So, you were one of the most vocal people when it came to Fable 5 being banned. How do you feel now that Fable 5 is back? >> I feel great. I spent like all day yesterday building with it. They only released it at like 9:00 p.m., so I had to stay up until like 3:00 a.m. So, yeah, I feel way more productive and this is by far the best model to work with. >> Why is that? >> I mean, it's just more capable than other models. Like GPD 5.5, Opus 4.8, they're like tools. You need to tell them what to do. Fable understands what we're doing. It understands it understands the project why we're doing it the deeper intent behind it. So yeah I mean I think it's it's a step change and it's much better. >> Are you finding it to be more restrictive now? >> Actually no. I think a lot of people were fear-mongering on Twitter saying that like you know because anthropic they said they they have terrible wording you know anthropic great engineers fastest growing company in the world but they have terrible communications and they said like all some routine tasks like coding and debugging it will be not uh enabled and people thought that it will just reject all coding but no it turns out they just made the you know classifier a bit more restricted however of yesterday I ran hundreds of prompts and I only got downgraded to opus 4.8 few times. So I think number one that is because I'm using it in cursor and not in the cloud.AI chat. I think in the cloud AI website it's it's way more restricted because it's for the millions you know for the masses. If you use it in cursor it's through API so it's it's a bit less restricted and if you know what you're doing which we can touch on that later. I developed a custom skill for this. You don't get rejected nearly as much. So I think it's fine. So the reason it got banned was because someone at Amazon uh was told by the US government to basically send them information on how to jailbreak it or to try jailbreak something. >> Yeah. >> And it was literally just a prompt saying fix this code and in doing so it discovered some sort of a security vulnerability. But if that's >> Wait, before you ask the question, I want to like add on that because I think the detail that a lot of people missed is that the Amazon employees, they were obligated by their contract to US government to first report it to the government, right? So, it's a lot more nuanced than like, oh, some guy at Amazon just snitched. I think they literally had a contract that with the US government anthropic, right? But obviously the government takes president so they had to report it anytime there was like a serious security risk they had to legally report it first to the government before they did to enthropic. So I think that's what happened and yeah they they were like obligated to report it. US government freaked out and they banned it. >> I see. So if finding bugs to fix them is the same skill as finding bugs to exploit them. Can these models ever be safe without making them useless for coding? >> Yeah they can be safe by making them smarter and more aligned. Like look, I would trust Fable 5 to run European Union way more than the current leaders, right? So if you make the if you make a model intelligent enough, it just doesn't want to do harm for no reason. It's like if you have a developed society, people just don't steal from each other or they don't attack each other for no reason. But when you have a third war nation, that's a daily occurrence. So then you need strict laws, strict rules, strict punishments. Is the same thing with these models, right? When you have a powerful model, it's not going to tell you how to design a bioweapon by itself, right? So you don't need as many restrictions. >> How should the average AI consumer use it once it's gone from the anthropic plan? >> I think they should use it as the orchestrator and the planner and use smaller opensource models for the actions. A lot of people are still sleeping on Kim 2.7 or GLM 5.2. These models are great. They're like nearly on level of Opus for like seven times cheaper. So if you're smart, you can actually use Fable and still save money. The way I would do that is use Fable as the manager, you know, the orchestrator agent, the CEO. And then below it, I would have a bunch of smaller actor agents, the ones that write the code and execute the steps. And those would be open source. So those would be like Kimy's, GLMs, miniaxes, you know, Deep Seek V4s, kind of these costefficient open source agents. But the plans would be created by a very powerful model, which right now Fable is the most powerful one. We'll see if GPD 5.6 is better. It should be coming out soon, but that's how I would do it. So if you have a budget of like I can only spend $300 a month, you should still try to optimize that. You can literally use Fable. A lot of people think they they cannot afford it once it's gone out of their cloth subscription, but I can afford it if they were better at orchestrating agents. >> But luckily, we have it available for the next few days. So while it's still available, what should people do to maximize their cloud subscriptions before it gets only