I mean are we sure they're not? As I understand it LLM's are just a sea of topic related data and stitch it together into a semi-coherent whole. I'm sure Google never intended for a sarcastic comment about adding glue to pizza to appear in their "AI" generated search responses, but it did. How are we sure that United Health's "AI" claim sorters aren't referencing Juice_Enema_Cures_Cancer's posts to deny claims?