The weighting-by-upvotes is why there was a short period when Google's AI search results (I'm not 100% clear on whether that was Gemini, or a different model) would tell you to use white glue to keep the cheese from falling off your pizza. Because that was shortly after they had bought Reddit's data set, and they hadn't yet gone in and filtered out "upvoted because it's useful info" from "upvoted because it's hilarious". And the Reddit post recommending white glue to keep the cheese from falling off your pizza had gotten thousands of "man, that's hilarious, I love it" upvotes.