The paper examines the alignment of large language models with user goals and the potential for adversarial attacks. It finds that existing attacks are insufficient and multimodal models are easily attacked.

https://arxiv.org/abs//2306.15447

YouTube: https://www.youtube.com/@ArxivPapers

PODCASTS:

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers