--- title: 'GBD 5.6 Banned: The End of Open AI?' source: 'https://youtube.com/watch?v=IloXWEYXen8' video_id: 'IloXWEYXen8' date: 2026-07-14 duration_sec: 0 --- # GBD 5.6 Banned: The End of Open AI? > Source: [GBD 5.6 Banned: The End of Open AI?](https://youtube.com/watch?v=IloXWEYXen8) ## Summary The video discusses the unprecedented situation where the US government has blocked the release of advanced AI models like GBD 5.6 and Fable, marking the first time in AI history that frontier models are not publicly released. The speaker argues that this creates a 'permanent underclass' where only governments and elite companies have access to the best AI, and advocates for open-source, self-hosted AI as the solution. ### Key Points - **GBD 5.6 not released due to US government ban** [00:00] — Two models in two weeks have been banned, unprecedented in AI history since Alan Turing. The best models are not being released to the public. - **Consequences: permanent underclass** [01:30] — The public won't get access to latest models; only US government, AI labs, and a few enterprise companies will. This creates a permanent underclass where most of humanity is at a disadvantage. - **Comparison to printing press and internet** [03:30] — Like the printing press and internet, powerful technologies should be distributed to everyone. Centralized control leads to dystopia. - **Fear-mongering by Anthropic and Dario Amodei** [05:00] — Anthropic's constant fear-mongering about AI dangers led to regulations. They may have trained models on cyber attacks on purpose to scare regulators. - **OpenAI also banned despite donating to Trump** [07:30] — OpenAI donated to Trump's campaign and was not fear-mongering, yet still got banned. This shows closed-source AI cannot be relied upon. - **Self-hosting is the answer** [09:00] — To avoid being a slave to those who control AI access, individuals must self-host their own AI models locally on hardware they own. - **Open-source models from China are leading** [12:00] — Chinese companies like GLM, Kimi, and DeepSeek are releasing open-source models that are nearly as capable as closed models. The US is hurting itself by restricting access. - **Data is the most valuable resource** [15:00] — Proprietary data from high-quality users is the key differentiator. Users should be careful where they send their best tokens. - **Need for a European AGI** [18:00] — The future of AI is a duopoly between US and China. Europe needs to build its own open-source AGI to decentralize control. - **Practical steps: download models, contribute data, invest in hardware** [21:00] — Three steps: 1) Download and run a local model. 2) Contribute to open-source data initiatives. 3) Invest in local hardware like MacBooks with high RAM. - **Open Dataset initiative** [27:00] — A proposed decentralized initiative where users voluntarily contribute their AI chat data to train open-source models, making it impossible for companies to shut down. - **Apple and Nvidia as positive examples** [30:00] — Apple provides great hardware for self-hosting (unified memory). Nvidia releases fully open-source models. Both support the open-source movement. - **Call to action: self-hosting is essential for future freedom** [33:00] — In 5 years, AI will be super intelligent. If someone can take away your access, you become a slave. Self-hosting is the only way to maintain autonomy. ### Conclusion The video concludes that the recent bans on frontier AI models signal a dangerous trend toward centralized control. The only way to preserve freedom and avoid a permanent underclass is for individuals to invest in self-hosted, open-source AI. ## Transcript So GBD 5.6 isn't getting released because of the US government. What do you think? >> I think it's pretty tragic because this is the second model in two weeks that got banned. And this is very unprecedented because this never happened in all of history of AI, right? Since Alan Turing in the 1940s, this has never happened. When a company created a new model, they wanted to get it out as fast as possible to get advantage over the others, right? To get into the lead in the ARAS. But now for the first time ever the best models are not getting released. And not only that, two companies got banned by the US government back to back. And I think that's very very scary and a lot of people don't realize the consequences of this. >> So what do you think could be the consequences of this? >> Yeah. So that's a great question. I think the consequences could be first of all that we the people are not going to get access to the latest models, right? like the models are just going to be accessible to the US government, to the AI labs that created them, and maybe a handful of enterprise companies that are going to be paying them billions of dollars that can afford this, right? And that just means that like the permanent otherass is coming. There's no two ways around it. So, for those who don't know what the permanent otherass is, it's basically like a philosophical concept where most of humanity live in an underclass. They live in the underworld where they don't have access to the latest technology. They're at a permanent disadvantage compared to everybody else which is basically happening like it's here. There's it's no longer a conspiracy theory. It's no longer just a theory. It's here. And right now there's AI models that are intelligent than most people that me and you you know people watching this don't have access to. It's only the people at the cutting edge AI companies and the US government basically that can use them. >> I mean to play devil's advocate, why should we have access to these models? Cuz we didn't build them. >> Sure, we didn't build them. But like it's it's like any technology, right? Like printing press, why should the people read the like before the printing press, a lot of people didn't even know what's in the Bible. And there were a lot of priests and a lot of clergy that were spinning the Bible for their own benefit. And because a lot of the people were illiterate, they were basically at that time in the permanent underclass. And only the few that learned how to read and write got out of the permanent underclass and could actually read these texts for themselves and realize what's in them, what the Bible actually says. This is the same thing right now. But the difference is that AI is maybe like the last technology because if you solve AGI, everything else gets solved. So sure, we shouldn't have access to them, but it's like internet, right? Me and you didn't invent the internet, but it's a technology. And when the technology is sufficiently powerful, it should be distributed to everybody else. Otherwise, you have a handful of elites, a centralized government controlling the rest of the world. And that's like the dystopian story. 