In-context learning in large language models (LLMs) is a powerful learning paradigm, but its underlying mechanism is not well understood. This paper shows that the functions learned by in-context learning have a simple structure and can be seen as compressing the training set into a single task vector. Experimental evidence is provided to support this claim.

https://arxiv.org/abs//2310.15916

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