Unverified50% confidenceFactExact time
In small language models, confidence signals based on token entropy almost entirely fail to distinguish between correct and incorrect answers.
1
Sources
50%
Confidence
Long-term
Relevance
9/21/2026
First Seen
Sources
Related Entities
Related Claims
Unverified对齐悖论指若模型真实理解比其能输出的语言更精细，则基于语言输出的对齐评估可能无法真正捕捉模型内部状态的对齐程度68% similarUnverifiedWhen a model lacks accurate knowledge of a particular library, it tends to generate syntactically correct but semantically incorrect code—so-called 'confident errors.'65% similarUnverified在缺乏真实接口响应样本的情况下，AI生成的用例无法完成有效断言，只能产出「状态码为200即通过」这类近乎无效的校验规则64% similarUnverifiedThe study observed a clear lack of correlation between the correctness of answers and the validity of reasoning traces, with models frequently arriving at correct answers even when the reasoning traces are completely invalid.64% similarUnverifiedAI hallucination is not a random failure but an inevitable result of low-quality context input; when the input context is sparse, the model lacks a handle to anchor itself to real information.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/938052API
curl https://kongchang.com/api/v1/knowledge/claims/938052MCP
get_claim(id=938052)