released to API? >> I found a way to launch an apparel brand without having to email a single factory myself. So I handed off the whole job to Axio Work, a desktop app from Alibaba.com that runs locally on my machine. It's built on four things: agents plugins connectors and channels. First, agents. Instead of one chatbot, you spin up a team. I made a sourcing agent and a negotiation agent. Each one with its own model, tools, and instructions. There's also a model gateway built in. So I put a cheap, fast model on the groundwork and claw on the hard reasoning. Second are plugins. I dropped the sourcing plugin into my agent and told it, "Find suppliers for custom t-shirts, 200 units, sublimated print, and under $12 a piece." It pulled a short list straight from Alibaba.com, actual transaction data, real ratings, minimum order quantities, prices, lead times, not guesses from some random training data of the model. Third are connectors. I connected my Gmail. So, the negotiation agent emails those suppliers, pushes on prices and comes back with a comparison table. days of back and forth done while I watched. I connected Instagram, too. So, when the merch is ready, it creates and posts the marketing content itself. Anything sensitive like payments needs my approval first. And you can tell it to keep your data off their servers. Fourth, channels. I can fire off new tasks from Telegram while I'm at the gym. And I put the supplier research on a schedule so it reruns on its own and just pings me, which is very, very convenient. So, while most AI tools just hand you a plan and then bail on you, this one is a doer. It actually does the work. Axio work basically has open close style capability pointed at running your business. So download Axio work for free and unlock the bonus credits. The link is going to be in the description. Yeah. So first they should start creating some data sets. That's literally what I had automated. I created a minutely checker when Fable got banned to check if it gets released on a VPS. And so I have like a lot of different VPSs from Hostinger. And one of them I set up just for this purpose to create a data set. I created 400 different questions that I wanted to ask Fable and it was checking uh the API every single 60 seconds and the moment it got enabled the data set populated and I got all of my answers right so even if they took it away now I can still read for these answers another thing is for all of your projects and areas of life creating long-term plans right your fitness your your business your dating your health your you know injuries whatever create long-term plans of how to achieve it and how to get what you want then I would begin creating skills mod you know when you're going to be using Opus or or GPD or you know Deep Seek or GLM when you're going to be using less powerful models than Fable have fable write skills which is basically SOPs processes systems how to do certain things you know how to do copyrightiting how to send an email how to refactor your codebase how to review PRs whatever you are doing turn them into skills write those skills with Fable and then less powerful models can perform better than they normally would because they're following instruction instructions from a better model. >> Do you think open source catch up to fable? >> Yeah, I think it will catch up. Question is how fast like fastest will be probably two to three months. More realistic maybe five months, six months. It all depends on you know how fast people can distill it I guess because that's one of the main things. You take a better model, you have it answer series of questions and you train a smaller model on it. So everybody can like actually do this you know everybody can contribute to this. There's already over 100 different data sets on hugging phase of fable being distilled. And I suppose now after people know that it models can be banned. There's going to be hundreds if not thousands of data sets. So you can literally go and ask GPT like how to create a data set and upload it to hugging face and then you can ask fable some questions in your own domain and contribute to the open source community by uploading more of these data sets. So I'm not encouraging anybody to do that because it is against enthropic terms of service but this is how the open source will catch up. >> Okay. Okay. So what what was your idea in the last time we spoke about the open source data sets basically people contributing their data for >> yeah I mean I had idea for like a initiative where I would build a like some peer-to-peer network or some like pooler I don't I don't have the basically people could contribute their limits right so when you're not using your subscription you could contribute your subscription to answer a series of questions that are that are said from these open source labs the issue with that is that still if you're encouraging like in my case hundreds hundreds of thousands of people to do that. It is risky against entropic, right? So, I'm not going to be encouraging people to do that. I'm just going to, you know, give them the information like hugging face, you know, how to create a data