2025-01-15 14:52 UTC
@aaron@chirp.zadzmo.org this looks interesting, I'll check it out later
I was thinking of building smth like this but it also uses an llm to generate garbage data to poison their datasets
it would need a bit more power, but it'd cache the pages, so it'd use a tad bit more cpu. but, I also plannes for it to be like the absolute tiniest llm I could find. it could also stream the pages very slowly to the bots, just to be annoying.
but this uses a markov chain, so that might be sufficient? unsure.
Replies (1)
-
@solonovamax@tech.lgbt 2025-01-15 14:56
@aaron@chirp.zadzmo.org also what I planned to do was to use a regex to match against the bot's user agent, but only when scraping pages that actually exist. then, if it matched it would return a page with links to the other fake pages. but, all the fake pages would use smth like /w/ as a prefix to it. and if any client made a request to /w/, then it would always return the fake page, because only the llm scrapers would ever know about those fake pages.