/games/ - Games

to start off the thread i am working on a mod for Limbus company. Gonna replace a cutscene in the game for Canto VII specifically for Don quixote vs Sancho(Don quixote) with characters from honkai star rail, specifically asat pramad and himeko.

i wasnt motivated to do it since it was a silly idea anda a lot of work but a recent edit on bilibili combining "hero" with some clips motivated me back to it.

I've been testing local LLMs and they're much faster than I thought they were. I've been thinking about making a game (or at least a tech demo project) that uses an LLM for NPCs, but with all the (understandable) hate that AI gets, this might be the worst possible environment to make such a game, I don't know…

>>47443
Do it. Worst case scenario it will be an interesting project to put on your CV.

> I've been thinking about making a game (or at least a tech demo project) that uses an LLM for NPCs
i've been thinking about this as well and i think the limiting factor is the user's specs. getting your LLM to output antyhing that isn't junk hallucinated blabber, you need substantial RAM and GPU cache so these things are able to hold and process all the context so that the output seems intelligent, or at least relevant. i think that's the reason why LLM mods for skyrim just didnt take off despite seeming super promising at first

>>47448
there is a new mod for behaviors instead of ANOTHER "chatbot but in a bideo game woooo"

>>47444
aight, bet!

>>47448
This is the thing. In the project I'm planning I don't want long complex and highly consistent text. I want to use a very light-weight LLM to provide very short but organic dialogue about the context they're in, such as the objects around them, what the player is doing and so forth. But maybe this requires more processing power than it seems. I'll find out… eventually. Oh yeah, >>47449 and using it for behavior would be rly cool

i remastered my M1912 originally meant to update my mod:

https://www.nexusmods.com/newvegas/mods/94026

but unfortunately the modder behind the framework i was using for the mod, has a lot of drama going on with some other chuds Xilandro and Hitman and they make compatibility with his weapon framework mod impossible for the stupidest reasons.

so i thought of porting it into PAYDAY 2, but it just got updated to the new 64bit engine

so we're gonna have to wait to see this anywhere because i'm also going through like trans menopause and i literally feel like i'm dying so.

>>47465
>I want to use a very light-weight LLM to provide very short but organic dialogue about the context they're in, such as the objects around them, what the player is doing and so forth.
maybe you could combine normal-ass decision trees with LLMs, i think a good approach to experiment with would be to procedurally assemble prompts, and then pass them to a tiny LLM and show the result to the player. it's basically dialog trees with extra steps, but maybe there's something cool there i dunno.

btw you wouldn't be the first to bundle an AI gen model, inzoi has stable diffusion built in to make custom assets. the results are horrible shit but it's there. dummies are trying to cancel inzoi for this for some reason

>>47471
>for some reason
>the results are horrible shit
sounds like a good reason

>>47473
ok i said "for some reason" but actually they are not mad at the lousy output but because they thought the stable diffusion model was an inhouse inzoi model and not, uh, stable diffusion, and they thought that was "more ethical". it is actually just a useless feature.