AI
Researchers build a benchmark for AI answers that admit uncertainty
The project measures whether a model knows when evidence is weak instead of rewarding confident language alone.
The project measures whether a model knows when evidence is weak instead of rewarding confident language alone.