Unverified50% confidenceOpinionExact time
"Knowing when to stop answering" is a fundamental capability of document-based AI applications; if a system cannot tell what it does not know, then all the answers it gives are suspect.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIf a system scores primarily on whether an answer is given rather than whether an answer should have been given, then responding 'I don't know' is counted as a failure.77% similarUnverified通过系统提示词约束智能体只回答知识库中有的问题，可有效防止智能体回答无关问题并引导转人工75% similarUnverifiedAny task with a verifiable result—where you may not know how to do it but can judge whether the answer is correct—can essentially be solved by AI.75% similarUnverifiedIn prompt design, prohibiting the AI from repeatedly asking questions while requiring every uncertainty to be written as an assumption and logged can prevent the AI from falling into an infinite loop when given free rein to ask questions, and also prevents directly having it guess and producing unexpected results.74% similarUnverifiedIn AI programming, using deterministic tools for post-hoc checking is more effective than piling rules into prompts, because rules in prompts tend to be ignored due to the 'lost in the middle' phenomenon.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1049654API
curl https://kongchang.com/api/v1/knowledge/claims/1049654MCP
get_claim(id=1049654)