This is what I and I'm sure many other gamers/gamedevs thought when LLMs/ChatGPT first came on the scene:

Dynamically-generated dialogue. Device-local models designed for games could be trained on the game's lore and given a different style for each NPC, and react to the player's random actions instead of blurting out canned responses each time.

I'm sure if the genre-starting veterans like Richard Garriott or Roberta Williams were still active in the industry they'd definitely want to use something like this. Imagine a King's Quest or Monkey Island where you could literally type or try anything and get a relevant response! instead of just "nope"

▲

standardly 5 days ago | parent | next [-]

I've floated that idea around in various internet comments and am usually met with the typical AI=BAD response. It would work so well in a CRPG type game, but I think a lot of folks would review bomb a game like that just because

▲

rkomorn 5 days ago | parent [-]

My main issue would be that, especially with dialog, I'd never trust whether it's something the game's writers would think is canon, correct, etc.

It would probably creep into my suspension of disbelief and mess with my enjoyment.

▲

Razengan 4 days ago | parent [-]

> It would probably creep into my suspension of disbelief and mess with my enjoyment.

You mean trying some interaction that should be plausible in a point-&-click adventure game but getting told "I can't do that" because the devs didn't think of it, or typing some slightly different grammar in a text adventure game and the parser not understanding common words, or getting the same responses over and over in an AAA+ RPG once the prewritten dialog tree runs out, does NOT creep into your suspension of disbelief?

▲

rkomorn 4 days ago | parent [-]

Honestly? No, it doesn't, but maybe suspension of disbelief isn't the right wording.

Typically, when I play games, I build up some sort of mental model of how the game's meant to function, and how the game's creators want me to experience it. eg: running into useless/limited dialog trees with an NPC tells me it's not important. Not being able to do something tells me that thing isn't relevant (although sometimes, I guess it is fun to try something fruitless for a while). It doesn't have to be perfect but it helps if it's consistent.

What I understand about programming and game design leads me to accept there are going to be limitations encountered and tradeoffs made, so they rarely bother me beyond the first time I experience it.

With a more open-ended, AI-driven experience, I think I'd just end up wondering if the interactions I'm getting make sense, if they should be taken seriously, etc. I do think games have a lot of artistry (design, assets, writing, etc) behind them, and I suspect having "live" AI-generated stuff would distract me by making me wonder if I'm actually getting the right vision.

It's all speculative, of course. Maybe I'd try it out and find it cool. It's certainly not something I think shouldn't be attempted. :)

	▲	standardly 3 days ago \| parent [-]
		I think the solution to your hang-up, which doesn't exist yet (AFAIK) is an LLM that is fine-tuned specifically for strict narrative context. The NPCs would then be trained on its own lore, and prompted accordingly. Hallucinations would for sure completely ruin the immersion, break the fourth wall, etc, so this thing would need to have some level of deterministic control. I'm hopeful for the immediate future though, seems like something that is totally doable even if it isn't 100% perfect at first.

▲

rkomorn 5 days ago | parent | prev [-]

Live commentary in sports games is a place where I think LLMs could shine, but I have no idea if they could perform, time wise, to the point that it would sound right.