AICohereabout 13 hours ago
Cohere Unveils CALIBER for AI Confidence
6 min read
Cohere introduces CALIBER, a method training reasoning models to report initial and final confidence estimates. On the BigMathDigits benchmark, the 7B model reduces calibration error by 52.5% compared to the best single-confidence baseline. The approach improves performance on GPQA, TriviaQA, and SimpleQA without extra training.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyCohere
