What if you could get Chia Butt, Claude, Gemini, and Grock to answer the same question, debate each other's responses, and rank who gave the best answer, and then have a fifth AI synthesize everything into one final battle tested response. That is not hypothetical anymore. Andre Carpathy, one of OpenAI's original co-founders and former head of AI at Tesla, just open sourced something called the LLM Council, and it's free. And in this video, I'm going to show you exactly how to set this up in about 10 minutes using cursor. And here's the thing, you do not need to know how to code at all. I'm going to show you how to do this with natural language. More importantly, I'm going to show you how business owners are already using this to make better strategic decisions, validate any expensive choices, and essentially build themselves a free AI board of directors. Now, this is quite literally one of the most useful AI tools I have seen this year. So, let's get into it. So here is the problem that most people have with AI right now. You ask CHBT a question. It gives you an answer. Sounds confident. So of course you run with it. But here's the thing. Every model has blind spots. Every model has biases. Every model will confidently give you wrong information sometimes. I have been running an AI consulting business for over two years now. We've worked with over 40 different companies. And one of the things that I learned early is that when you are making important decisions, you do not want just one opinion. Of course, you want multiple perspectives that challenge each other. So, think about how actual boards of directors work where they do not hire four people who all think the same way. You of course want diverse viewpoints. You want someone who plays devil's advocate. You want somebody who's not going to be yes ma'am. You want people to poke holes in your ideas before you commit real money or any resources towards anything. Well, that is exactly what the LLM council does. But with AI models and because each model was trained differently on different types of data with different approaches, they genuinely bring different perspectives to the table where Chacht may tend to be more creative or maybe broad or formal. Claude is often more careful, nuanced, better at writing. Gemini has their own nuances. Grock is better at so and so, whatever it may be. Now, when you force each ones to answer the same questions, you can see each other's responses and then critique and rank each other where you get something much much more valuable than any single AI could give you alone. Now, to break that down a little bit further, in a business setting, this tool is taking those big make orb breakak business decisions like maybe it's expanding or hiring someone critical or launching a huge project and turns them from a single AI guest into a diverse battle tested plan. So, think of it as your own free AI board of directors. So, when you have expensive choices, the council gets models like Chad GBT gets Claude, it gets Gemini, it gets Grock to challenge each other. Now this whole process it is of course resulting in just one comprehensive final answer that looks at all the angles like the risks, the return on investment and everything else. So you do not get blindsided by one model's bias. Now the same goes for fixing a problem too. So if sales drops or people are leaving and you guys have high churn, the console can give you a few distinct theories to investigate instead of getting stuck on the first thing that sounds right. It's also going to be your lowerc cost secret weapon for validating any expensive ideas. So before you drop 15,000 on a consultant strategy or 50k on a new campaign, run the core idea through the LLM council, ask it to poke any holes or spot the blind spots and tell you what could go wrong. You're essentially getting a second, third, fourth, and fifth expert opinion for just a few cents. Plus, it's going to be amazing for reviewing contracts and proposals. Additionally, the critique phase, it is anonymous, so the AI models will ruthlessly, I should add, point out unfavorable terms or missing details in your important documents, which is a luxury you don't always get from human reviewers who are trying to be polite or even using just one AI model. Now, really quick before we break down how the LOM console actually works, I just wanted to quickly mention that if you are interested in AI news and you wants to either monetize this or create your own business in the AI space or just you're interested in learning how to use AI to grow your own business, then check out our paid inner circle. Maybe more information down below in the description where you can check all of that out. Now, anyways, let me just break down exactly what happens when you use the LLM Council. So, there's essentially four different stages. Stage one, this is first opinion. So this is where you ask your question and it gets sent to four different frontier models all simultaneously where each one it writes its response independently without seeing what the other has said and you can inspect each response individually in a tab view. Stage two, this is the review phase. So this is just where it gets rather interesting. I should add each model is now shown all the other responses. But the key is all identities as I mentioned earlier, they're anonymized. So Claw does not know which responses came from GPT. GPT doesn't know which ones came from Gemini. This prevents any favoritism or bias on model reputation where each model can then rank the responses based on accuracy and insight. Moving on to stage three. This is the critique stage. So this is where the models aren't just ranking each other, but instead they are actively criticizing mistakes and pointing out weaknesses in each other's arguments. So this is where you find out if one model has hallucinated something or made a logical leap that really doesn't hold up or or check out. And stage four, last one is the chairman synthesis. So this is a designated chairman model, usually one that you trust the most for synthesis. And then it reads the entire debate, considers all the rankings and the critiques, and it produces a final combined answer that incorporates the best elements for everybody. And what you ultimately end up with is a response that has been pressure tested by four different AI systems before you have ever acted on it. Now, normally setting something like this up would require you to open a terminal, run a bunch of commands, install Python, deal with dependencies, all that technical stuff that makes most business owners eyes, I would say, glaze over. But I'm going to be showing you guys how to set this up using Cursor's IDE. Now I know for some of you using cursor may seem a bit intimidating but I promise you if you can use chatbt then you can use cursor and your mind will be blown by how easy it actually is to use and more importantly how much more efficient and transformational it actually is. Now for those of you who do not know what cursor and more specifically what an IDE is it just stands for an AI powered development environment. Now, what that means in plain English is that