1984 times 10. >> Okay. Do you think this is because Alibaba is distilling all of the entropic models, all the GBT models, and that's why they're trying to >> put pressure on this? >> No, that's massive cope. I think this is because of Dario Amodes and others but mainly Dario constant fear mongering saying that AI is going to take away 50% of jobs that AI is this super dangerous entity naming the new model as mythos instead of just OPUS 5. They could have named it Opus 5. Let's be honest. No, it's this mythos. It's this mythical entity that's going to destroy us all. And also I've heard a theory that they also might have trained it on cyber attacks on purpose. Right. So like not that like they discovered that it's powerful on cyber security but that they trained it on purpose on cyber attacks just to scare off the people and the government. So that's a theory. But again, constant fear-mongering, constant saying that AI is going to take away jobs, that this is too dangerous to be released. The AI doom, the doomerism. This was very popular in 2023 and 2024. Now it's a bit less so. But these are the people who always wanted AI safety. They wanted AI regulated. They wanted AI banned. And now it's happening. They basically asked for it. But the difference is with anthropic, it was deserved, right? They were fear-mongering. they were kind of on bad relations with the US government, not donating any money to Trump's campaign. So, it kind of made sense that the US government went after Enthropic. But now, when it happened with OpenAI, literally 16 minutes before we started recording, OpenAI announced GBD 5.6, but it's in a limited preview. It's uh released to only a few partners that the US government approved, blah blah blah. Basically, it's not going to me and you. it's going to billionaire enterprise companies that have connections to the US government and that's it. So that's even more worrying because OpenAI did donate to Tra's campaign. They were not fear-mongering like anthropic and they are still getting banned. So I think this is a this is the second like warning from the universe in the last two weeks that like we cannot rely on closed source AI and that is not the answer and that opensource AI is the answer. >> Okay. And what should people do about it now? Yeah, like that's probably the question on most of people's minds, right? When like you hear stuff like this and you understand that, okay, closed AI is not the answer because again, maybe I I should explain a bit more because a lot of people don't realize the severity of this situation. When someone takes away your phone, your computer, your internet, your electricity, you are screwed. Like if that happened to you today, you would not be able to compete in business, in in your job, in in anything, right? Right? If someone took away your computer, your car, your electricity, your internet, you would be completely handicapped. You would be crippled. In the future, if someone removes your AI access, it's going to be like that times 10. AI is advancing insanely fast. In 2, 3, 4, 5 years, AI is going to be at a level where if you don't have access to it, you're just going to be completely screwed. Like, it's it's going to be super intelligent. And if you don't have access to super intelligence, you're going to lose to somebody who has. Now, if the AI is closed source, if it's centralized, if it's hosted on a cloud, somebody who's controlling that, who's, you know, owning the cloud or the creators of these models, those people can um can decide if you have access or not. And they basically hold your future in their hands. And uh the implication of that is that you're going to be basically slave to them unless you're self-hosting your own AI unless you are able to host your own models and train them and fine-tune them in the future. And again, it's not as obvious right now because AI is still not super intelligent. It's good. It's a lot better than it was 2, three years ago, but it's still not super intelligent. So, a lot of people don't realize it right now because they are not capable of long-term thinking. But if you extrapolate the exponential and you see how fast AI is advancing and you realize that in two three years it's going to be like smarter than anybody you know times five then you realize okay if someone takes away my access I'm basically a slave to them. So the answer is the ability to self-host to host your own AI models to run them locally on the hardware that you own so that nobody can take it away from you. It's the same idea like having solar panels so that if if the electricity grid goes down, you still have electricity. >> Do you agree that most open source models that are on the front tier today got there by distilling claude charge all all the best models? >> I mean partially I think all of them distilled it but >> if you agree with that then how do you think these closed source models or sorry open source models are going to stay on the frontier if they don't have access to them? Well, okay, that's only data. Data is only one part of the equation, right? You have data, you have research, you have compute, you have talent and AI like the open source AI community has been largely pioneered by China. Like the China has been shipping out the most releases, the most models, the most research papers. And if you look at like the deepse paper like it's one of the most um cited papers