primer for how LLM's work
this is grossly simplified, k
acquire huge volumes of text — it's a large LANGUAGE model — but don't bother about the context, that's an out of domain subtlety yet to be addressed
this is a ginormous array of lexemes, inappropriately constrained to mean words
1/4
¹that was the fuckup, human intelligence doesn't rand(), it attempts a closeness within the author's domain of understanding, which can be interpreted by the broad context of the author's writing
3/4
make another array of the lexeme that's most likely to occur after the current lexeme
oops, lots of them have equal weighting
use rand() to choose¹ the next lexeme
branch back and consider the total result
still statistically correct, continue
2/4