Maybe this goes in the Tech thread, but I'll just drop it here....
I've mentioned a number of times how crap Google's AI is and it shouldn't be trusted. Yesterday I Google'd the score of the CWS g2 and it told me UNC had defeated OU by a score of 6-5 in g2 to force a g3. This was while the game was still in the top of the 7th inning.
These AI's have started sourcing their garble, with links to pages you can click on. If the model is working as advertised, it's telling us which pages it's trying to summarize with all its AI wisdom. One of the pages had no text, it was just an ongoing box score, which also indicated the game was not done (and the score was also not 6-5). Well, nothing will trip up a LLM summarizing something like having no text to summarize. That didn't stop it from trying.
It had more paragraphs, each linking to different pages it thought it was summarizing. They were all either score updates or info billing the game the way that articles read prior to the contests. Gemini, however, does not let a lack of information stop it. It just hallucinated right on to certainty.
otoh, it did accurately pick the winner, I guess, so there's that.