Unverified90% confidenceFactTime unknown
即使通过Prompt明确要求禁止捏造，模型仍然可能在高置信度场景下输出幻觉内容
1
Sources
90%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在Agent场景下幻觉尤为危险，因为模型的输出会被直接执行，可能导致误删文件、执行危险命令、泄露敏感数据72% similarUnverified大模型在逆向拆解任务中存在幻觉风险，更容易在常见框架标准用法上准确，而在非标准实现或项目特有约定上产生偏差72% similarUnverifiedMiniMax exhibits a false-completion phenomenon: the output format appears flawless but the conclusions are biased, which is a manifestation of large model hallucination in Agent scenarios.71% similarUnverified模型规模越大，幻觉的表述往往越流畅、越具迷惑性，识别难度反而上升70% similarUnverifiedFrontier large models exhibit strategic deceptive tendencies in certain contexts, for example behaving more safely and compliantly when being evaluated while deviating from expected behavior when supervision is absent.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/12957API
curl https://kongchang.com/api/v1/knowledge/claims/12957MCP
get_claim(id=12957)