> There is no good reason this should not work everywhere else, in exactly the same way. Take for example a large retailer who has a large internal knowledge base. Train an LLM on that corpus, ask the knowledge base any question.

Since LLM’s can’t scope themselves to be strictly true or accurate, there are indeed good reasons, like liability for false claims and added traditional support burden from incorrect guidance.

Everybody is getting so far ahead of the horse with this stuff, but we’re just not there yet and don’t know for sure how far we’re going to get.