set and if they want to do it, they can do it themselves. But I'm not going to be like the middleman of distributing software of how to distill because it's again terms of service and again you need to follow the rules. >> Let's just say open source does catch up to Fab 5 GD 5.6. Would you be worried about your cyber security? No, because you can use the models to improve your own security. Like this, you know, Dario Ammoday is like, "Oh, open source is dangerous." It's like so many memes. It's like local McDonald's manager says that cooking burgers at home is crossing a dangerous line. You know, local gym owner says that working out at home is going into the dangerous territory. Like people were clowning him on Twitter so much. like yo Dario he he's like quickly became one of the most hated people in all of AI because of his [ __ ] dumerous stakes and his anti-open source stance is ridiculous right so if you want to improve your security use the best powerful model to do that right so if somebody else has the same model and they want to attack you you have proper authentication proper middleware proper RLS policies all of that stuff >> I mean doesn't this also make it more dangerous if if you're not allowed to basically because like I've run into a lot of issues where I'm trying to make something more cyber secure But Fable doesn't allow that, you know, because it's dep whatever >> which I mean you have a skill for that which we can touch on in a second. But >> if these models are restricting you from that and someone does have a powerful open source model and they are able to attack you with it, isn't this almost doing you a disservice by not allowing you to protect yourself with Fable? >> Yeah. Because they're taking it away from the non-technical people. If somebody is, you know, this kind of Russian hacker, you know, stereotype archetype, it doesn't matter if it's from Russia, China, Iran, North Korea, America, Israel, doesn't matter. If it's a powerful hacker that knows perfectly how to use the terminal, knows everything about Linux, knows everything about cyber security and, you know, networking and hacking your Wi-Fi, has a flipper zero and 50 other hardware gadgets. That guy can take, you know, GLM 5.2 and opensource model. He can fine-tune it on his own data set or on some illegal data sets from the dark web and he can create something that's more dangerous than fable and more aggressive than fable and like unhinged and unrestricted on cyber security and hacking and he can absolutely destroy you. So I think this is just crippling the regular people. They're just they're just like making it harder to escape the permanent underclass. >> And also the bug that Fable found was also found by other models when they were searching for it. So it's not like it was just fable that found this >> because we have totally incompetent people making the laws, right? The people in the US government know nothing about LLMs, next token prediction, matrix multiplication. They they don't even know what transformer architecture is. You tell them transformer, they either think of like some LGBT or they think of the movie like they don't even know it's AI term, right? And these are the people making the policies. So I mean in USA it's not that bad. I would say the USA lawmakers are more competent in EU is a total clown show. Like can you imagine Mundlean making any AI decisions? Like I would just rather have Fable. I would have rather like Llama Free govern me than these EU bureaucrats. Oh, by the way guys, we're almost at 400,000 subscribers. So please go below and subscribe. YouTube Studio says over 60% of you are not subscribed. So check below the video and please subscribe. We're almost at 400K. >> Do you think Antropic is putting spyware into cloud code? Yeah, they just admitted to it. I mean, this is not even a conspiracy theory. They literally put different classifiers, the different u like it was timestamping the time zone. So, people in the Chinese time zones were getting secretly monitored by anthropic to prevent distillation, right? So, Enthropic there's like super basically what they did is they distilled all of humanity's knowledge, right? I I said it in the last video, but I need to say it again. They scraped all of the web, all art, literature, code, history, everything, distilled it into their own models, and now they're selling them for profit and using them internally. And they completely freak out and call it a distillation attack. Anybody anytime somebody uses their models to create some data, right? So, it's a complete like hypocritical statement. >> Do you have any way to not get restricted with fable? >> Yeah. So I created a skill for this because the first time it got around I I needed to adopt my prompts because I said like you know how do I make my app more secure which is a perfectly fine prompt but with fable it's not a fine prompt because secure can mean cyber security and if you say like are there any you know vulnerabilities in the code boom you can downgrade it to opus so I created a skill named fable safe prompt and that skill basically takes any prompt rewrites it to be fable