you can literally tell it what you want in natural language and it will build it for you. So, there's going to be zero coding required, no terminal commands you have to enter, no technical knowledge needed. There's also going to be range of different idees out there just to add that. So, other popular ones being Google's anti-gravity. So, you do not have to just limit yourself to cursor, but each one has its pros and cons. I prefer to use cursor. But anyways, with these IDEs, you just describe what you're trying to do and they will figure out the rest. It's quite literally like having a developer on demand who works for free and never gets frustrated with any of your questions. Well, you do have to pay them, but it will be substantially minimized compared to what you would pay a normal person. So, if you ever thought that this AI stuff looks cool, but I'm not technical enough, this literally is your solution. Idees, they remove that barrier completely. So, let's move on and show you guys exactly how this works and let's start setting this thing up. Okay, so first thing that you'll want to do to get started is just going on to cursor.com. Once you're on here, create your free account and then you will be able to download here. I'll have a link down below in the description where you guys can sign up. But from there, you'll just want to download. All right, now that you download Cursor, what you're going to do is you will see first a few different options. Open project, cloning a repository, connecting via SSH. We will clone a repository and we're going to type out is github.com/carpathy/lom console. I'm actually going to click on that. Now you'll find it right here. So let's select this and let's choose the lm console. We just created a brand new folder. Use this as a brand new project. So create a new folder. Doesn't matter where you create it. And then we're going to select as a repository destination. And we are going to open this up. Now before we move on, what we're going to do is set up the rest of the system. So go ahead and find yourself on the GitHub. Again, link will be down below in the description. Inside of this git, we're just going to scroll down to the setup, and we're just copying the setup right here. So, we're just going to copy this, and all we're going to do is grab this, bring it back into LLM Council. We're going to put this in here. I'm going to say, I need you to set this up for me. Now, if you noticed, there was an API key that we need to grab. So, let's go ahead and just grab this right now. With this API key, this is going to allow us to essentially use Open Router. Now, Open Router, it allows us to access a range of different LLMs. So, make your way to open router.ai. Again, I'll have a link down below in the description for that as well. And once you create your account, you will be able to go into your API keys. Once you go into the setting, then go to keys. And we're going to create a brand new API key here. I'm going to call it whatever I would like. demo delete and expiration. I'm going to set it for an hour so none of you guys try to do anything malicious with this or steal my API key. So, let's grab this and let's go back to our LLM. We'll say down at the bottom or we can just insert it right here. Let's find it within the text. We'll expand this a little bit further. There we go. Okay, now let's enter. And from here, it's going to set up the rest of the environment where we're not going to have to do anything except for just allow for the permissions and actually give it access to mess with any of the files that you guys will see on the lefth hand side. So again, all of this stuff you can see this is the main folder, the LLM council folder that we have created earlier. Below this, this is the back end. So just some more files inside of the folder. So if we actually open up our finder right here, we can go to the LLM council demo, you can see all this stuff. It is just represented on this tab right here. Okay, so that just finished installing and we had to add our open router API key separately aside from what we had set within here. But again, alls I did is just I said, "I need you to set this up for me." It is going to do everything where I had to, you know, allow for some certain permissions and give it permissions to access all the files and stuff like that. But it took maybe 2 minutes and now everything is set up. Our environment is good to go. Now, at this point, if you want to customize which AI models actually sit on your console, you can say something just like this. I wanted to change the council models to GBT40 Claude. Actually, we'll place that with 4.5 sonnet and whatever llama model. So, can you update the config files for me? In which case, it will then update everything. So, you could have three different companies, three different training approaches, three different perspectives on every question that you ask. Now, for the chairman model that synthesizes the final answer, I am going to be keeping uh one particular model. So we can set that by just saying set GBT 5.1 as the chairman model for the final synthesis. We're going to run this off and then from here it's going to set up everything necessary. Okay. Now let's quickly just talk about the cost of this. So open router they actually charge based on tokens used. So since you are running each query through four different models plus a chairman synthesis you are using roughly five times the tokens of a single query in practice. So since you're running through each query through four different models plus a chairman synthesis, you're using roughly five times the tokens of a single query. So in practice for any kinds of strategic questions, you can expect to pay maybe about five to 20 cents per council meeting depending on how long the responses are. Now for any important business decisions where being wrong could cost you thousands or tens of thousands of dollars, spending a few cents to get four expert AI opinions genuinely is one of the best ROI decisions that you can make. I've probably spent $30 to $40 total on Open Router just running this over the past few weeks and that includes a lot of testing. So, for real business use, you can run dozens of important decisions through this for the cost of a nice launch. But in any case, with that being said, just pick one important decision that you're facing in your business right now. Something where you would normally just ask Chad or maybe ask a friend or mentor. Go to cursor set up LLM Council using the natural language approach that I just showed you. It'll take you maybe 10, 20 minutes to set up if you follow along with this video. around that decision through the council. Compare the individual responses, read all the critiques, look at the final synthesis. I think you guys would be surprised by how much more confident you feel in your decision when it has been pressure tested by four different AI perspectives rather than just one. And if you build something cool with this or use it for an interesting decision, drop a comment below. Love to hear how you guys are using it. But in any case, thank you guys for watching and I will see you in the next