you actually need an entire LLM specialized for this WITH reasoning, because it will just never catch those silent quirks i just mentioned
yes, sure, associating somebody with a specific stylometry is impressive, but is it really useful if the person tries to spoof it?
youre not suddenly going from hating music to quoting lyrics from the '80s songs in the chat
nor will you deliberately argue worse then you get in an argument or differently
(nor will you suddenly change talking points)
yeah, if you have ONE specific spelling mistake on a word its a full giveaway, but does the model catch that? probably not
what im saying, if you want to fine-tune an LLM, then just your training thing isnt enough, i dont think it will be good because it will become too slow and too expensive for you to run this as a hobby, since surely youll need glm-5.1 distills and stuff