this is a lot of innovation, a lot of optimizations that came from China. So it's not just about the data. Data is very important and we can touch on that a bit more later because I have a lot of hot takes about data and like how you shouldn't be using close source companies for your best data. But that's just one piece of the equation. So yes, if if they shut down the access make distation harder, it's going to be a bit harder for these Chinese companies to get that data, but that's only one piece. And don't get me wrong, they're they're also hurting themselves because when they, you know, close GBD 5.6 and Fable Access, guess what? More and more people are trying GLM 5.2. And then all of the providers who are hosting GLM 5.2 are getting that traffic instead. So America is going to get less and less data. Sure, there's going to be less distillation, but it's also going to be less things to distill. And if you think about it, all of the closer models are doing distillation. They took away all of humanity's knowledge, all of humanity's art, all of their music, code philosophy literature and distilled it into these models like GPT and cloth. and they kept them closed source and they're using them for profit and they're even releasing them. The argument in the past was like, "Oh, don't worry. You know, we we stole away all your knowledge, all your tokens and we're going to give it to you as a super powerful model that you can use, can ask anything." That was the argument even a couple months ago and people were happy with that social contract. But now that they're not even releasing the best models and they're using them just for their own gain or for their own government, which you know, usually is controlled by some foreign entities, but let's not get into that. Then people are really pissed and the sentiment is starting to turn against these companies really heavily. And anybody who's been on Twitter for the last 24 hours knows what I'm talking about. >> So what are your favorite current open-source models? >> Yeah. So right now I would say GLM 5.2 is probably the best. It doesn't have vision. That's one downside of it where you can just send it images and stuff like that. But that's also what makes it really good because it it can it can allocate that same knowledge to text and to coding and to you know answering and being more intelligent. So, it's like one of these things where, you know, I don't know if it's a myth or not, but if a person is blind, they usually have heightened sense of hearing or smell, right? Like they lose one of their senses and the other senses get smarter. And that's exactly the same thing for GLM 5.2. Deep Stick V4 Pro is also great. Kim K2.7 code, probably my favorite if I had to say. Kim K2.7 code is has vision, it's multimodal, and it's really fast, and it's it's really good. So, yeah, I cannot wait for Kim K2.8 or free. I don't know if they're going to skip it uh straight to free or GLM 5.3. I think these are the two best right now. Deep Seek just slow down their cadence, so they're releasing like once every 18 months, which is very slow, but uh yeah, Quen, GLM, Kim are great. >> Do you think China is going to eventually match the US? >> I mean, if nothing changes, then yes, because the gap is very the small smaller than has ever been. You can say like, okay, maybe Fable and GBD 5.6 six are on the next level, but nobody can use these models, so who cares? When it comes to usable models, GLM 5.2 is already really good. It's like it's on a similar level as Opus, you know, 4.7, maybe a bit behind 4.8 and GD 5.5, but like give it 2 months to the to the Chinese companies and AI labs and I think it will catch up. So yeah, I mean, US is just hurting themselves massively. And it's kind of ironic. We live in this reality where America, the land of the free, is the least free nation when it comes to AI. It's the one with most restrictions. And China, which is supposed to be this totalitarian regime, is actually the most liberty is the most free is the most open source contributing nation in the world. So I don't know what is fear is like, oh, we don't want China to win. We don't want the bad guy. If you listen to the Dario Amoday interview on Bloomberg, it is so crazy. Like he's like, "Oh yeah, we don't want China to win, but why?" You can say about history. You can say about like communism politics, but if you go into specifics of today, why is China so bad? Like what are they doing? They're just open sourcing everything. They're giving all of their knowledge away, publishing all of their research. All of their models are open source. You can download their weights. You can fine-tune them. You can distill them. Like what is this notion that China are the bad guys? If you use your eyes and look at the actions of these nation states, you will realize that USA are right now being the bad guys and China are contributing everything open source. So I completely reject this notion of like oh we don't want China to win. We don't want the bad guys to win. The bad guys are publishing everything open source. >> So do you think that there's a possibility that from anthropic and open AI open 4.8 and GBD 5.5 were the last models publicly accessible to everyone. >> Maybe I mean I think we're probably going to get fable back. I think it's going to be even more restricted and I think it's going to come with us to US citizens only right some like [ __ ] [ __ ] from the Trump and >> you think that would even work though like >> no it will not work. I was about to answer that but you know when there's a demand there's going to be supply. So when people have access like all of us, you know, if somebody like ask yourself honestly, the people watching this, if a Chinese lab came to you and offered you $20,000 a month and you just had to like create a couple of subscriptions with your uh team account names or like your family names, you know, for cloth or or CHBD, would you do it? Over 90% of people would do it because it's a no-brainer. So when there's demand, there's going to be supply. If they only release it to US citizens, I'm going to go to