Hacker News Reader

Comment by OtherShrezzing | Hacker News Reader

I’ve been thinking about how traditional game AI can be improved by generative models. One of the biggest problems with games like Civ is that the AI strategy is predictable - especially if you’ve played a few dozen hours.

LLMs with some decent harnesses could build up unpredictable - but internally consistent - strategies per each new game you play.

This is close to a proof of concept for those improvements.

View on HN

I wonder how could you keep the LLM from going bonkers as the game progresses? I have a feeling it's possibly better to re-create the prompts after some time, and have the LLM work more like one of those "reasoning models" with the game as something it can interact with.

Otherwise you run into the risk of "TOTAL NUCLEAR FINANCIAL LEGAL DESTRUCTION" ;)

This is something we've been working on and are planning to release a "decision" update to the game which should allow for multi-step, configurable options to choose if the LLM actually gets to contribute to the current world / chat. There's a lot of trial and error involved and we're all ears, if you have ideas we'd love to hear them! We actively monitor our discord https://discord.com/invite/theinterface

Probably still performs more rationally than the lategame AI in Civ.

Absolutely! Max and I were huge Civ fans and always tried to make the game AI deviate from its programmed strategy. We also believe you can get some really interesting story arcs by adjusting parameters like temperature and how context is presented. Some of the things you'll notice in the game is we have a no-holds barred approach -- you can fully modify system prompts and adjust how the LLM interprets the state of the world.

As someone that plays those games pretty heavily: I’d rather not have LLMs take over game AI like that. If I want different gameplay I’d play online. We don’t need to bog down already heavy games with LLMs.

Odd take, the stale dialogue and static quests of most rpgs could certainly benefit from llm enhancements

Nah, they’ll still be stale. Many people play RPGs that haven’t changed in 30 years, so static isn’t an issue either.

And many people don’t, there are already skyrim mods for this so your point doesnt really hold water.

However, no one has ever praised the Elder Scrolls storyline.

They win by the sheer quantity and by giving you a lot of subsystems to play with.

So LLM generated quest text probably feels it belongs here. It wouldn't, for example, in something with the Witcher 3 story quality.

except the hundreds of hours of youtube videos discussing the story of the franchise

It’s so funny to reference a game that has like 12 editions and is on every platform including a refrigerator and think “this game is missing something”

By the way there are LLM dialog mods for Skyrim and everyone thinks they’re a joke because they suck.

you seem to enjoy speaking in absolutes, that's a really bad habit

You can't even put your name on the table, brother.

I think the staleness comes from the fact that it’s the 60 billionth time you’ve done some “quest” to go gather some crap up or kill the same thing in a loop for an hour.

No amount of dialogue is going to save that.

The actual story dialogue is usually interesting enough already

Quality writing is what is most important in an RPG, something that LLMs are distinctly terrible at.

Same take here - these games are addictive due to certain repetitive predictable patterns. I expect more and less complex automatons populating a game world and emergent story resulting from my flawed meat brain inputs.

Another thought that follows is that any kind of generative behavior, not just LLM, runs this risk of an endless pointless blandness. I.e. like with any artform we want there to be a point.

If those games are to feature LLM AI it would have to stand on it's own, with someone like these guys having thought it through.

I can't help it, the first thought that came to mind was "Huh...talk about sheer senseless brute force." Why use a Large Language Model on something as clearly defined in scope as a game instead of a model designed and trained for the task/ruleset? Sure, there's the argument of not having to train that model, but OTOH, "decent harnesses" does some very heavy lifting there...

I think it's a compelling argument. You would need a large dataset of completed games on which to train, which may have something to do with why the games considered solved by AI are also among those where exist a very rich and heavily annotated corpus of completed games in algebraic notation.

Of course - but in practice you won't be aiming towards fully a "solved" game or that kind of player skill for something like Civ - and even so, I severely doubt an LLM realistically can hope to even get in the vicinity unless the aforementioned "harness" does something similar anywayas part of its heavy lifting I mentioned.

Definitely the case. That being said, I think it would be hard, at least in the immediate future, to translate the concept of difficulty to a universal LLM for a bespoke/specific game. I assume most game AIs are tuned by hand to feel fair for a given difficulty level...but if you just give an LLM some new game, explain the rules and what resources/abilities it has available to it, you're stuck with adding some addendum to the tune of "and you're meant to represent an entity of 'medium' difficulty." For very well established games, it might have a sense of how given actions might fall into a skill-level hierarchy, but not for anything new.

Fine tuned LLMs though with actual experience with the game, maybe?

You can just have the old AI doing stuff like resource management while the LLM handles the diplomacy and it wouldy be a lot better already.

From a player perspective, oftentimes the best AI systems are the most trivial ones. You can get really far with an agent that is allowed to cheat. It's a hell of a lot easier to build and troubleshoot a model that manipulates the amount of in-game resources received per unit time than it is to implement actual strategic intelligence.

I play strategy games a lot and cheating AI can be fun to play against at first, but the more you learn a game the more cheating AI sucks. When you're new to the game it just feels like you're playing against a good player, but you soon learn that what they are achieving isn't possible with the resources available. Once you hit that realisation it can be fun to beat them as a challenge but it never feels like a fair game.

Agreed, this is an instant turn-off for me when I realize this in e.g. an RTS game. Red Alert or C&C come to mind on higher difficulty, can't remember which.

IIRC the RA1 skirmish mode AIs always had perfect information and resource multipliers based on difficulty. RA2 did it a little differently with "virtual ore purifiers" added for the high difficulty AIs. I'm sure a similar thing was done for the Tiberian Dawn campaign and the Tiberian Sun multiplayer/skirmish AIs.

OpenRA's bots are a bit more clever, and also don't need to magically see into fog-of-war.

I never played much of RA2, but played hundreds of hours of RA1 skirmish. Must have been that. Thanks for the insight!

Skirmish was a blast- I'd turtle until I had the enormous battleships (cruisers?) that could fire onto land. Loads of fun when I was like 12.

Civilization uses a similar technique, and it’s the reason I’ve been thinking about the potential here.

The AI on higher difficulty starts a few centuries more technologically advanced than you, and gets multipliers on the starting resources like cities.

It’s not particularly fun to compete against.

Cheating AI turns every game into a puzzle game. The game turns into figuring out what the weaknesses of the AI are and taking advantage of them at every step. That is the only way you can compete against the massive advantages cheating gives.

Typically there are some easy micro and macro tricks that make the AI do something very stupid. That's why kiting is so ubiquitous in games - the AI just keeps following you while you whittle it down. Doesn't really work against a real player if they're microing the units.

even being able to scheme with / backstab leaders, and they would "understand" all that's happened (and acts accordingly) would be so fun

I'd love to see LLM based versions procedural tasks like "radiant quests" which are generally disappointing, though I've heard it discussed before and the real challenge is keeping it from going way off the rails.

The other challenge I think you'll run into in general is that there's a huge knee jerk reaction against any use of LLMs or other popular types of gen AI in games in places like Reddit or Bluesky.

Reply on news.ycombinator.com