Large language models (LLMs) like Llama-2 learn coherent models of data generating processes, as evidenced by their linear representations of space and time across multiple scales, robustness to prompting variations, and identification of "space neurons" and "time neurons".

https://arxiv.org/abs//2310.02207

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