> The rocket launch was only achieved through significant ‘manual’ effort on Sol’s part as it papered over its late-game failures to connect certain production systems by repeatedly ferrying supplies between areas of the factory.
Hah I dunno, I think most human players wouldn't find that very fun and would instead take that as an opportunity to refactor the factory for optimized performance (i.e. the core gameplay loop)
When you start a new game in Factorio, if you never played it, you can’t know how much of each resource you’ll have to extract and neither how much of an item you’ll need to produce.
I know it’s easy to forgot after thousands hours of playtime, but the game starts with your character having to manually mine coal one by one.
I hope to experiment with using classic tabular reinforcement learning in game AI one day. Imagine a framework that makes setting up small Q-learning tables easy, and you can just sprinkle dozens of Q-learning algorithms throughout your traditional AI scripting.
These old basic algorithms often have mathematical proofs that they will converge to optimal behavior, and they do so fairly quickly. Turning the complex world of the game into a Q-table would require some creativity and judgement, but is not that hard I think.
In the end we would have a controllable AI that actually does adapt. That's my hope anyway.
There has to be a AAAA studio building the new, greatest MMO, that also utilizes LLMs as part of the game loop, surely?
(Not sure how prohibitive it would be to do anything interesting with local LLMs of any kind in a big game, given how many platforms/setups you have to make the stuff run smoothly on, and the sheer amount of RAM they would bind)
https://proxiai.com/ Will Wright (The Sims, etc) is dabbling with AI in his latest game. No idea if it'll be any good, but there's def companies out there trying.
I have been playing with this. Having tiny models provide more contextual information when an NPC speaks is cool but the "small enough" models tend to go off the rails.
It's fun if you have a handful of people playing, but it would quickly go towards negative reviews if you tried to sell it.
I don't think an LLM should ever be having a two-way conversation with the player. It would be more suited to 2 npcs having conversations amongst themselves. All dialogue delivered to the player should be human written.
At most it should be mildly tweaking existing dialogue, e.g updating a line to reflect game events. But as soon as you present a prompt window to the player the entire illusion that you are in a game world instantly dies.
i think it can work for procedurally generated situations, and the llm is acting more like a translator between data model and player rather than being a data storage layer.
thats close to tweaking existing dialog, but with a bit more freedom.
you still dont have to give the player arbitrary response
Most dialogue in games is crap. I would much rather have an LLM writing the dialogue so that it can include references to player choices over what currently exists.
With the aid of SOTA LLMs, you can encode specialist intelligence into computer programs without going through the roundabout route of training small LLMs. And even when you do go the route of ML, it doesn't necessarily have to be an LLM or even a Neural network.
i think hooking in an llm to dwarf fortress for turning the very mechanistic descriptions into prose. same with the adventure mode talk. instead of a constant "well weve got beasts" an llm could read a bunch of the language tree to make a unique introduction for each.
same with the df artworks. get kitfox to design some art styles for dwarves, elves, goblins, and humans, and then instead of just reading
"this is a fine engraving of a dwarf. it concerns the events of 102 when ulrist fastbuttons was elected mayor. ulrist fastbuttons is sobbing"
impossible, unfortunately, because LLMs suck at creative writing. without handholding, they are incapable of producing anything but cliche-ridden purple prose for young adults. you can feed it the first half of a book with lots of sex, violence, and profanity, ask it to continue the story, and it will degrade into safe slop within two chapters.
both American and Chinese labs aim their models at coding and knowledge work, so they penalize adult themes in post-training.
with better being more convincing, rather than just stronger
its the same for like a chess bot. i dont want a perfect stockfish that occasionally makes a serious blunder when a dice rolls a certain way. i want one that actually falls into traps
You could play against an AI right now on the Brood War ladder right now. Granted, you have to make it to S rank first because that's where the bots end up.
The biggest thing the bugs introduce from an LLMs perspective is a necessity for realtime response. Without them, the LLM can take as long as it wants to decide and do its next operation on the factory. With them, not responding fast enough results in death (particularly early game where a lot of biter combat is player driven). I presume it doesn't do great at this.
That was exactly my thought when reading the headline, before even looking into the article. Sure it's still somewhat complex to get to a rocket launch, but without the biters where is the challenge, other than complexity? It's like a big excel sheet with conveyor belts, pipes and electricity poles at that point only.
The more interesting challenge is the one you eliminated, complexity, in getting efficient logistics and arrangements. Dyson Sphere Program (until late in its existence) and Satisfactory don't have waves of enemies attacking your bases but remain challenging and a satisfying puzzle. I remember I used a constraint solver to help me optimize some late game recipes for Satisfactory.
The biters are just a line item in the spreadsheet. Early players might build a wall with an ammo belt, for example. There's lots of options that don't involve responding in real-time.
