Supervised fine-tuning of large language models introduces new factual knowledge, impacting model behavior. New knowledge is learned slower, leading to increased tendency to hallucinate factually incorrect responses.

https://arxiv.org/abs//2405.05904

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