The paper proposes a fine-tuning strategy for large language models that maximizes the marginal log-likelihood of generating a correct answer using a "chain-of-thought" prompt. The strategy involves sampling from the posterior over rationales conditioned on the correct answer using a Markov-chain Monte Carlo algorithm. The technique improves the model's accuracy on various tasks compared to other fine-tuning methods.

https://arxiv.org/abs//2312.02179

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

TikTok: https://www.tiktok.com/@arxiv_papers

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

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