i've started collecting material to train an LLM... so far i've got mises, tesla, szabo, a bunch of DST theory papers, some space weather, the bible, enoch, jasher, jubilees, critias... gonna just pause every so often and trawl for more resources until i've got way too much for me to make sense of. i'm going to make a little model that runs on my 16gb radeon 7800xt, that is specifically focused on economics, cryptography, mathematics, chemistry, physics, computer science, and i've still gotta dig up a ton of Go stuff, repos, papers, documentation of various kinds. i might start training it before i've collected all the data tho. from what i've read it's a weeks long process. but if i specialise it in the kinds of topics that i have use for being able to dig through to back a hypothesis, it can help me check what i'm doing and make sure it matches up with the various models i've fed it to build from. i'm not convinced that general purpose AIs like Claude and Gemini et al are even slightly optimized for science. i'm betting the woke content percentage of those models is too damn high. too much marxist theory, too much mainstream narrative bias. especially not computer programming. oh yeah, they say claude knows basically all the languages but my eyes must be lying when i watch it constantly forget salient syntax matters over and over again like a green n00b learning to program.