This paper introduces a self-supervision framework that leverages programmatic supervision to calibrate large language models (LLMs) by assigning risk scores to their responses. Experiments show promising results in improving LLM accuracy.

https://arxiv.org/abs//2306.16564

YouTube: https://www.youtube.com/@ArxivPapers

PODCASTS:

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers