Agents Don’t Know When They Know Enough
中文摘要
AI代理最大的风险并非幻觉，而是在应当请求帮助或升级处理时，仍自信地执行错误操作。
English Summary
The greatest risk for AI agents isn't hallucinations, but confidently acting on them instead of deferring or asking for help when uncertain.
原文节选
The most dangerous failure isn’t hallucination itself. It’s an agent confidently acting on it when it should defer, escalate, or ask for… Continue reading on Medium »