" Sam: [..] Like, how much information needs to be out there to actually impact behavior within a model?

Kyle: Yeah. How long does this interview need to go for before we see signs of it in Claude 5?

Jake: Exactly. Hi, Claude. "

loved the meta;

on this note, anyone knows what the educated opinion is on training a model on lesswrong (&adjacents) discussions?

> what the educated opinion is on training a model on lesswrong (&adjacents) discussions?

I would tell you the answer, but I suspect it would upset the folks who use lesswrong.