dedicated AND fucking crazy

i am not particularly interested in reading fanfiction generated by an LLM. this is because i am in fandom to read fic that engages interestingly with a canon i love, and an LLM definitionally cannot do this. there are also human-written fic that do not do this! this is not a unique quality of LLM-generated text. "there are fics not to my taste" was a problem long before LLMs entered the picture.

what is new is the scale. the quantity of text an LLM can produce will always outstrip what humans can write by orders of magnitude. maybe this is a problem. readers don't have infinite time -- what's the sustainability of backbuttoning from fics i don't find engaging for one reason or another, if the scale is so dramatically tipped to one side? i genuinely, deeply adore discovering fics with a unique voice or an unexpected perspective, but how much harder is that going to be to find as the volume of fics to sift through grows ever larger?

on the other hand: my preferences are not everyone's preferences. people clearly do read LLM-generated text, and like it.

i've generally found the morality dimension to this discourse somewhat unproductive because my question is always: how does an action materially lead to changes you want to see? and i don't think it's helpful to say, if you use a tool that is widely available then you are a bad person. famously this is not a tactic that makes people eagerly change their ways.

(i am always thinking about daryl davis, a Black man who deradicalized ku klux klan members by actively choosing to have conversations with them. i am thinking about how a child of the founder of stormfront disavowed the movement because of extensive discussion over shabbat dinners. people find it hard to be wrong. to change someone's mind you have to put in the work; you need to approach with empathy.)

i'm equally uninterested in saying that the problem with LLM-generated text is that it is bad. many fics are bad with no LLM involvement whatsover. probably the LLMs will get better, since a lot of money is being invested into the effort. but if you frame the conversation as, LLM-generated text is bad, and therefore liking it means you have bad taste, or you are a fool, or, or. again, what is the end goal here? what happens is that people get defensive. people say, i liked this thing i read, and therefore it is good, and therefore it can't be machine-produced. reading is an active act between the reader and the text -- people aren't wrong for bringing themselves to that conversation, regardless of the source. we can't make not recognizing LLM usage a moral failing.

on a third hand: claude has a voice. gemini has a voice. chatgpt has a voice. much as human writers have tics or an oft-used turn of phrase, these tools do, too. i say this not because i am seeking to identify specific texts as LLM-generated, but because i care about the effects in aggregate.

fandom is not the first space i wrote in, but it's one where i've written consistently, often badly, and over time developed a voice i feel confident about. i really think that's valuable and worth preserving. but the increasing volume of LLM-generated text makes this harder for someone who is just starting to write today. how do you develop your own voice when everything around you sounds the same? the first step to learning something is mimicry, but to build on that you need diversity.

play like a white onion