HN: 354M LLM with better epistemic calibration than GPT-2 Medium on OOD datatruthagi.ai·1 pts·felipemayamuniz·1