#nlu benchmarks

Do Small Language Models Know What They Don't Know? New ResearchNEW⭐9

New research shows token entropy fails as a confidence signal in sub-3B language models, while semantic entropy enables routing gains up to +50 percentage point...