Discussion
Things I hate about LLMs: that novelists are being challenged to prove they wrote their own work.
Things I don't hate about LLMs: that they give me an excuse to put "trap streets" in my work—really obscure archaisms and eldritch neologisms that no AI would come up with.
@cstross Sometimes I think about how very mild pinkie-depth references to history crop up in my writing, and often enough people don't even pick them up and are like "hey that was emphasized like a proper noun, is it something big".
I imagine the average LLM would spit out multiple paragraphs of emphasis on such a reference, but I've deliberately not interfaced with these systems for any writing related notions. The output I've seen cited is banal and self-aggrandizing, and that's my thing!
@cstross street mimics street mimics street mimics.
this dark alleyway isn't *supposed* to be there, it's only on the map to catch plagiarists... if you walk in, you might not come back out. and when they come looking for you, all they will find here is a plain wall. afterall, it's only on the map to catch copycats.
@cstross heck, that's been my gimmick as a mostly online writer. neologisms FTMFW
@cstross
How I recognize a real Charles Stross novel: the unabridged Webster besides my bed doesn't contain all the words I need to look up.
@cstross Only until some fiend distils a, "Stross-bot," trained to write in your particular style, complete with pitfalls and foibles.
<insert Mr. Burns gif here>
Install bonfire.cafe
Get the full app experience
@cstross If the prompt said "Write a 20,000 word novella in the style of Charles Stross including occasional archaisms and eldritch neologisms," it would try to do exactly that. It would probably only take a few minutes. In my one test of AI I asked it to write a freshman level paper on Kipling's Rikki-Tikki-Tavi and it did exactly that in 15 seconds. I would have given it a B+. Then I asked it to do a senior level paper using post-colonial theory and it did that too, in another 15 seconds. It cited Gaytari Spivak, Homi Baba, and another theorist, hitting all the obvious points for another B+, maybe even an A. And this was using an inferior version of Co-Pilot. The new ones are far better. My conclusion was that it was useless to continue assigning such papers.