Reply to thread

Message: <blockquote data-quote="Ruin Explorer" data-source="post: 9312503" data-attributes="member: 18">The central and much-overlooked problem with current AI remains.Somebody has to pay for it.This stuff isn't genuinely free (the "free" models that exist are "free" because you are the product - by using them, you're increasing their value and helping them to refine and publicize their product). It doesn't grown on trees. It's all processed extremely heavily off-site. It means that any service which offers this is basically going to have to be either a loss-leader, or subscription-based, and if it doesn't make enough money, it's going to to go away. This wouldn't be true if it ran locally, but we're a pretty long way from that (despite Microsoft really wanting it to happen - ask again in 5-10 years).Two other, lesser issues:AI is extremely bad at this, currently.To call it "unimaginative and trope-y" would be insulting to stuff that unimaginative and trope-y. You can see this very easily for example if you ask an LLM to come up with a list of foods for a fantasy setting, or a potential locations. They'll all be both completely over-the-top to often insane degrees, fail to understand things like "eating mermaids is probably going to be considered cannibalism, dude", and they'll also be compleeeeeeetely generic and trope-y. Like the most obvious and dumbest ideas you could come up with. Just really staggeringly generic stuff.Tables written by humans currently are quite a bit better than what AI tends to come up with here, at least with the current LLMs - looking at you particularly ChatGPT. LLMs are very good at doing things like giving you a generic overview of a situation, or a completely generic example of something. They're horrible at originality or real specificity, because they literally don't understand anything - they just put words together. This is why they'll think mermaid stew is totally cool (fantasy creature + food), because they don't know that a mermaid is also a person and we don't eat people.With a human in-between, that's okay because you can filter out the AI's absolutely dumbass ideas, but with an AI DM, you stare into the naked idiocy of an LLM AI.The other major issue is that any AI owned by a company like WotC, is likely to have quite a significant collection of "stop-words" and "no-go-areas", and some of these are going to be extremely dumb and silly. This has been an issue since the beginning - the ancient procedural AI game Facade (about visiting a couple for dinner), for example, would always make the hosts boot you, the guest, out, if you said the word "melon" in any context. I'm sure any WotC AI won't be that crude, but it'll absolutely be boneheaded about certain things, and no doubt people who use it will have to learn elaborate ways to "work around" it, rather than being able to consistently use it naturalistically.</blockquote>

Verification