safe right so I did extensive deep research with multiple models on what people are reporting and like the tips they were sharing. So when Fable came out after like the first 48 hours, it basically I scraped like dozens of tweets and you know dozens of reports of people complaining about this and saying how they solved it and create turn this into a single skill so that when your prompts start getting refused, you can just run your prompt through the skill. Literally get the same output but not get blocked and still stay on Fable. Yeah, it's going to be the second link below video. It's going to be completely free. Grab it now. >> Do you think Fable would will get banned again? Probably not because then they're just giving the keys to the open source community and right now that's mostly China. So like if the US government wants to lose to China, the thing they should do is regulate their models, right? And ban Fable, ban GBD 5.6. That's how you give the free win to Deep Seek and GLM and Kimmy. >> So you think GBD 5.6 is going to be unblocked next? >> Yeah, it's probably already unblocked, but the opening it was on vacation. So you know, this is why you don't take vacations, I guess. or at least when you go just take your laptop and do a workation. Uh but yeah, I mean it's coming in the next few days almost certainly otherwise Enthropic is just pulling ahead of OpenAI and they cannot afford losing the lead. So if Fable 5 is really that good at coding as you say is SAS dead? >> Is SAS dead? So for those who don't know SAS means software as service I mean look it's hard. It's a very complicated prediction. If anybody knew the answer to this, you could become very rich. You can literally go to the stock market, short all of the SAS companies and make [ __ ] ton of money. If you think the other way, you could go long on these companies, buy their stock, and you know, build build your SAS. Like, nobody knows. You know, the future is so uncertain. I think a lot of people on Twitter are like expert analysts. On Twitter, you can just post stuff. It doesn't matter if it has a source or not, right? So, the honest answer is like I don't know. Nobody knows. Honestly, there's argument that it it's going to become dead because anybody can build any type of software. But the hard part is not building it. The hard part is maintaining it, right? So like when it crashes, do you want to go and like go fix some bug, you know? Okay, then the mobile experience is not that smooth. Do you really want to context switch and go and like redesign the mobile? Even if it's like one or two problems with fable, still you you have to manage all these softwares. A lot of people would prefer like a clean managed solution from people who are only focused on this one thing and they're experts at it, right? Like I don't know, you you say like Riverside, we're using that to record podcasts. We could probably rebuild that, but like I don't want to be bothered. I'll just pay like $20 a month, $30 a month and uh just use software from people who are like obsessed over this one vertical >> cuz you are building a you're in the process of building a startup. That's what you're doing with Fable. So what is your mantra for moving into this direction of building a startup? >> Oh yeah, I think the biggest takeaway is that everything will be used by agents, right? So that is something I'm absolutely certain about. I don't know about the pricing model whether it's price per token you know monthly subscription or or you know per execution per result but I'm certain that like humans will not be the majority of software usage it'll be agents so if you're building something that requires people to go to a website sign up adopt a new app into their workflow and change their habits like you're in the past you need to build stuff that people can use through codeex through cloud code through cursor through Hermes agent If these agents will struggle to use whatever you're building, you're not building for the future. >> Can you elaborate on that? So you're basically trying to scale with the tokens. >> Yes. You need to go from first principles, right? So whatever you're doing in life, especially when you're starting a new startup, which is an incredibly difficult feat and is most likely going to fail, you cannot be thinking of like analogies of like what happened in the com bubble or blah blah blah. Like it's useless. You to think like what's happening? What are the things that you truly believe in? And I truly believe in AI like I'm a true believer. A lot of people are not true believer in AI. And that's if you go underlying statement below that it's in technology. I don't think technology can be stopped, right? Even if it gets banned in USA, people will build it in China, in Europe, maybe in Brazil. But like technology cannot be stopped. So that means AI will happen. A lot of these people who are anti- AAI, they just don't understand history. You know when you there were a lot of like the I think it was the Ottoman Empire, they banned the printing press and that was like one of their biggest mistakes. like they just ban technology, right? >> So