on llms

a note about how the notes get written, because i’d want to know.

everything here is mine. the ideas, the opinions, the arguments, the things i turn out to be wrong about. all of it started in my head and stayed there through however many drafts it took.

a language model is sometimes involved in what happens after that. mostly a local one, running on my own machine rather than somebody else’s. specifically, it may:

• rephrase a sentence i couldn’t get right on my own
• fix grammar, punctuation and the typos i stop being able to see after the third read
• formalise something i wrote too casually, or loosen something i wrote too stiffly
• talk me out of saying something that will get me in trouble, or cut a sentence that came out stronger than i actually meant it
• handle the mechanical parts of publishing: files, formatting, the site itself

what it does not do is decide what i think. if a note argues something, i am the one arguing it. if a note is wrong, i am the one who is wrong. nothing here is a model’s opinion wearing my name, and nothing here went up without me having read every word of it and agreed to it.

i don’t think using a language model this way requires an apology. i do think it requires saying so.