Unverified50% confidenceFactExact time
In autoregressive language models, logits are the unnormalized scores the model gives to each token in the vocabulary, which only become probabilities after softmax.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe confidence scores output by language models are softmax-scaled probabilities reflecting relative preference rather than calibrated true probabilities, and models that have undergone RLHF or instruction fine-tuning are generally poorly calibrated.69% similarUnverified模型对所有Token计算原始得分Logit后，通过Softmax函数将得分归一化为概率分布，所有Token的概率之和等于168% similarUnverifiedSynthID-Text使用密钥和前文token序列生成伪随机分数，叠加到模型输出的logits上以微调token被选中概率68% similarUnverified大语言模型是概率性、随机性的，它根据之前出现的词元来预测下一个词元68% similarUnverifiedFor multiple-choice questions, the logits over each option token can be normalized via softmax to obtain a probability distribution, where a highly concentrated distribution indicates the model is more confident in the answer67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1016203API
curl https://kongchang.com/api/v1/knowledge/claims/1016203MCP
get_claim(id=1016203)