Large language models struggle with generalizing from "A has B" to "B is a feature of A" due to Zipf's law. Reverse training doubles tokens, improving performance on reversal tasks.

https://arxiv.org/abs//2403.13799

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