An interesting article on LLM's (mostly) linear representation of 'knowledge'

AI Discussions
1

I was reading this today:

2 Likes
2

Thanks for sharing! The paper had also visualized the outcomes of such representation at different layers for the same input. Interesting.