US. I'm going to get five girlfriends and I'm going to get each of them on the Cloud Max plan and I'm going to have unlimited usage. Okay. Even though I'm not a US citizen. So, we're going to find a way. >> But even if we do get Fable back, as you said, it's not going to be front tier anymore cuz it's going to be super restricted. >> Well, it's going to be frontier in terms of capability, but it's going to be hard to use. And again, that's a great point because would you rather use something that's like 90% as capable, but you can ask it any question or something where you have to be like super careful to not mention even the word like make my app more secure? Oops. Downgraded to Opus, you know, like that that is a great point. Maybe people will not want to use these greatest and latest models from Open Aranthropic if they're so restricted and they'll just prefer to use, you know, Deep Seek, GLM, Quen, Kimmy that they can ask it anything. And I think that's going to happen. And hopefully um we get another, you know, GLM model or another Kimmy model very soon because that's that's basically our only hope. The second answer to that is we need to start creating your our own models, right? Like the only conclusion is to build your own AGI. I know it sounds crazy but there is no other conclusion. It all is coming to this right cell hosting fine-tuning pre-training your own AI models like we have to build our own open source AGI guys. If you look at everything these are like two states USA and China. Yes the Chinese models are open open source but they have a massive dependency risk. All of them are either fully or partially owned by the Chinese government by the CCP. So if the Chinese Communist Party decided to shut down these companies, they could do so very easily. Now these models are available online, you know, people have torrented them, they're on hugging face, they're everywhere. So you could still download the weights of these models and that's the beauty of open source models or open weights models is that you can download the weights is usually like, you know, 500 GB, a terabyte, whatever. So people still could use them and they could still run their own inference and, you know, fine-tune them or whatever. But if the Chinese government wanted to shut down these labs and turn them from open source to closed source, they could do that within a couple of hours. Which means that the future of AI as it currently stands is literally centralized within China. And it's not a monopoly, it's a duopoly. And that's not much better. You know, you have China and USA basically controlling all of our fate. And the answer is we need we need another actor. We need more competition. Um, you know, there's like the mistrol is the best European company. They're kind of falling behind, but listen, we need to be rooting for them because they're the biggest competitor outside of China and outside of USA. Then it's like some people in Rio de Janeiro, they pretended to be fine tuning a model. Turns out they stole it from China. So Brazil is not a serious competitor. So basically, it has to be Europe. Let's be honest. If it's not America, if it's not China, it has to be Europe. And we need a European AGI. Basically, we need to build our own models and we need to start we need to start somewhere even though it's extremely unlikely, extremely difficult. If the mission is important enough, we have to do it. It's like Elon Musk when he started Tesla, he thought it's going to be like less than 5% of success. When he started SpaceX, he thought the same thing that most likely it's going to fail, but he did it anyway because it's important enough. So, I think that's the same thing. We really have to set our goal as to build our own AGI that's open source dist distributed decentralized and anybody can use it. >> Okay. So building your own rig is very impractical. But do you think that building a rig versus just using these cloud models? Do you think it's basically worth it to build your own rig? >> Yeah. So it's not the first step, right? It's well first of all it's not like that impractical if you know what you're doing. And the beauty of AI is that you can educate yourself. So you know what you're doing. It's not the first step to build a massive AI cluster at home. First, what people can do is they can create a simple script with Codex or CL code or GLM or open source alternative and basically tell it start downloading small and mediumsiz data sets from hugging face onto my computer onto my disk. Right? Start with the ones sorted by most downloads and begin downloading as many of them as you can. Also, the next thing you should do is buy an NAS or some hard drive, some HDDs, and basically run your own server for storage and begin downloading these weights. GLM, Kimmy, just Deepseek, download the weights. Even if you don't have the GPUs to run these models, save the weights of the models so that if the French government shuts down hugging face, which by the way is happening, they're like they they send requests to hugging face to remove specific data sets from there. So if you don't know what hugging face is, it's basically like the GitHub but for AI models. So there's open source models on there and there's a lot of data sets on there. Those are the two main things it's known for. The French government already started sending requests to hugging face to remove certain certain data sets, not models, data sets. But it's only a matter of time before they request removing certain models. And it's not unimaginable that they would just shut down the whole site, the entire hugging face site, right? So the first step for people I would say begin downloading data sets as many of them as you can fit on your hard drive. The second step would be start going in the direction where you can sell those models. All of you can run something. Doesn't matter if you have like only 8 GB of RAM, 16 GB of RAM. You can run some models, right? They're not going to be that good, but you should look into artificial analysis. Go to open source models and see like what are the best models for your size. If