| ▲ | fzeroracer 14 hours ago | |
> Games are literally one of the absolute best uses for this tech. Someone is going to figure out a way to leverage this stuff as a first class function of the gameplay. Player directed assets or procedural world creation, dynamic quests like never before seen. They're going to hit the nail on the head and produce a banger with this tech and the landscape is never going to be the same again. I see the AI boosters say this over and over again. People do not want it. Games have tried the whole dynamic NPC stuff along with generated quests and it's immersion breaking and more importantly, not fun. | ||
| ▲ | ticulatedspline 7 hours ago | parent | next [-] | |
People definitely want it. I can say I want to see it happen. Also people don't know to want what they don't know exists. There just hasn't been the Edison light-bulb/Watt's steam engine/Apple iPhone moment yet. I'm not even sure generative AI is capable of it yet but I see the potential there. While not all dynamic games are "not fun" (lotta people really like Rimworld) that kind of content tends to feel flat, like the never-ending quests in Skyrim. But I think it's easy to squint and see how those shallow Skyrim quests could become far more interesting with a system that was capable of basically writing a story as you play. IMHO It's only a matter of time before someone gets it right. | ||
| ▲ | tancop 6 hours ago | parent | prev [-] | |
i think good dynamic quests will eventually be a thing but not coming from gen AI or old school complicated rule books. if you generate the whole thing with a LLM you get a lot of results that make no sense. if you go with a deterministic rule based generator or a constraint solver you get low diversity because the rules are fixed. my solution is a hybrid approach using a machine learning model trained to estimate player ratings for a quest script. you have a simple RNG that generates a ton of different script, the model picks out the ones that are likely to be fun and you give one of them to the player. the meta might be using all the methods (maybe not LLMs) together to get a result thats fun, logically consistent and even fits into a larger narrative. like tracking NPC relationships to assign a target for combat missions, coming up with extra plot twists along the way then using ML to rate all the generated options. | ||