The Worlds First AI Labs Simulator RTS Game - Kimi K3.1 Pro
Hello. Hope you are doing well. So, there's a lot of things happening in the gaming space and I wanted to kind of jump on the hype, too. So, today I'm going to show you kind of the game I've been playing around with using Kim K 3.1 Pro. Uh, this is AI Labs Sim. So, let me just…
- 01.Hope you are doing well
- 02.So, there's a lot of things happening in the gaming space and I wanted to kind of jump on the hype, too
- 03.So, today I'm going to show you kind of the game I've been playing around with using Kim K 3.1 Pro

Hello. Hope you are doing well. So, there's a lot of things happening in the gaming space and I wanted to kind of jump on the hype, too. So, today I'm going to show you kind of the game I've been playing around with using Kim K 3.1 Pro. Uh, this is AI Labs Sim. So, let me just play the trailer and then we talk a bit more about it and I'm going to show you everything I have done so far to yeah, build this game. So, my idea for a game, you run your own AI lab. You build out data centers, optimize compute, do research, train models, and make money while another Frontier Lab races you. And I wanted the real code feel. So, there's a real coding agent right inside the game. Unit Zero is your own clawed code. It builds, scouts, hires, and ships code. You start in a garage. You build power, server halls, and racks. And every new era of compute opens a slot for one more AI agent. Compute you're not using gets rented out by the minute. That's your income, minus the power bill, and the price moves all day. Each agent can run a department, power, compute, research, or models. You decide how much it can do from adviser to full delegate. And nobody tells them how. They try things, watch their score, and write what worked in a playbook. A bad move costs points. A good one sticks. The agents write real code, too. Python optimizers in a git repo, benchmarked by the game on hidden tests. Better code, more money. Then come the hard choices. Every bit of compute on research is rent you don't earn. Solar dies at night. And somewhere in the fog, Vanta Labs is racing you. The goal, grow from a garage to 10 gigawatts and finish the frontier run before Vanta does. That's AI labs sim. So the the model we are using today is uh the Kimik 3.1 Pro. I think they released this yesterday, not yesterday. Yes, last week uh from Moonshot AI. And basically I'm just using this via Open Code. We are their API. Uh that works pretty good. And uh I I like um the Kimmy models actually for building out these games. You can see some examples here, but they are really good at kind of this actually. I don't know if they optimized for this, but uh I kind of like their way of doing this. Uh I also think Opus is really good at this, but uh I kind of prefer the pricing for Kim True 3.1 Pro when I'm using Open Code. So let's have a quick look at our setup now. Like I said, we are on open code with this model and basically this is the prompt kind of I started with. So I had a bit of help shaping up this prompt using 3.1. But uh I kind of did like a voice input where I told it everything I want in the game. So you can see build AI Sims Lab isometric real-time strategy game that's kind of inspired by Factorio. Uh I don't know if I've mentioned that but we have some goals. So, we're going to reach 10 GW and finish the Frontier run. Then we will beat our rival lab. If we can do that before that model, right? And there's, of course, a bunch of like upgrades. So, we wanted to do a bit of a mix, but focus mostly on C++. So, we can do it on our Mac. So, we have this 20 version plus this SDL3. Yeah, we have some security testing. The core gameplay loop is we're going to build, we can build out racks, compute, money. Here we have different eras. As we scale, we get more agent slots and yeah, so we have some decision making. I'm not going to go through all of this, but I'm going to scroll through it if you kind of want to read it. But basically, this prompt was kind of designed uh by using voice and fleshed out by Kimmy. So, here is where we are at the game now. So, I'm going to open up the game and I'm going to show you a bit because I want to expand this now. So, this is kind of the state of the game. Now, I'm going to show you the save I've been working on. Uh, yeah, you can see I've come quite far here. And this is the big you can see all the the things I've built out here. So, we are at 1.5 GB of compute now. We have four agent slots and now we are really ramping up. You can see our model training run. Uh, yeah, if it check out, we have a model now that is how big is it? We have trained five different models. So the latest one we trained is uh 179 billion uh parameter model. So when we train you can see let's just start a training run. Now you can see we have this log loss right and we run the compute the more