1 paper · 2 filters
Yoo Yeon Sung, Eve Fleisig, Yu Hou +2
Language models are often miscalibrated, leading to confidently incorrect answers. We introduce GRACE, a benchmark for language model calibration that incorporates comparison with…