48 ms·
If you read the GPT-4 paper, they claim that the base model was well calibrated but the calibration was completely broken by RLHF
by sebzim4500 3y ago
If you read the GPT-4 paper, they claim that the base model was well calibrated but the calibration was completely broken by RLHF