Unverified50% confidenceFactExact time
Under a shared-covariance Gaussian model, LogCDM's sigmoid gate is equivalent to a Bayesian posterior.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGCD uses the model's own Jacobian matrix as a constraint to solve an optimization problem, making the direction of the logit gradient exposed to the attacker maximally misaligned with the direction of the true adversarial loss gradient.63% similarUnverifiedLeveraging the logprobs interface of existing LLMs to request only the log probabilities of specific candidate tokens can achieve the effect of a decision model without replacing the model; OpenAI's logprobs parameter supports this mode.62% 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."62% similarUnverifiedThe core idea of contrastive decoding is to run a strong model and a weak model simultaneously and subtract their output logits to amplify the strong model's distinctive tendencies relative to the weak model.61% 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 answer60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1154289API
curl https://kongchang.com/api/v1/knowledge/claims/1154289MCP
get_claim(id=1154289)