compute you have of course you can perform better. And I'm just going to I'm just going to pause this because yeah I don't care. We just released this now. It doesn't matter. But that when you have trained a good model, you can search where you want to if you want to serve that model, right? So if we serve the model, we make more money. I'm not going to go through every single detail, but we can serve compute or rent out compute or we can serve compute. We can allocate compute to research and everything, but basically uh and we have the research. So you can see here I done a lot of research like on the model side the software side and we have the workplace here I have uh allocated like uh models I have my agent car Ilia like Ilia Skever and he has control of uh doing the model training and there's like a bunch of stats here but this is run by um in this case it's an open code so we have an AI agent like uh an open code version of like a local model that is actually controlling this. So, we do have like an AI agent inside this now that kind of controls. You can see they have not some experiments they've done and stuff. But I don't want to spend too much time talking about everything I've done so far because and here are some more work the agents can do to improve, right? We can set an agent to improve that. Uh, and of course, we have the build here. You can just click on a field and build out something. Here's the other lab. So we can see what they are doing, right? So we can kind of spy on them because we have here we have something called we can spy poach prize war lobby and stuff like that. Gets really expensive though. But uh there's something we can do. We have our clo code here if we want to do something in the game. And we can swap this to open code or other harnesses if you want to. Now I set up with plot code here to yeah help me out with the game if I want to. But we are going to focus now on actually uh improving the game because I I feel there are some more things we can do. So I had some suggestions here from Kim K3.1 Pro. So we want to actually expand the game so we can do more. So we can do chip plus supply, lead times, grid queue and power deals, hardware failures and stuff like that, infront rack selling waste heat. So we want to expand our game a bit more. Data as a resource, we can have API customers, a model lineup instead of only one model, benchmark gaming, weight theft, that was pretty cool. Uh export control and regulations, star researchers and everything. So it has a recommended implementation order here. You can see that. So P1, M1, M2, H1, and P2. So I think we're just going to run this now and see what we can actually do uh upgrade our game and then we're going to test it out and see how well it works. So I'm just going to say start on these implementations and yeah, let's start working on this. Okay. So now you can see we have our P1 M1 data as a resource API and customers ship supply and a visible next goal. That's pretty cool. So let's see how this got implemented now. So if we do continue, you can see we already have a big customer. That's pretty cool. And give us like 30 seconds to So let's just take the contract. I think we're going to do that. So this will increase our earnings. But let's take a look at what we did change now. So I think if we go to research, we added a couple of new things. So now we have reasoning models. Let's just start that research. Okay, so I had that lined up. That's pretty cool. So we added reasoning models, research, and we have immersion cooling. 20 million. Wow, that was expensive. But that's pretty cool. That is two things. And if we go to And if you go to models now uh let me find that let me close this. You can see we now have something called license. So we can do some licensing of models. We can turn on and off this side if we want to do a reasoning model or not in training. If you want to do more chat coding or enterprise that's pretty cool. So we can kind of adjust towards that. I think I kind of like that. We can kind of see we can adjust if you want more synthetic data in our training run. That's pretty cool. And license we can change. We can do uh book corpus license code science papers. So this is more expensive. 