Concept置信度估计
Confidence Estimation
A technique for evaluating the degree of certainty a large language model has in a generated answer, commonly using log probabilities or self-consistency sampling as metrics.
Timeline (last 90 days)
Oct 22
In confidence assessment scenarios, RAG can perform entailment judgments or factual comparisons between the model's answer and retrieved authoritative text, treating an answer as low-confidence if it cannot be supported in the evidence.
Unverified50%