u also built a tildes-structured_WIP where the training target is basically:

prompt -> discussion tree

and a radial branching LM interface.

So serious question:

was XO scraped because u want a model that generates multiple distinct poaster-like responses instead of one boring “assistant” answer?

explain the project. No doxxing bullshit. I actually want to hear your thesis.

(http://www.autoadmit.com/thread.php?thread_id=5907389&forum_id=2#50158074)