300 million per. Wow, that was really expensive. Um but yeah, that gives us like a lot of new options. And this is what's so cool about using like Kimik True or Kim K 3.1 in this case to kind of adjust to get the game exactly where you want it to be because this is my game. I don't have any plans to sell this but I want to play it. So now this is kind of my experntally run and I just want to add like a bunch of cool features. Maybe in the future I want to sell this but now uh I just want to make it and have fun with this. Was there anything else we did? Yeah, you can see we added like now you can see we added a few more. So under here we are like campus. So here you can kind of see now a bit more interesting stats on where we are. I found it very interesting that one thing we built was I saw this in the news today. Here is our AI campus, right? Because I think OpenAI is trying to actually change the name to some kind of AI compute campus or something for data centers. So, this is already in the game because I think we picked it up when Kim Kry did some online research. So, why don't we just build out a couple of new features I want maybe do some wish roll features. You can see here's my wind farm for power. So, let's think of a few new expansions uh for the game and let's see if we can do a bit more here. So what I went ahead and did is overnight I ran Kimmy 3.1 over the night on some few things I wanted to upgrade. So uh one thing we wanted was uh more animations in the map. Uh we also wanted star researchers. So if you kind of go here now uh you can see if we go to team now we have the option here to hire more uh we have more heads and rolls. So here we have all these heads and we also have down here we have um guess I can do this. We have our researchers. So we can hire energy hardware and software researchers and these kind of we pay for per minute. We also did ask for some the site architect. So this uh feature is kind of takes um our setup and kind of disturbs it into a more like a better flow for both uh you can see power flow and data flow. So we kind of optimize you can see now we don't have any buildings but if we build out uh we can use or hire the site architect to kind of optimize the flow of both power and data to get more uh out of it. And also you can see now we have something called uh use AI behind our um roles here. So we can decide if we want to hire like a human and pay them like u $40 a minute 10 minutes of it up front something like that. But if we train a model we can kind of use if so now we don't have any trained models but if we do we can use that model. So AI so the better model we have the better these AI agents of head of MQ CFO head of models will be. So if you have a really good model we can get like up to 100% uh performance uh in these type of roles. So this is something I wanted to also include. Uh we have a HQ. So now we can if you go to build I don't think we have it yet but we can grow our this is our garage. So we can kind of update this into like a big headquarters. I think I I kind of like that in these games we have more business layers. We have something called finance. So if you go to finance now you can see now we can kind of our companies has a value. We can do funding rounds to um raise money. We have a quarter earnings. We can do depth and leases contract startups for sale. And we also have a news feed. I don't know if you had that before, but our Vanta takes a shot at us. We can clap back. Okay, that didn't go too well. Uh yeah, we're going to skip that. So, we can kind of interact with the news and we have some kind of hype meter how popular our company is. So, basically uh I ran uh Kim K3 uh in like a loop overnight in like a almost like a slash goal and it built everything out. So, uh I thought we can do some live changes now. So, the first thing I want to do is be able to toggle on and off the mini map. I don't like that. So, let's do uh add a mini map toggle uh on off in settings. Something like that. Let's just do that. Uh because I don't like this mini mapping up all the time. Kind of takes up some real estate here. So, let's just build like a grid connection and let's build a server closet so we can get some compute. And after that, we can start uh yeah, we can do a rack, too. So, let's buy a couple of racks. Yeah. So, now we have something to start with here. So, while Kim is working on that, and here you can see we have a rack coming into the server. Perfect. And yeah, we got some more racks. Now, we have a bit more compute. So, you can see there's a small truck coming with the racks here. I thought it was pretty funny. Another thing I want kind of the old style plus money thing. That's that's pretty cool. Also add the plus money uh bigger. Uh let me add that. Okay. So you can see now we added a mini map on off button. Uh and we also did add the plus uh like I will show you soon. So let's do a new game now. So let's call our s intelligence. that. So here is kind of start. We have like a a tutorial. So build a grid connector if we wanted to do that. So let's just do that and do next. Uh uh build a server closet. Yeah, we can do that. Buy four racks. Two, three, four. Perfect. And yeah, they should come in soon. So let's check out now if our features worked. Let's go to menu settings. Uh, mini map off. Yeah, it worked. That's great. So, we can skip the tutorial for now. You can now see uh our racks should come in here. Yeah, perfect. Uh, and I wanted to see if Yeah, you can see now we kind of changed