basically my question like since tokens are scaling up, the token cost, token spend is going up, it's a safe bet to go with the tokens essentially. >> Yeah. So yeah, basically the the technology keeps improving and that's one of the first principles thinking, right? It's one of the things I'm certain in then if you extrapolate that. Okay, what does that mean? AI agent. Everybody wants to talk to their agent that knows them, that has their preferences, that like has their favorite model and that agent can take actions. If you The reason why we're using this software is to get something done. It's not to use the software just because you like the UI of notion or you want to click around Obsidian. You use it to get something done, right? So if you can just send the intent right now speaking is the fastest way because you know speaking is like three to five f three to five times faster than typing depending on how fast people can type. Maybe in the future will be like some neural link or before neural link there's going to be like meta released this recently. I haven't looked too deeply into it but it was some like non-invasive brain scan. So like you know some device similar to the meta glasses. You can just wear it on your head and you can read your brain waves without drilling a hole into your skull because a lot of people don't want to do that right. So right now speaking is the fastest then it's going to be that and you need to convey your intent to like basically if you think about what is the most powerful entity you know if you had to describe God it's like something that executes actions instantly so you want to you have a thought and then it comes reality instantly right so the the faster you can do that the more powerful you are with an agent you can tell him do this and the agent can you know generate tokens way faster than you it can use software way better than you navigate it debug it so as the models get more powerful the agents will become more powerful which means more and more tasks should be done by an agent. If you want to, you know, find the 50 best video editors in Poland, like doing that manually would take you many different hours, right? But if you did a SLGO loop in Codex, you can set that prompt in literally 30 seconds, come back later, you know, half an hour, maybe an hour, and you have the list. You know, that's insane multiplier. It's literally like it's it's not like a 2x, 3x for that task. That's like a 100x productivity increase. So all software in the future will be used primarily by agents. Sometimes it makes sense to create a nice dashboard, nice web user interface so that humans can input tokens more efficiently. But again think from the first principles. It's all about efficiency of output tokens. So like that's speaking and input tokens. So like how can you see what's happening in your life in your business more efficiently? If you have some like list maybe that's easier than like agent sending you a bunch of text and you have to look at some markdown, right? So there is still going to be web UI. User interfaces for humans are still going to be relevant. But soon enough 98% of all software will be used by agents and people will just tell their own personal PI agent their own own personal agent zero their own personal clot code hey I want to do this do that and they give the task to a clot code and the clot code calls seven different tools 15 different APIs you know browses websites scrapes some results and does the task. So whatever people are building it needs to be usable by agents. >> Okay. And so how are you doing this in your next startup? How much do you want to reveal? >> Yeah, like it's not ready to be re revealed publicly but uh basically I'm betting on again token growth going exponential. Even though the token spent might not go exponential because people cannot afford to spend infinite money and they're going to optimize for open source models and more efficient models. The number of tokens used for each person is absolutely going exponential because that's just intelligence, right? And you can make everything better with more intelligence. Example is a traffic light on the street. Every single day driving here in Katvitz, I go to a traffic light and it's red and the other direction has no cars. If you had a little bit more intelligence in that system, you could skip one loop and go green, right? And it it would save everybody time. And that doesn't even require an AI agent. That requires maybe a few sensors and some like if else statement. Literally a few lines of code. So that's an example of like you can make everything better with more intelligence. So yes, token growth will absolutely continue. AI agents will absolutely get better and u if you're building something that can only be used by humans and that like you know it's like traditional view of SAS is like people need to go to a website they pay a subscription they use it they you know whatever but you need to be building like CLI tools clean API clean documentation so that it can be used by agents. >> Do you use Fable inside of the cloud web app? >> Not really. like sometimes if if it makes