Though these days I just manually clear the pollution cloud with a pocket full of turrets. The forward deployed engineer model, if you will.
Patience, for one. I reach a point where I'm looking at a mess of coal-fed sorters that need a complete rip-and-replace with electric ones, and that's usually the point where I close the game and go back to something easier like Kenshi.
I know it's a pipe dream, but these kinds of games (Factorio, Satisfactory, etc) would all benefit from sort of brain-game interface where I could just think about what I wanted to do.
The tedium and repetition is really what kills the enjoyment for me. The problem solving, planning, forethought, etc are super fun but it's killed by the former.
Factorio rewards the players progression through the game by taking away points of pain. The helper bots being a key transition. Without that pain there would be no reward. The fact that refactoring is expensive in time and complexity is what makes it realistic - it makes the earlier decisions important and consequential.
You can always enter creative mode and make whatever changes you want in the game instantly, but in my experience that does make it less fun.
It's basically programming with really bad ergonomics. I'm a programmer by profession and I was immediately frustrated by Factorio because I realized I'd rather just write code and build a project because it would scratch the exact same itch while actually being useful.
I tried really hard to like Factorio, and I bounce off of it each time after playing a save for 4-8 hours for exactly this reason. I know a lot of programmers who love it, so I almost wish I could get into it.
Now I'm tempted to try again, because it feels like the kind of game I should really enjoy.
It doesn't seem to understand the core mechanic of Factorio, which is putting outputs onto belts and routing them to other inputs. It does this in a few places, but for the most part everything is put into chests and moved around by the player.
It's just easier to do manual stuff when the complexity of scaling the automation becomes cognitively demanding. A new player does mostly everything manually while experienced players have a natural intuition for planning and scaling the automation.
If you don't have enemies, you have unlimited time. Doing it thru player is what it learned and wasn't pressured to do better (it has no fatigue as opposed to humans).
It is weird and almost alien how AI designs factories. When Claude announced Fable 5 they also did a Factorio demo that was the same kind of designs and a lot of hand feeding and direct insertion.
In some sense it proves to me that llms are not just copying and are actually creating a bespoke solution. But I honestly expected more from frontier models
To them the player character is just as good as a belt for transporting materials.
The player character actually does have an extremely high capacity so if you are constantly running between chests your factory will still be reasonably high throughput.
> In some sense it proves to me that llms are not just copying and are actually creating a bespoke solution.
Is the weirdness of the designs meaningfully beneficial in any way that the AI has figured out, or could it for example be that the training data the LLMs learned from are too different from the type of data it would need to learn from in order to interface properly and play the game better? And then because of that incompleteness, the LLM has to hallucinate a lot and it results in weird outcomes.
I think one of the problems is that AI has no sense of laziness or time. Unless it's in their training data or prompted they're not going to bother automating something
This is a much better benchmark as a concept (without knowing the details of how the problem is framed to the agent) than some of these ridiculous things the frontier labs are reporting (vending machine simulator? come on).
Can you report more in depth at stumbling blocks? It's a surprisingly linear curve.
I'm going to try it on a local run.
I wonder if there is room for "recommendorio" where it gives you AI advice on your current factory.
Is the model pre-trained with any knowledge of the game?
> The rocket launch was only achieved through significant ‘manual’ effort on Sol’s part as it papered over its late-game failures to connect certain production systems by repeatedly ferrying supplies between areas of the factory.
Robots: they're just like us!
> Robots: they're just like us!
Hah I dunno, I think most human players wouldn't find that very fun and would instead take that as an opportunity to refactor the factory for optimized performance (i.e. the core gameplay loop)
i think its more of an effect of factorio's hidden information.
youre not really sure how to set up some of the logistics/how much of what you will need where ahead of time
What information is hidden in Factorio? I can't think of any stat that isn't shown to the player in some way.
He is not talking about stats.
When you start a new game in Factorio, if you never played it, you can’t know how much of each resource you’ll have to extract and neither how much of an item you’ll need to produce.
I know it’s easy to forgot after thousands hours of playtime, but the game starts with your character having to manually mine coal one by one.
How do you mean? (I have played 1 hour of Factorio)
I didnt think was possible to play for less than 1k hours.
What do you mean? Factorio is an extremely calculable game and tends to show most of its information (ratios aside, those are in the wikis)
Unless I misunderstand, ratios are second-order and can be calculated; production rates are presented in-game.
Watching AI play games is fun, but I can't wait for someone to crack having small LLMs in games to enrich them in useful and interesting ways.
I hope to experiment with using classic tabular reinforcement learning in game AI one day. Imagine a framework that makes setting up small Q-learning tables easy, and you can just sprinkle dozens of Q-learning algorithms throughout your traditional AI scripting.
