Unverified50% confidenceOpinionExact time
In LLM-based agents, the model's confidence in its own output tends to be too high, making it easy to mistake "looks reasonable" for "already verified".
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
Unverified语言模型对工具输出有一定判断力，外部知识与模型内部信念的偏差程度越大模型越不相信77% similarUnverifiedAlignment techniques such as RLHF may in some cases cause models to overconfidently generate content that appears plausible but is actually incorrect, training models to 'sound correct' rather than 'be correct.'77% similarUnverifiedLanguage models tend to smooth vague expressions such as 'plan' into more confident, more fluent assertions, causing facts to be upgraded into less truthful statements during generation.74% similarUnverifiedLLM-based solutions can self-evaluate the reasonableness of intermediate steps through internal semantic understanding without fully relying on external reward signals, giving them an advantage in sparse feedback environments73% similarUnverifiedWhat distinguishes the verification loop from simply "asking the model again whether it is right" is that the basis for checking is the source text rather than the model's own judgment, so it can catch transcriptions that the model is confident about but that are wrong.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1101510API
curl https://kongchang.com/api/v1/knowledge/claims/1101510MCP
get_claim(id=1101510)