I think the Othello GPT paper and the platonic representation hypothesis papers are the most interesting. There's also a lot of digital humanities work inadvertently demonstrating these mechanisms. As well as a lot of the jailbreaking "literature" imo

Not minds. Fight me on it, please. Jfc. I'm saying many of you are not only using these things wrong but also especially in the valley, there's genuine mass delusion going about what "emergent behaviors" are and where they're coming from. So tell me where I'm wrong, if you think I am. All feedback is always welcome.

Sigh

[dead]

[dead]