that. This is kind of the old plus $16 style, right? So now we have an uh agent seat available. So we can staff someone here. Yeah, we can't staff now, but uh we we need a bit more money before we can do that. So yeah, that worked pretty good. So now we have something up and running here. And we kind of want to handle this, I think, afterwards. So you can see super easy to make changes in game now. And of course we can fire up our agent here and we can help him. I think this is like a really cool feature. So we can have cloud code in this case could be open code could be codeex. We could just came here but this was what I had set up before. So I just wanted to so we can ask something like what is the next best move for us? If you wanted some extra help, right? And you can see we get some advice on what we should do next. Next hire a worker. dispatch batching start a research project save for a data center and stuff like that. So we can just close that. Yeah, it's been really nice to actually use chem Kim K3 Pro for this and they are I think they are not optimized for game development of course but they work really good for it I got to say and all the sprites and stuff we Here's an investor call. Yeah, we don't have the money for that. Or we can get 3,000. Yeah, let's do that. So now we have some money, right? And then we can go to work. We can staff our como hire. So we delegate. And now they will handle all our um yeah news and social media. I don't want to deal with that. So now we can start looking at our team. Should we hire a site architect? N a bit early for that. But we could look to hire more people here, right? If we wanted to. And I think that's pretty good. So basically, I think this is kind of what I wanted to share because this game is far from finished. So I'm just going to keep on building this out and it's kind of going to be I I just think it's so interesting that you can kind of now build out a game you want. So it doesn't really matter too much. Like for now, I don't have any plan to sell this, but I really want to just build out um build it out into a game I like to play that is just for me. Maybe I share it with some friends and stuff. So nowadays with models like Kimmy and stuff like that, it's super easy to just build a game you want to play. And I found that so liberating, right? It's just so fun. I kind of like both sides. I like playing the game, but I also like developing the game, coming up with new ideas, asking. So we can ask something like suggest a cool feature, something like that. So here's the suggestion. So feature suggestion the lab reind cinematic time lapse that replays the company's growth uh on the actual map. Okay. Yeah, maybe we don't need that, right? But this you could could spawn some ideas around that if you wanted to, right? So you're you kind of create like a video on this. So what next? I saw that Kimmy has pretty good stats on like um 3D modeling. So I was thinking maybe creating some assets for our game. maybe using Blender or something. Uh haven't tried that but that might be something I'm going to try. Now we have a heat wave here. So you can see if we look at power now we are struggling with cooling but this is something I might look into and I might do a followup on this and see where we are in like uh whoa our server closet took fire. That's not good. Fire in the server. I'm probably going to lose this one. I don't have anything to do with that. Maybe we need some kind of security or like insurance or something that could be interesting to add in. But yeah, that is what I wanted to do. Uh really enjoyed playing around with Open Code again. Uh it's been a while since last time and this new release was something I wanted to test out. So yeah, I had a a good experiments with Kim K 3.1 actually developing this. So that was pretty fun. Haven't tried those since Kim K 2 I think. But uh yeah, good experience. Hope you enjoyed it and I'll see you again
Related Intelligence
All stories →Ultimate Gemini Guide: How to Use Gemini 4 Argon (Beginner to Pro)
Try HeyGen https://www.heygen.com/?sid=rewardful&utm_content=creator&utm_medium=influencera&via=artur-vishnevskii 🧠 Never miss latest Gemini Updates! Join to my 15,000+ students → https://aimaster.me/yt/argon 🤝 Exclusive brand partnership - https://aimaster.me/collab Gemini 4…
How to predict the behavior of molecules from scant data
Discovering new molecules for specific uses is a major driver of progress in a wide variety of industries, whether it’s pharmaceutical companies looking for new drugs, or electronics or energy companies seeking new materials with particular properties. But it’s also largely a…
Hundreds of Artex AI hacking tools found publicly exposed to the web
Security researchers found hundreds of servers hosting the penetration testing tool Artex exposed to the internet, allowing anyone access to the hacking tool.