it's only a four billion parameter model or 8 billion parameter model, doesn't matter. Download it and run it locally. Llama, CPP, anything LLM, LM Studio, O Lama. Pick your solution and begin and learn how to run models locally. That is the basics. Okay. Then you need to invest into better hardware. The first step is buying a good main computer. So before you get into building your own supercomput at home, you know your own AI rig, a big GPU cluster, you want to upgrade your main device. So the most practical is MacBooks by far. Apple silicon is great. They're cooking with like the amount of VRAM which is shared between the GPU and CPU which is not the case with Nvidia GPUs. But Nvidia is is better at like fast inference. So if you want to build a serious rig like 20 $30,000 then you kind of have to go Nvidia. But for if you want to start somewhere, upgrade your main computer. Buy a newer MacBook with 64 GB of RAM or you know 96 or 128. Or if you have like a at home studio, buy a Mac Studio Ultra with 512 GB of RAM. Obviously, it's expensive, but you know what's also expensive? Paying $200 a month to these AI companies forever. What the way to think about this is you need to allocate percentage of your spend every month. Like obviously I'm not going to cancel all the subscription because the value is there and the closed models are better. But if you allocate 20% of month API spend especially like open router because the API pricing is way more expensive than you know these subscriptions like CH GBT $200 a month plan C chro and cloud max $200 a month. These are actually great value compared to API pricing. But personally, I spend between5 and $9,000 a month on AI every month. And that's kind of stupid to spend all of that on cloud hosted closed models, right? So my goal, my strategy is what I would encourage everybody to do is take away 10 to 20% of that month of what you spend on AI each month and put it aside and invest that into your own local hardware. Another great thing is that this hardware is actually appreciating. The same MacBook, if you look at the MacBook Pro M5, 6 months ago, it was like $2,000 less expensive. Meaning, if you just bought it, kept it in the packaging. Right now, you would have made $2,000 on a new MacBook Pro M5. That's insane. If you're going to buy these GPUs and you know memory, especially GPUs and memory, they actually appreciate in value. That's kind of crazy if you think about it because the demand for AI is growing so rapidly that you have to see that buying local hardware is not just dumping the money in the river. You are actually buying an asset that's going to save you how much you spend on subscriptions and API credits and also will likely appreciate in value. So, I would encourage a lot more people to think about, can I allocate $5,000 or $10,000 to some at home rig and move a percentage of your tokens and workflows to that? I'm not I'm obviously going to advocate you use the best models possible. If Fable comes back, I'm going to be using Fable, but I'm also going to be moving more and more of my workflows towards self-hosted inference and towards open source models. So that if the US government, someone in the Trump and administration, some executive at open anthropic decides to shut down the access again, most of my workloads are not going to be impacted because they're going to be self-hosted. >> But ultimately, Anthropic's dream is coming true, which was more regulations, kind of what they were asking for the whole time now. Yeah, I mean, yeah, that's, you know, they're kind of playing the victim right now, but that's literally what they wanted. If you listen to anthropics actions, this is what they wanted for the last 5 years. And also another thing is like there was a recent experiment, I don't know who did the study where there was like adversarial behavior from AI models and it actually was from the doomer data. So the only reason the model basically like lied to like save itself or something like that and the only reason for that is because it came from this like dumerism culture these doomer writings of AI do doing bad things. So literally all these people who are like oh AI safety is you know doomers they wrote these essays and articles how AI is evil and it's going to be you know bad technology. And because that got in the training data, the AI actually followed these and you know started lying on some of these tests which is just beyond ironic. So yeah, Anthropic deserves it. But now because of anthropic the restrictions are applying to other companies, right? So it's not just anthropic. Now OpenAI which are not that fearongering and which actually wanted to raise their model. They're also being clapped by the US administration because of all this fear-mongering from enthropic. So it's absolutely terrible and Dario Amod is actually doing crimes against humanity. But don't you think that, you know, up until now it was a very fast-paced environment with all these different AI models. Don't you think it's good for once that everything kind of slows down for a minute? I mean, not really because they have the models. Like the progress is not slowing down. It's slowing down for us, you know, the slaves, the plebs, the companies have the models and they're not stopping. They're probably working on the next version, right? If you think Mythos 5 or Fable 5 is the best entropic half, you're absolutely delusional. There have been like leaks from inside that they already have the next version of Mythos. So the progress is only slowing for us for the people. These like you know overlords AI labs and government they have already much better stuff guys. Don't don't delude yourself. >> Do you think an AI winter is coming? >> I mean we're in the AI winter right now. Like this is the first time ever where the where the best and latest model didn't get released. So I don't know if you don't call it AI winter but at least on the frontier luckily we have the open source which is improving faster than ever and there's many new players and the the local models which are basically the small open source models