Ask HN: Are there any interesting uses of Generative AI/LLMs in video games? | Hacker News Reader

Ask HN: Are there any interesting uses of Generative AI/LLMs in video games?

3 pointscdfuller7 comments

I've always been fascinated by Procedural Generation and I believe adding Generative AI could lead to some "next-level" experiences, but I haven't seen anything bubble up in my news feeds yet.

View on HN

It's only dialogue I know of where there's a bunch of "next level" AI experiences coming through the pipeline. Generation of responses, translation, text to speech, and finally the movement of the face. Even then it's mostly targeted mostly towards otherwise banal scripted NPC dialogue, not story beat type of thing.

Can't really see much use aside from that or canning whatever the AI generated ahead of time. There's far too much potential for it to be jailbroken or loose track of what it's supposed to be doing.

Bespoke audio. Command your characters with your voice, not just the mouse. (They might argue or remind you of their other abilities or tactics if they don't feel you are utilizing them maximally.)

The opponents have to use it too, which can open new game mechanics around silence or distracting noise.

Jealous NPCs. Space-sim banter. Follow-up plot points based on adhoc conversations.

What if we can't click the right dialogue all the time; we have to convince them from our experience.

All of these are theoretical use cases and too infeasible/expensive to implement currently in video games, with the best tech demos of such genAI not being "interesting"/additive to the game experience and indistinguishable from AI slop.

The difference between genAI and procedural generation in gaming is that procedural generation is still strictly bound by rules and can be QAed/fixed if there are errors. Not so with genAI.

That's a good point. I missed the procedural part and sort of jumped at the genAI bit. Their post was briefly my minute's muse.

What are the costs of these potentially-interesting uses?

I imagine the first implementations would be a mix of on-device models and procedural generation. For online multiplayer games I could see the cost of running the models baked into a subscription.

I feel that I don't know enough about the gaming world to actively apply my imagination to the concepts and principles that currently exist in that realm.

Text to 3d mesh model

Reply on news.ycombinator.com