6 ms·

Unrelated but are there any games using LLMs or ML for the NPCs? I really want enemies that adapt to the player.

hackable_sand 15d ago
Related, but we've had multiplayer for decades now.
joegibbs 15d ago
I did it for story events in my grand strategy game (Fall of an Empire) with a set of finetuned local models. It doesn't affect performance (for most people) since the game is bottlenecked on other things. Players seem to like it, there are some issues since the models are so small they are more prone to hallucination, and you also have to give them preset topics or they go on the same track again and again. I haven't been able to use it for more than events since the tokens per second aren't high enough - I would like to use the models to e.g. generate actions for characters based on their history of interactions with other characters, but that ends up taking so much time that the player never gets any events. I also did a demo of an LLM-powered NPC, I think the issue here is that you need more than just the LLM in e.g. a 3D RPG where you can actually go up and talk to the character. You can generate the text, but from there you also need to generate speech, from that you need to dynamically generate realistic facial and body animations that match up with it - this means you need 3 or 4 models chained together in a pipeline.
AshamedCaptain 15d ago
Right until you release it and a minor gets the NPC to talk about rape or something.
daemin 15d ago
I don't know of any that use LLMs and my educated opinion would be that there won't be any in the next decade or ever. The main reasons are that games already run with tight margins of both frame time and memory consumption, and that designers/game developers want to control the interactions with the player and what the player experiences in the game. So an LLM won't be used for any story relevant NPCs in the game as what they say is tightly curated, and it's never going to be good/fast/slim enough to get used for crowd NPCs. Using machine learning to make enemies smarter has been thought about and talked about for decades, but I don't see it being used either for similar reasons. Training a ML algorithm takes resources which are in short supply and time which could take longer than for the player to finish the game (if you want smarter enemies play against real human players). There's also the fact that the game has to run on the minimum specification computer, which means different players will have differently skilled enemies and their experience of the game will suffer. There have been some games in the past which used some form of machine learning as a core gameplay feature, but they have been pet breeding simulators.
- drugstorecowboy 15d ago
  In my completely unqualified and baseless opinion, I think the reason machine learning in games isn't more popular, is that ruthlessly difficult enemies that learn and adapt to the player just aren't.. fun. Level playing fields between characters and enemies just don't work. For example, the Dark Souls series is generally known for being "difficult but fair", however enemies are not smart by any means, they ultimately have a very limited number of attacks and very basic AI that consists of "Wait until player is in range, Run in a straight line toward the player until in attack range, attack". Even with that simplistic AI its pretty difficult and enjoyable.The player is meant to triumph over hundreds of enemies, its the player that learns and adapts to a static set of obstacles. If enemies in Dark Souls learned and adapted to the player, it would go from difficult to impossible very quickly and would not be enjoyable. The purpose of AI in games isn't making something thats impossible to beat, its making something that is ALWAYS possible to beat while remaining fun and engaging.
  - luke5441 15d ago
    There are enough people playing multiplayer games where this is not the case. Those are actually kind of annoying because ELO systems cause you to always loose about half the games (except if you are the top player of all the players) no matter how much better you get. You could tune this to handicap the LLM such that the player wins 80% of the time or so and it should be more fun (except for the competitive aspect).
    - drugstorecowboy 14d ago
      Thats a good point about multiplayer, I was really thinking single player only. I guess the main thrust of my point is that the reason there isn't more ML in games isn't related to performance, its that the goal of game design(single player anyway) isn't to make enemies into unbeatable gods so the usefulness of ML is somewhat limited.
addandsubtract 15d ago
I've only seen proof-of-concepts, like Vaudeville[1]. I think more work is currently done in the modding scene, for example with Mantella[2] which, "brings every NPC in Skyrim and Fallout 4 to life with AI." [1] https://store.steampowered.com/app/2240920/Vaudeville/ https://store.steampowered.com/app/2240920/Vaudeville/ [2] https://art-from-the-machine.github.io/Mantella/ https://art-from-the-machine.github.io/Mantella/
polytely 15d ago
I think there is a skyrim mod that lets hook up an LLM to the NPCs so you can talk to them, the problem is that the rest of the game design doesn't really support that. So the novelty wears off quite quick. There is probably some limited utility for using ML in games, lifelike multiplayer bots trained on behaviour of different types of players. I think in most scenarios good gameplay comes from iteration which comes from tweak-ability, you see players get stuck in area x so you do a minor or major adjustment to area layout, enemy composition, or give the players more tools. If something is controlled by an LLM it becomes unpredictable and hard to reason about for both player and designer, and consequently also hard to adjust. I think it is possible there is some role for LLM's in games, like how Left 4 Dead had an AI (this is classic AI not ML) director who would decide how to structure enemy waves and for example send a boss after a player who strays too far from the rest. But even there rules based AI is probably easier to adjust for the designer.
nitwit005 15d ago
Excepting strategy games, the industry's experience has been that most players won't notice clever AI. There have been a number of games that have tried AI conversations, but you're basically trying to sell a Chat GPT prompt, and many players are aware they don't have to pay you to get a similar result on their own.