These old basic algorithms often have mathematical proofs that they will converge to optimal behavior, and they do so fairly quickly. Turning the complex world of the game into a Q-table would require some creativity and judgement, but is not that hard I think.
In the end we would have a controllable AI that actually does adapt. That's my hope anyway.
There has to be a AAAA studio building the new, greatest MMO, that also utilizes LLMs as part of the game loop, surely?
(Not sure how prohibitive it would be to do anything interesting with local LLMs of any kind in a big game, given how many platforms/setups you have to make the stuff run smoothly on, and the sheer amount of RAM they would bind)
https://proxiai.com/ Will Wright (The Sims, etc) is dabbling with AI in his latest game. No idea if it'll be any good, but there's def companies out there trying.
In "Where Winds Meet" you can talk to NPCs
I have been playing with this. Having tiny models provide more contextual information when an NPC speaks is cool but the "small enough" models tend to go off the rails.
It's fun if you have a handful of people playing, but it would quickly go towards negative reviews if you tried to sell it.
I don't think an LLM should ever be having a two-way conversation with the player. It would be more suited to 2 npcs having conversations amongst themselves. All dialogue delivered to the player should be human written.
At most it should be mildly tweaking existing dialogue, e.g updating a line to reflect game events. But as soon as you present a prompt window to the player the entire illusion that you are in a game world instantly dies.
i think it can work for procedurally generated situations, and the llm is acting more like a translator between data model and player rather than being a data storage layer.
thats close to tweaking existing dialog, but with a bit more freedom.
you still dont have to give the player arbitrary response
Most dialogue in games is crap. I would much rather have an LLM writing the dialogue so that it can include references to player choices over what currently exists.
With the aid of SOTA LLMs, you can encode specialist intelligence into computer programs without going through the roundabout route of training small LLMs. And even when you do go the route of ML, it doesn't necessarily have to be an LLM or even a Neural network.
i think hooking in an llm to dwarf fortress for turning the very mechanistic descriptions into prose. same with the adventure mode talk. instead of a constant "well weve got beasts" an llm could read a bunch of the language tree to make a unique introduction for each.
same with the df artworks. get kitfox to design some art styles for dwarves, elves, goblins, and humans, and then instead of just reading
"this is a fine engraving of a dwarf. it concerns the events of 102 when ulrist fastbuttons was elected mayor. ulrist fastbuttons is sobbing"
SOTA LLMs would probably be fun opponents in turn based 4x.
Will be interesting to find out what they can bring over simply having better SOTA LLM coded opponents (I am sure, we will).
impossible, unfortunately, because LLMs suck at creative writing. without handholding, they are incapable of producing anything but cliche-ridden purple prose for young adults. you can feed it the first half of a book with lots of sex, violence, and profanity, ask it to continue the story, and it will degrade into safe slop within two chapters.
both American and Chinese labs aim their models at coding and knowledge work, so they penalize adult themes in post-training.
I would love the revival of the RTS genre with better computer enemies.
with better being more convincing, rather than just stronger
its the same for like a chess bot. i dont want a perfect stockfish that occasionally makes a serious blunder when a dice rolls a certain way. i want one that actually falls into traps
You could play against an AI right now on the Brood War ladder right now. Granted, you have to make it to S rank first because that's where the bots end up.
Such AIs have existed in Brood War and Dota 2 for a very long time now (pre-LLM), problem is they aren't very fun for humans to play against.
bah, such a pity we didn't have those in wc3 times. bots sucked badly.
I could play online, but my dial-up costed $1 per hour for 250ms latency and near guarantee of at least one disconnect per hour.
There are already mods for games like Grand Theft Auto V that make NPCs talk back and follow your verbal commands: https://www.youtube.com/watch?v=3_Eng-F4akI&t=106s
A little unpolished because it was bolted on by fans and not the original game developers, but otherwise very impressive.
SS13 but the ship's AI is actually an AI instead of a player in disguise
Factorio is a much easier game without the bugs being a pain in the arse as you expand - i wonder how Sol would handle the bugs
The biggest thing the bugs introduce from an LLMs perspective is a necessity for realtime response. Without them, the LLM can take as long as it wants to decide and do its next operation on the factory. With them, not responding fast enough results in death (particularly early game where a lot of biter combat is player driven). I presume it doesn't do great at this.
you could pause the game between actions, while the model reasons about what to do next.
Glad to hear we've got LLMs controlling military drones in real life :(
That was exactly my thought when reading the headline, before even looking into the article. Sure it's still somewhat complex to get to a rocket launch, but without the biters where is the challenge, other than complexity? It's like a big excel sheet with conveyor belts, pipes and electricity poles at that point only.
