"Just phenomenal stuff here, folks. Every time someone like me reports on Google’s AI getting something wrong, we’re training the AI to be wronger." 🧪 www.theverge.com/2024/6/11/24...
Google bombing is back, but now with AI.
Is it training anything when the checkpoint is in the same state? A simple script that runs an idiotically powerful cluster to couch the first search result in an "according to" sentence will do the same thing whether the first result is a joke post or an article about the joke post
Inevitable without constant human curation, even if that's nothing fancier than a "suggest a better response" button a la Google Translate. However, that would betray the core lie of the grift, that these machines need no human monitoring, so they won't fix it until there's a body count.