This is one of the ways in which generative AI in games is going to be awesome as the tech matures.
I remember playing the Zork-like text dungeon AI using GPT-3 as the backbone.
It started with me as a wizard looking for a certain book.
When I found my way to a library, I could have it read the various titles of each book on each bookcase, and even flip open to a page, which in turn led to a completely different path of surprise and mystery.
Pairing procedural generation with generative AI within a properly bounded set of constraints and caching variations is going to lead to another level of immersion and discovery from anything we’ve seen yet.
It’s become nearly frustrating playing games now and seeing where they fall short of what it’s going to be by this time next ‘generation’ of games.
World density and that sense of discovery around each corner that’s not being spoon fed to you in advance via behaviorally targeted YouTube video previews or junk game articles summarizing Reddit posts is going to be a wonderful thing.
Games are going to be going through a whole new Renaissance as the industry adjusts and adapts to emerging tech.
The issue is more generation is leaned on the less powerful it becomes. To create those stores that you were reading in books it was fed lots of other fantasy and game information, as more of its used the less ‘new’ stuff it can make.
The breadth of the modern LLMs is wide enough there’s no need for further broad training.
There’s certainly room for fine tuning (likely coupled with a discriminator) around specific world lore, like feeding it all dialogue and book text from all the Elder Scrolls games if you wanted it to generate things fitting with that universe.
But there’s no limits on how ‘new’ its stuff can be as long as properly seeding its prompts.
You’ll generally see development moving towards manual development for core story and generative AI for the extended universe. Almost like a literary foveated rendering, filling in the peripheral world with generated content while the core story beats are manually directed.
Also, keep in mind that by the time AAA games release where their design is predicated on generative AI existing, we’ll be about 2-3 generations ahead of where we are today in generative AI products.
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: [email protected]
Sub for any gaming related content!
Rules:
1: No spam or advertising. This basically means no linking to your own content on blogs, YouTube, Twitch, etc.
2: No bigotry or gatekeeping. This should be obvious, but neither of those things will be tolerated. This goes for linked content too; if the site has some heavy “anti-woke” energy, you probably shouldn’t be posting it here.
3: No untagged game spoilers. If the game was recently released or not released at all yet, use the Spoiler tag (the little ⚠️ button) in the body text, and avoid typing spoilers in the title. It should also be avoided to openly talk about major story spoilers, even in old games.
This is one of the ways in which generative AI in games is going to be awesome as the tech matures.
I remember playing the Zork-like text dungeon AI using GPT-3 as the backbone.
It started with me as a wizard looking for a certain book.
When I found my way to a library, I could have it read the various titles of each book on each bookcase, and even flip open to a page, which in turn led to a completely different path of surprise and mystery.
Pairing procedural generation with generative AI within a properly bounded set of constraints and caching variations is going to lead to another level of immersion and discovery from anything we’ve seen yet.
It’s become nearly frustrating playing games now and seeing where they fall short of what it’s going to be by this time next ‘generation’ of games.
World density and that sense of discovery around each corner that’s not being spoon fed to you in advance via behaviorally targeted YouTube video previews or junk game articles summarizing Reddit posts is going to be a wonderful thing.
Games are going to be going through a whole new Renaissance as the industry adjusts and adapts to emerging tech.
The issue is more generation is leaned on the less powerful it becomes. To create those stores that you were reading in books it was fed lots of other fantasy and game information, as more of its used the less ‘new’ stuff it can make.
That’s really not the issue.
The breadth of the modern LLMs is wide enough there’s no need for further broad training.
There’s certainly room for fine tuning (likely coupled with a discriminator) around specific world lore, like feeding it all dialogue and book text from all the Elder Scrolls games if you wanted it to generate things fitting with that universe.
But there’s no limits on how ‘new’ its stuff can be as long as properly seeding its prompts.
You’ll generally see development moving towards manual development for core story and generative AI for the extended universe. Almost like a literary foveated rendering, filling in the peripheral world with generated content while the core story beats are manually directed.
Also, keep in mind that by the time AAA games release where their design is predicated on generative AI existing, we’ll be about 2-3 generations ahead of where we are today in generative AI products.