The more interesting challenge is the one you eliminated, complexity, in getting efficient logistics and arrangements. Dyson Sphere Program (until late in its existence) and Satisfactory don't have waves of enemies attacking your bases but remain challenging and a satisfying puzzle. I remember I used a constraint solver to help me optimize some late game recipes for Satisfactory.
The biters are just a line item in the spreadsheet. Early players might build a wall with an ammo belt, for example. There's lots of options that don't involve responding in real-time.
Though these days I just manually clear the pollution cloud with a pocket full of turrets. The forward deployed engineer model, if you will.
> where is the challenge, other than complexity
Patience, for one. I reach a point where I'm looking at a mess of coal-fed sorters that need a complete rip-and-replace with electric ones, and that's usually the point where I close the game and go back to something easier like Kenshi.
I know it's a pipe dream, but these kinds of games (Factorio, Satisfactory, etc) would all benefit from sort of brain-game interface where I could just think about what I wanted to do.
The tedium and repetition is really what kills the enjoyment for me. The problem solving, planning, forethought, etc are super fun but it's killed by the former.
Factorio rewards the players progression through the game by taking away points of pain. The helper bots being a key transition. Without that pain there would be no reward. The fact that refactoring is expensive in time and complexity is what makes it realistic - it makes the earlier decisions important and consequential.
You can always enter creative mode and make whatever changes you want in the game instantly, but in my experience that does make it less fun.
It's basically programming with really bad ergonomics. I'm a programmer by profession and I was immediately frustrated by Factorio because I realized I'd rather just write code and build a project because it would scratch the exact same itch while actually being useful.
I tried really hard to like Factorio, and I bounce off of it each time after playing a save for 4-8 hours for exactly this reason. I know a lot of programmers who love it, so I almost wish I could get into it.
Now I'm tempted to try again, because it feels like the kind of game I should really enjoy.
Its puzzle solving. For a lot of people, that _is_ the fun part.
It doesn't seem to understand the core mechanic of Factorio, which is putting outputs onto belts and routing them to other inputs. It does this in a few places, but for the most part everything is put into chests and moved around by the player.
What if the player's inventory has a higher throughput than belts?
Part of the reason players try to automate manual feeding is the annoyance factor, which AI doesn't care about.
It might have higher throughput than a single belt over short distances, but not in general. Have you seen what a Factorio speedrun looks like?
It's just easier to do manual stuff when the complexity of scaling the automation becomes cognitively demanding. A new player does mostly everything manually while experienced players have a natural intuition for planning and scaling the automation.
If you don't have enemies, you have unlimited time. Doing it thru player is what it learned and wasn't pressured to do better (it has no fatigue as opposed to humans).
Coding agents are bad at architecture, Exhibit no. 12057.
To be fair, I do do the same.
It is weird and almost alien how AI designs factories. When Claude announced Fable 5 they also did a Factorio demo that was the same kind of designs and a lot of hand feeding and direct insertion. In some sense it proves to me that llms are not just copying and are actually creating a bespoke solution. But I honestly expected more from frontier models
To them the player character is just as good as a belt for transporting materials.
The player character actually does have an extremely high capacity so if you are constantly running between chests your factory will still be reasonably high throughput.
> In some sense it proves to me that llms are not just copying and are actually creating a bespoke solution.
Is the weirdness of the designs meaningfully beneficial in any way that the AI has figured out, or could it for example be that the training data the LLMs learned from are too different from the type of data it would need to learn from in order to interface properly and play the game better? And then because of that incompleteness, the LLM has to hallucinate a lot and it results in weird outcomes.
I think one of the problems is that AI has no sense of laziness or time. Unless it's in their training data or prompted they're not going to bother automating something
This is where human-directed looping revision fits in on a lot of my projects.
I tried a few recently on Civ 6 Deity, and they all lose pretty quickly.
Sounds like an experiment I'd like to give a try--what integration layer did you use to get LLMs to control the civ?
To be fair, Deity on Civ games is brutal. I'd be interested in where the cutoff is.
Doesn't the AI (the computer players, I mean) "cheat" at that high of difficulty level?
This is a much better benchmark as a concept (without knowing the details of how the problem is framed to the agent) than some of these ridiculous things the frontier labs are reporting (vending machine simulator? come on).
Can you report more in depth at stumbling blocks? It's a surprisingly linear curve.
I'm going to try it on a local run.
I wonder if there is room for "recommendorio" where it gives you AI advice on your current factory.
Is the model pre-trained with any knowledge of the game?
how much did it cost I wonder
interesting artifact of over investment and subsidies of AI and datacenters
Who determines "over investment"
the market, and investors opinions on their returns.
Another thing AI can do that I can't...
Good to know it's good at something, I suppose? Sol seems like it always just wastes my time and writes terrible code.