Unverified50% confidenceOpinionExact time
In high-stakes domains such as healthcare and law, the abstention capability—where the model prefers to give no answer over a wrong answer—is far more valuable than answering more questions.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
Unverified医疗AI中的选择性预测机制在低置信度时拒绝答案转介人工审核，被证明比强制输出答案更能降低临床风险75% similarUnverifiedIn high-risk scenarios such as medical, health, and safety domains, the cost of errors is high and model outputs cannot be trusted blindly.74% similarUnverifiedAgentic scenarios place far higher demands on a model's rejection rate and hallucination rate than ordinary Q&A, where stability and controllability are often more important than peak capability.73% similarUnverifiedThe greatest risk in having AI agents directly answer data questions is generating answers that appear plausible but are actually incorrect.73% similarUnverifiedWhen using large models to handle questions in rigorous fields such as law and medicine, one should not blindly trust their output.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961122API
curl https://kongchang.com/api/v1/knowledge/claims/961122MCP
get_claim(id=961122)