Guys, what do you think about the integration of AI technology to the games right now?

[zqw975]

Except the crappy AI-generated picture games, my ideal game model doesn't seem to have arrived yet.

My ideal game model right now is it should give me more of realism sence and engagement in the game. For example, I can use AI to describe what clothes, figure,age or other characteristics of the female protagonist or the harem of the male MC I operate——my description should at least be presented in the form of pictures, and have more realistic response from passers-by. Or the AI program could give me more and freer feedback on my actions, allowing me to write as many endings as I want in those erotic, epic, or tragic game stories.

But in fact, the only I can find on the Internet is some AI chatbots, and they have no high quality, and these bots generally have poor memories, and their reactions are chaotic and formulaic, this is especially true for SillyTavern, which uses a free model. Even for the best of them, Friends & Fables, AI Dungeon and NovelAI, there are few that can even combine AI chat with generated images, I can only seem to find a website

You must be registered to see the links

, but the generated content has little to do with the game (such as generating modern city streets and nightclubs in medieval games). And they often have strange prohibitions, such as "8-year old piano" or "12-year weapon", or if you want to have sex with a race such as a described smaller, like dwarf, often trigger prohibitions of the site. It makes people feel strange, and it causes a lot of trouble.

In short, the current AI technology is far from my ideal, and the only AI-related technology I know of that is relatively practical for games is AI translation. Moreover, the existence of those inferior AI picture games has greatly affected my overall impression of AI's application in games. How do you feel about the integration of AI and games in general now?

[Please forgive my poor English, because I am a Chinese.]

 

[Doorknob22]

Perhaps AI will never (or at least not for many years) be as powerful as we imagine it about to become "any minute now". We humans have a tendency to imagine technology improvements as linear but they are often not: a leap in technology takes place followed by years or even decades of trying to reach the next phase.

In other words: perhaps current AI is as good as is gets for a long while.

 

[ayylmao1313]

what i think of is one of those text based/real porn games with Life Sim genre, with AI it can create bunch of characters with various dialogues that user can choose or write, oh man i cant wait for those days

 

[DuniX]

Perhaps AI will never (or at least not for many years) be as powerful as we imagine it about to become "any minute now". We humans have a tendency to imagine technology improvements as linear but they are often not: a leap in technology takes place followed by years or even decades of trying to reach the next phase.

The problem right now is not the technology.
If you ask it for a character it will give you a character.
If you ask it for a BG CG it will give you that.
If you ask it for a plot it will write you a plot.

The problem is massaging some of it's flaws and finding some workarounds around some problems and limitations, this does not mean there are no ways to use it.
And things are indeed improving in terms of context,memory and consistency. There are already papers and demos on that.

Once we get text to 3D models that are trained on good topology and procedural materials most of the issues with the art and consistency will be solved since you can feed that back into the 2D generator and tell it it's wrong and how to fix things.
In the first place most CG images that are created nowadays are based on a composition of multiple layers, similarly if we had multiple steps into the process and break things down into more parts and layers for the AI that will fix most of the issues.

There is no need for a "key breakthrough" for the AI technology, just finding a better Process for how the AI works.
You already see some of this if you look at the workflow for Stable Diffusion with working with diffrent artstyles.
There is no need to do that manually, all of that can be automated tomorrow, the reason we don't is we are still experimenting on what is the right workflow to turn it into the right process.

 

[ogr blanc]

genuinely have no issue with using AI in games. my issue is with the HOW.

 

[DuniX]

Even for the best of them, Friends & Fables, AI Dungeon and NovelAI,

That's mostly because they are long obsolete already.
It's just the cutting edge AIs aren't as specialized and integrated as they are.

The more fundamental problem is LLMs text generation is very hard to utilize, at best all we can think of is use them as a chatbot but talking for the sake of talking is meaningless.
The AIs cannot have a goal by itself that gives a conversation a purpose, that has to be implemented by the developers.
The problem is we haven't figured out how to do that yet, but it shouldn't be impossible to achive. You have tools like context, prompts and restrictions that could achive it.

 

[Uthuriel]

Perhaps AI will never (or at least not for many years) be as powerful as we imagine it about to become "any minute now". We humans have a tendency to imagine technology improvements as linear but they are often not: a leap in technology takes place followed by years or even decades of trying to reach the next phase.

In other words: perhaps current AI is as good as is gets for a long while.

Nuh uh...

I want my Holodeck so I can play "Vulcan Love Slave".

 

[♍VoidTraveler]

Perhaps AI will never (or at least not for many years) be as powerful as we imagine it about to become "any minute now". We humans have a tendency to imagine technology improvements as linear but they are often not: a leap in technology takes place followed by years or even decades of trying to reach the next phase.

In other words: perhaps current AI is as good as is gets for a long while.

You must be registered to see the links

And who knows what the companies and governments are cooking in secret even now.

 

[Ashira13]

the ai itself is not bad these days. the technology has been in development since the 1960's. the real problem is the hardware running it. it is still no powerful enough to do generate "new" content on the fly, but it is getting there, slowly but surely. the best way is probably dual "entity", one part local, and one part cloud-based computing. it is very interesting to see how the technology develops, and i hope to see it to fruition in my lifetime, but that is highly doubtful.

 

[DuniX]

the best way is probably dual "entity", one part local, and one part cloud-based computing. it is very interesting to see how the technology develops, and i hope to see it to fruition in my lifetime, but that is highly doubtful.

Best bet is to link the game with their own private subscriptions that they have to those AI sites.
So it is technically useable right now.