Hacker News Reader

That's because we don't leave them running, right? We could, though, yes?

Current LLMs decohere rapidly, and as far as I am aware, this problem seems like a fundamental to the architecture one.

Yup.