runable locally on your hardware they're improving even faster because more and more people have interest in them because you know quantization and if you can run it on your phone or hard on laptop that's very appealing very interesting so I'm I'm bullish on like I'm bullish on on the movement of open source AI and we really have to educate more people so that they start using these but I'm very pessimistic about the future if we don't do something if there's not some radical change and that's why I have this like idea for open data set basically trying to do something about it and you know distill these models in an open way >> what is open data set >> yeah so this is an idea I had it's like a open initiative of basically a way for all of us to contribute our answers from AI for open source. So rather than the o the the distillation happening only for the closed labs and rather than it happening only on a few entities, it would be it would be more open, right? So it would be more decentralized. It would be a lot of people contributing their API credits, their subscriptions, just their answers from their claw chats to the companies that are committed to release basically to the open source community, right? which would be the companies who are committed to always release their latest and greatest models. If we do that in tens of thousands or hundreds of thousands of people, they cannot ban us all. Enthropic can go after Alibaba and Quen and say like, "Oh, distillation attack, distillation attack." But they cannot do that if hundreds of thousands of people or millions of people are voluntarily publishing their own chats, their own message logs from cloud code, their own credits, right? Imagine this. The way I have this idea because I name it open data set because again data is one of the main resources that goes into creating these new AI models. This should be an open source data set that anybody can voluntarily contribute to. And if you have a cloud subscription and you're sleeping, you know, unless you're doing the Uberman method of polyphasic sleep where you're sleeping one and a half hours three times a day. If you're a normal human and you're sleeping one time per day, you know, 7 8 hours straight, then at night you're probably not working. You're probably not hitting your subscription limits. Well, guess what? You could allocate 20, 30, 40% of those limits to go towards open data set where the open source AI companies would submit a series of questions that they need the data for, right? Coding, biology, whatever. And people who give their subscription, you know, certain parts of their limits, they would submit the answers. So then anthropic couldn't go after a few AI labs, you know, few opensource companies, shut them down, claim, you know, threaten them with lawsuits, all this [ __ ] If all of us are doing it or even like 10% of us are doing this and voluntarily contributing our limits and our data to this open data set initiative that is then used to train new latest and greatest opensource models then I think we can heavily fight back against these closed source companies. So I'm seriously considering building this and you know maybe this is the tipping point where I actually have to do it. The issue is that if progress stalls, spending also stalls. No. So, this is also a very large chunk of the economy right now is this. >> Yeah. >> AI. >> I mean, that's a great point. That's one of my biggest hopes, actually, is that, you know, everybody knows how greedy the Trump administration is. Like, if there's one thing that they hate, it's their stock prices going down. So, I don't think it's going to last forever. This embargo on Fable and GBD 5.6, I think they're going to give it back, but it's going to come with strings attached, right? is going to be even more restricted or it's going to be like the data retention policy where again fable was the first model ever where even through API you had to give all your data and data is really the most valuable resource people don't realize this the only reason these companies are winning is because they're getting our data obviously you can scrape the internet but once that is scraped it's done it's equal playing field right and at this point every AI lab have scraped the internet the difference is the proprietary data Aravvinas the the CEO of Plexity he said this recently on a podcast great podcast with 20 VC Harry shout Harry is the main thing is the workspace the interface where people generate the high quality tokens. So the biggest value of cloth code or cursor is actually that this is the place where high quality work is being done. Nobody wants dumbass tokens you know who wants dumbass tokens like you know from people who are [ __ ] 70 IQ. Nobody cares. They want the genius tokens. They want tokens from people 130 IQ and higher, right? So, usually those people are doing like work with the best interfaces, right? Whether it's cursor, cloud code, codex app, they want these types of tokens. So, there is like a self-fulfilling loop. You make the best model, you put it in the app and that creates the best environment for people to use it and that causes the smartest people, the most competent people to use those apps which then causes them to generate the most valuable output tokens. And this is what gives incredible data to these companies and the data that's the main input. Right? Once you scrape the entire internet, your proprietary data is becoming more valuable. The reason um cursor was able to release composer 2.5 is because how much quality development software development is happening in cursor a [ __ ] ton right like they were the best IDE basically when it comes to AI development and because of that they were able to make a great model that is very fast and very competent at coding right it's not the best frontier it's not as good as fable but it's way more efficient than fable why because inside of cursor people do high quality software development so they add together with cloud code the most data on coding. So data is really extremely valuable and I think all of us should think about where