Unverified50% confidenceFactExact time
Confidence scores generated by LLMs are not necessarily calibrated true probabilities
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThere is an essential difference between the probability scores (logprobs) output by LLMs and calibrated confidence; large language models are widely known to be overconfident, particularly on out-of-distribution samples.73% similarUnverifiedLarge language models do not inherently possess calibrated probability estimation capabilities, and the probability values output by a model do not represent actual occurrence rates.73% similarUnverifiedThe true/false probability pair (F = 1 − T) in traditional classification frameworks is algebraically redundant, because the probability of false is completely determined by the probability of true and provides no independent information.72% similarUnverifiedThe 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.70% similarUnverifiedLLM生成阶段对词汇表（通常包含32,000至128,000个token）计算Softmax概率分布，形成总和为1的概率向量68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1000298API
curl https://kongchang.com/api/v1/knowledge/claims/1000298MCP
get_claim(id=1000298)