Unverified50% confidenceFactExact time
d3-ultra is a discriminative scoring model, whose typical structure is an encoder or fusion encoder followed by a scoring head that computes a logit for each candidate answer and obtains a probability distribution via softmax.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFor 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 answer68% similarUnverifiedLogprobs are the logarithmic form of the probability distribution a language model gives when generating each token; when enabled, the API additionally returns candidate tokens and their log probabilities, and pricehunt uses this to soften binary yes/no judgments into a 0~100% match percentage.67% similarUnverifiedLogits are the raw unnormalized scores output by the model's final layer; after softmax they become the probability distribution over each token, and they are the most observable quantity closest to "model confidence."66% similarUnverifiedAn autoregressive language model produces a logit vector of vocabulary size at each prediction step, and its structure determines the conditional probability distribution of the next token.66% similarUnverifiedThe study compared three strategies for LLM-generated codebooks: top-down (theory-driven), bottom-up (data-driven) and hybrid.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102929API
curl https://kongchang.com/api/v1/knowledge/claims/1102929MCP
get_claim(id=1102929)