignment may cause a model's refusal rate on harmless questions to rise from under 1% to over 10%.70% similarVerifiedIn a 2024 research paper, Anthropic quantified the phenomenon of model over-refusal, finding that safety-trained models had false positive refusal rates as high as 15–30% in certain scenarios70% similarUnverified分层模型路由策略若分类不准确或过度偏向成本控制，会导致用户感知到的质量下降69% similarUnverifiedMost general-purpose large models suffer from overconfidence issues, where their output probabilities reflect token distribution preferences more than genuine uncertainty estimates.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999012API
curl https://kongchang.com/api/v1/knowledge/claims/999012MCP
get_claim(id=999012)