Related, but we've had multiplayer for decades now.

I did it for story events in my grand strategy game (Fall of an Empire) with a set of finetuned local models. It doesn't affect performance (for most people) since the game is bottlenecked on other things. Players seem to like it, there are some issues since the models are so small they are more prone to hallucination, and you also have to give them preset topics or they go on the same track again and again. I haven't been able to use it for more than events since the tokens per second aren't high enough - I would like to use the models to e.g. generate actions for characters based on their history of interactions with other characters, but that ends up taking so much time that the player never gets any events. I also did a demo of an LLM-powered NPC, I think the issue here is that you need more than just the LLM in e.g. a 3D RPG where you can actually go up and talk to the character. You can generate the text, but from there you also need to generate speech, from that you need to dynamically generate realistic facial and body animations that match up with it - this means you need 3 or 4 models chained together in a pipeline.

AshamedCaptain 15d ago

Right until you release it and a minor gets the NPC to talk about rape or something.

daemin 15d ago

I don't know of any that use LLMs and my educated opinion would be that there won't be any in the next decade or ever. The main reasons are that games already run with tight margins of both frame time and memory consumption, and that designers/game developers want to control the interactions with the player and what the player experiences in the game. So an LLM won't be used for any story relevant NPCs in the game as what they say is tightly curated, and it's never going to be good/fast/slim enough to get used for crowd NPCs. Using machine learning to make enemies smarter has been thought about and talked about for decades, but I don't see it being used either for similar reasons. Training a ML algorithm takes resources which are in short supply and time which could take longer than for the player to finish the game (if you want smarter enemies play against real human players). There's also the fact that the game has to run on the minimum specification computer, which means different players will have differently skilled enemies and their experience of the game will suffer. There have been some games in the past which used some form of machine learning as a core gameplay feature, but they have been pet breeding simulators.

drugstorecowboy 15d ago
In my completely unqualified and baseless opinion, I think the reason machine learning in games isn't more popular, is that ruthlessly difficult enemies that learn and adapt to the player just aren't.. fun. Level playing fields between characters and enemies just don't work. For example, the Dark Souls series is generally known for being "difficult but fair", however enemies are not smart by any means, they ultimately have a very limited number of attacks and very basic AI that consists of "Wait until player is in range, Run in a straight line toward the player until in attack range, attack". Even with that simplistic AI its pretty difficult and enjoyable.The player is meant to triumph over hundreds of enemies, its the player that learns and adapts to a static set of obstacles. If enemies in Dark Souls learned and adapted to the player, it would go from difficult to impossible very quickly and would not be enjoyable. The purpose of AI in games isn't making something thats impossible to beat, its making something that is ALWAYS possible to beat while remaining fun and engaging.
- luke5441 15d ago
  There are enough people playing multiplayer games where this is not the case. Those are actually kind of annoying because ELO systems cause you to always loose about half the games (except if you are the top player of all the players) no matter how much better you get. You could tune this to handicap the LLM such that the player wins 80% of the time or so and it should be more fun (except for the competitive aspect).
  - drugstorecowboy 14d ago
    Thats a good point about multiplayer, I was really thinking single player only. I guess the main thrust of my point is that the reason there isn't more ML in games isn't related to performance, its that the goal of game design(single player anyway) isn't to make enemies into unbeatable gods so the usefulness of ML is somewhat limited.

addandsubtract 15d ago

I've only seen proof-of-concepts, like Vaudeville[1]. I think more work is currently done in the modding scene, for example with Mantella[2] which, "brings every NPC in Skyrim and Fallout 4 to life with AI." [1] https://store.steampowered.com/app/2240920/Vaudeville/ https://store.steampowered.com/app/2240920/Vaudeville/ [2] https://art-from-the-machine.github.io/Mantella/ https://art-from-the-machine.github.io/Mantella/

polytely 15d ago

I think there is a skyrim mod that lets hook up an LLM to the NPCs so you can talk to them, the problem is that the rest of the game design doesn't really support that. So the novelty wears off quite quick. There is probably some limited utility for using ML in games, lifelike multiplayer bots trained on behaviour of different types of players. I think in most scenarios good gameplay comes from iteration which comes from tweak-ability, you see players get stuck in area x so you do a minor or major adjustment to area layout, enemy composition, or give the players more tools. If something is controlled by an LLM it becomes unpredictable and hard to reason about for both player and designer, and consequently also hard to adjust. I think it is possible there is some role for LLM's in games, like how Left 4 Dead had an AI (this is classic AI not ML) director who would decide how to structure enemy waves and for example send a boss after a player who strays too far from the rest. But even there rules based AI is probably easier to adjust for the designer.

nitwit005 15d ago

Excepting strategy games, the industry's experience has been that most players won't notice clever AI. There have been a number of games that have tried AI conversations, but you're basically trying to sell a Chat GPT prompt, and many players are aware they don't have to pay you to get a similar result on their own.