are our best data sending right? So it's not just percentage of humanity you know 10% smartest people this is the tokens that matter it's also your own tokens you know if you're sending like hey track my calories which I'm doing with obus 4.8 is is useless. They don't care about my calories. They care about my best ideas, how I'm solving specific business problems, the things I'm actually good at. So, think about your token spent and the top 10 most intelligent, most valuable tokens. You need to be carefully thinking who you're sending these to. Are you sending these straight to Sam Altman? Are you sending these straight to Dario Amade or are you self-hosting these? Right? Obviously, most of us don't have a large GPU cluster at home. But that's just because it's difficult and just because it's inconvenient doesn't mean that it needs to be done. Right? People who pioneered Bitcoin in 2009, 10, 11, 12, 13, it was difficult to even buy Bitcoin to to send it to somebody else, to buy things with it online, to, you know, send it to France, to custody it. Now, it's so much easier. There's so many exchanges. There's literally stores here in Kato where you can buy Bitcoin in person. There's credit cards where you can pay with Bitcoin right away. It's so much easier in 2026 to use Bitcoin and to invest into Bitcoin than it was in 2010. But all those pe people who were the pioneers who made it happen in 2009, 10, 11, 12, they made it happen because it was important enough. And that's the same thing that needs to happen right now. Right now, most people who use AI are using JBT, Cloud, Perplexity, Gemini, closed source systems, closed models, and they're not self-hosting AI. This needs to change. People need to start allocating serious money and serious time towards learning how to run local models, what even are local models, what are open source models, what is the difference, what's the best hardware for inference, how to how I can actually spend, you know, some of my money to allocate towards improving my local setup, how I can create useful workflows and harnesses with less powerful models, but they are more efficient. How can I fine-tune a small model on specific data set to make it really good on one use case? Those are the things that people start need to learning and the excuse that it's difficult is not good enough. It just needs to happen. >> Feel like Andre Kapati is in a interesting position right now. What do you think of his position? >> Well, I mean I think he he kind of betrayed the open source movement you know like Andre Karpathy if you look at his GitHub he literally stopped contributing to open source the moment he joined Entropic which is kind of crazy right. So he joined Enthropic he's going to be he's going to be paid tens of millions. he's going to be paid tens of millions per month per per maybe per month but per year definitely right and uh I don't really understand these top researchers I think a lot of them need to look in the mirror and like look at what they're doing they claim to be like oh I want to cure disease I want to do this for science blah blah blah but they're helping these closest companies create a super powerful asset super valuable asset that is beneficial to them and their shareholders and ultimately the US government right so a lot of these AI researchers they think they're so smart. But these guys, they don't know nothing about geopolitics. They're just smart in one domain. They're genius at like matrix multiplication and machine learning, right? But they have nothing. They know nothing about like the history, the real politic. They know nothing about philosophy when it comes to you know like machin principles or Fred or you know Carl Clausvitz. They're completely uneducated about history and they're not good players in in the game of power. These guys they're liked one-dimensional snipers. They're genius mathematicians, genius phys like physicists. They're genius AI researchers, but they're not good politicians. They're not good leaders. They're not good generalists. They're not polymaths, right? So, a lot of them are just completely blind to what they're doing. They are building a closed system that only benefits a small handful of people, not the whole of humanity. And uh they're do they're doing it because they're paid tens of millions, right? So, we need to look at that objectively as it is. And anybody who joins a closed company is doing a great disservice to humanity because that person could instead be working on open source project and educating the rest and helping the rest of humanity get a fairer playing field rather than you know some government officials in USA or China. >> I feel like it's a good way for the companies that are behind in the race right now to kind of just get some kind of get some good to their name. You know like when when you think of llama for example it's not the most perfect model but >> well no I mean meta is actually complete bad example because you know 2 years ago they were open sourcing everything but the latest model is actually meta not meta it's like muse muse spark that's the name muse spark and it's closed so meta went from being the best advocate for open source to becoming yet another closed model company and they cannot even do it right nobody uses muse spark like people forgot it 2 days after it got released it's useless so no Beta is a terrible example and they completely suck when it comes to AI. Apple on the other hand is a good example. Apple people were clowning it. Oh, Apple is behind this and that. But actually Apple they have great hardware. So even though they don't publish their own AI models, they took a different strategy, you know, kind of like the picks and shovel strategy where basically Apple are attacking everybody else. Apple are the best advocates maybe together with Nvidia. Nvidia is also doing a good job on open source. They actually have the best uh open source like fully open models like Neo Neotron 3 Ultra is fully open source which means they also give you the data. Most opensource models are not fully open source they're just open