Glue pizza and eat rocks: Google AI search errors go viral

Not posted because it's a technical deep dive, but because it might prompt an interesting discussion on society moving from deterministically-correct responses to probabilistically-correct ones.

Which seems like a distinction (a) the general public isn't aware of & (b) LLM product companies aren't clearly discussing/educating.

Especially as they get more embedded in products, and it becomes less clear when you are/aren't getting LLM responses.

Something that's right 99% of the time but convincingly suggests you make spicy spaghetti with gasoline 1% of the time is fundamentally unlike what came before.