davisyoshida.bsky.social

global x = 1: value set to 1 process-wide
stellar x = 1: value set to 1 across all processes running this program, wherever they are
galactic x = 1: value set to 1 across all programs in this language starting now
universal x = 1: value has always been 1, since the big bang

I run a chatbot for my friends in our Discord. I get the best results by training in chronological order, since I want it to be as "up to date" as possible. The old data helps, but you don't want it to define the final learned distribution.

This is only vaguely related to the original question.