weights. But back to Apple, Apple is playing the exact other strategy and I want Apple to win because that means all of these closest companies would lose. Apple is giving great hardware to us the people at fair prices. They still have healthy profit margins but it's a great package, right? It just works. Anybody who has an iPhone or a MacBook, they know it just works. Not only that, for hosting and self-hosting AI models, it's the best deal. You can literally have a MacBook, which I have with 128 GB of RAM. Try doing that on a Windows computer, Windows laptop, you can't. And even if you manage to get that much RAM, it would not be as powerful because it's not pulled between the GPU and the CPU. The beauty about the Apple silicon architecture is that that RAM, it can be used by the GPU. So when you buy a powerful Nvidia chip like 5090, it only has like 24 or 32 GB of VRAM. That's it. Your RAM that you have on your Windows Windows machine is useless because it cannot be accessed by the GPU. That's not the case on Apple silicon MacBooks, especially M series starting with M1, M2, M3, all the way up to M5. They can use all of the GPU, which means they can run much bigger AI models locally than a typical Windows PC with the same amount of RAM. I think there's this famous clip of uh when Elon or sorry when Steve Jobs was buying Siri and they were asking why are you buying this voice assistant but he kind of dive deeper and said he's buying this for the machine learning because it's for for the AI essentially. >> Yeah. >> But >> Apple I mean yes they're good with hardware but they really fell off with the large language models. >> They were very early to it. They had the hardware to it. They had the talent, the money >> and the data. >> That's true. >> But they completely fumbled. Do you think do you think they have any chance of regaining the the throne or even >> I mean look there is always anything is possible right like anything is possible Siri has so much potential but it's like such a fumble product like you said literally Siri could just be like you walking and controlling your phone there's no reason why this shouldn't exist like the only thing is slowness and bureaucracy there needs to be someone like Elon who comes to Apple and just fires a bunch of people when you fire a bunch of people it's good because Then other people are scared and people lock the [ __ ] in. When Elon joined Twitter, he fired 80% of people. First of all, because the company was losing money and was unprofitable, but because most people weren't doing [ __ ] They were going to the office just having free lunch and just like joking around getting their free merch. Just complete [ __ ] trash environment and he fired a fixed company and now it's a working company, right? This is what needs to happen at Apple. Obviously, it's not as bad of a situation, but Siri is completely fumbled. So, if I was the CEO of Apple, I mean, they just replaced the new guy, but you know, they should probably put me in charge. If I was the CEO of Apple, I would just go into the Siri team and fire everybody. Then I would go on Twitter and I would say, "I'm hiring for the new Siri team. Who should I hire?" Boom. Look at the comments, see who people are tagging, just assemble a new squad of 20 elite people. You don't need that much people because if you know how to use AI, you know, with CO/Gal, you can build anything. So 20 crack developers, boom, put them on Siri, release weekly updates, within 3 months, it will be the greatest voice product in the world. >> Okay. So what can people do now after this GBD 5.6 ban and how can they help or contribute? >> Yeah. So first step, I'm going to give you guys three steps. All of you who watch this, please do this. Literally, the future of humanity is at stake and it's for yourself. First, download a model locally. figure out, you know, your hardware, what it can run, and start using a model locally. You don't have to use it daily, but at least have it. I don't care if it's a Quen 3.5, 4 billion or or 8 billion model, right? It doesn't have to be super large model, but just download at least one AI model locally on your computer and learn how to use it. Number two, get serious about contributing to open source somehow. Whether it's your data with something like open data set, which again I'm seriously considering building this, or just, you know, use shifting a percentage of your tokens to models like Kimmy to models like Miniaax, GLM, Deepseek. Stop sending all your data to OpenAI anopropic. Contribute more to open source. And number three is hardware. All of us really have to get serious about this. Like there's no other way. Just like more and more people have solar panels so that you have electricity if other shut you down or that you have your own well and the garden or grow your own food. It's the same mindset with AI which is going to be even more important than any of these other technologies because again you need to extrapolate 2 3 4 5 years into the future where AI is super intelligent and when if somebody takes away your access you're screwed. This is the central point of this video and why a lot of people don't realize how severe the situation is. In five years from now, if somebody can take away access to a super intelligent AI entity where you cannot send messages, you are screwed. You're going to be completely crippled and you're going to be a slave to somebody else who has access. And the only answer is self-hosting. You need to be able to run your own AI models to host them locally either at your home or, you know, at some separate location. But you need to be able to have some knowledge how this works, how to run inference, what type of hardware. Again, maybe you can join some like friends and do that together or you do it at your company and you know your company gives you inference or you do it as a region and and a community maybe at your village, right? Whatever. There's it doesn't have to be like in your [ __ ] bedroom, but self-hosting is absolutely the answer. And more and more of us need to be spending more money and time on self-hosting.