Unverified50% confidenceFactExact time
The correlation between token-level probabilities (logprob) output by a model and factual accuracy is not strong, and fabricated facts may receive the same high generation probability as true information.
1
Sources
50%
Confidence
Long-term
Relevance
8/9/2026
First Seen
Sources
AI Agent可观测性：生产环境调试与幻觉治理新范式
Related Claims
Unverified泰特洛克的研究得出结论：专家预测的准确率仅略高于随机猜测，甚至不如简单的统计基准线68% similarUnverified作者验证有93%的概率质量正确对齐，信息损失并非映射逻辑的bug67% similarUnverifiedMultimodal large models have extremely low factual recall accuracy in video question answering, which is difficult to improve even with supervised fine-tuning.66% similarUnverified'99% 准确率'的模型在生产环境中仍可能频繁出错，因为剩余的 1% 往往对应现实中最关键的长尾细节场景65% similarUnverifiedIn class-imbalance scenarios, a model can achieve seemingly impressive accuracy by always predicting the majority class, a phenomenon known as the Accuracy Paradox.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/712732API
curl https://kongchang.com/api/v1/knowledge/claims/712732MCP
get_claim(id=712732)