This study investigates how sentiment is represented in Large Language Models (LLMs). The authors find that sentiment is represented linearly and identify the mechanisms involved, including a phenomenon called the summarization motif. Ablating the sentiment direction significantly reduces classification accuracy.

https://arxiv.org/abs//2310.15154

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