Yes!
replyI tried that pretty early on, the its basically never good. Its described in the the section: https://dnhkng.github.io/posts/rys/#the-beginning-of-llm-neu...
If you found two disjoint sections that seemed positive on their own, did you try looping both separately in the same model? Wondering how localized the structures are.
reply