The Reason That Google's AI Suggests Using Glue on Pizza Shows a Deep Flaw With Tech Companies' AI Obsession, someone looked up "cheese not sticking to pizza" on Google. The search engine's AI Overview authoritatively began that "cheese can slide off pizza for a number of reasons."
Mr. Fucksmith was responding to a thread titled "My cheese slides off the pizza too easily." It's easy to see why that would make its content appear relevant to the query, but it's hard to make sense of why the AI algorithm singled out this comment in particular, which only had eight upvotes at the time, from a thread that's equally obscure.
That the AI can't discern sincerity from shitpost is no surprise. It underscores a fundamental flaw to the predominant approach of training generative AI models, which is in a nutshell to feed them as much information scraped from the internet as possible. That inevitably leads to a lot of garbage being ingested, and then eventually regurgitated. In this case, the faults of this approach have a clear lineage: in February, Google struck ato train its AI on users' posts.