[lupusreal]

> for example, I don't want to see yuppie shit on news.ycombinator.com. right now, I have a userscript that filters out links from the worst offenders (e.g. newyorker), but every day, there are many more that shit up the front page and waste my time and attention. this could be solved with a very low parameter LLM by asking it to evaluate whether the text is related to science and technology or not."

FWIW, I once used naive Bayes classification solely on HN headlines and it worked reasonably well for solving the problem you're describing.