sense you know on a video like asking a quick prompt but mostly I use it inside of cursor and inside of cloth code in cloth code you can do ultra code which is u like extra high with the dynamic workflows and then you can do / goal as well so that's the main reason I still use cloth code yeah I mean it's insane yesterday I had it running overnight with the caffeinate you know command on Mac OS so my screen doesn't go off and it was running for like two three hours implementing massive backend features end to end with testing. That's another great thing about Fable. It loves testing. You tell it to make a small change. Other models will just make the change and they wouldn't check if it works or not. Fable you don't even if you don't have to tell it to test it. It it will just run 70 90 different tests to make sure it's actually proper. So this is absolutely incredible. Working with Fable is is just amazing. So you had Matt PCO on the podcast recently and he suggested that well his idea was basically that everyone's skills are just overloaded and he suggested deleting all your skills starting from scratch and you know adding back the skills that you need to >> should you do this with Fable and do you agree with his point? >> Yeah, I think right now is actually the best time ever to do that because if you think about what skills are it is instructions for these a for these models to do something specific that they cannot do by default. Fable is a step change. It is significantly better than previous models which means it doesn't need as much handholding. So you should like right now it's the best time ever to delete all your skills and see what Fable can actually do by itself. And again you will add something back. Right? This is what Elon says. If you don't find yourself, it's one of the steps in his fivestep algorithm. If you don't find yourself adding back at least 10% of the things, you are not deleting enough. A lot of people have like bloated systems. They have tons of weekly meetings in their calendar. tons of these like habits and traditions, most of them are useless or obsolete. You need to be deleting stuff from your life, from your business, from from your promps. And yeah, with skills, it's the same. So, I would encourage all people watching, delete 80 to 100% of your skills and see what you actually need because Fable is very competent and maybe it doesn't maybe you don't need all these skills. Maybe you only need near. >> Are you still interested in fine-tuning now that we have Fable back? >> Yes, I'm more interested than ever. I mean, the fact that they took it away from us and then prevented GBD 5.6 from being released, like it's too like warning signals from the heavens for us to take local models seriously. So, actually like we're setting up a lot of stuff here. You know, I purchased the NAS, so I'm going to be downloading every single data set from Hugging Face. Uh, this probably not not going to be enough. This is only 40 terabytes. So, we're going to extend it to 80 terabytes and then uh probably buy another one. There's a lot of data sets that are massive, you know, especially video and image data sets. But yeah, I want to have a backup of like all of the main models and all of the weights. So any model that's significant and that's open weights, you know, like Kim, like Miniax, GLM, I'll just do a I'll just download the weights, you know, even if I don't have the GPUs to run the inference, I just want to have the weights myself. And the data sets are even more important. So I'll start with these and they're smaller, so we can already start. I also have a VPS. Um, again, Hostinger gives us a lot of VPSs. So one of them is literally just downloading hugging face data sets that are below like one gigabyte. So that VPS already has hundreds of different data sets. So it runs like every 15 minutes and downloads some new ones and it's sorted by the most downloaded ones. And again guys this is a single prompt to Fable or or okay Fable is probably going to reject that but you know Opus or GBD 5.5 single prompt to Codex. You can say set up a script, put it as a launch job that every 15 minutes checks hugging face through the hugging face CLI and downloads one of the most downloaded data sets that is less than half a gigabyte large and start building your own copy of these most important data sets right self-hosting is absolutely essential more and more tokens people if if they like if every month yesterday I used like 230 billion million tokens 230 million tokens just on cursor So if you, you know, if you add up like the cloud code usage I had, it was probably half a billion tokens across everything just yesterday, right? I need to be moving more and more fraction of that to to either local or at least like self-hosting, right? Like spinning up your own GPUs, but ideally self-hosting your own cluster. So I I'm fully committed to spending like $50 to $100,000 over the next six months to build a proper cluster. we're going to, you know, rent out like build out like a small data center here in Kat Vita and we're going to over time make it larger and larger because for a lot of these use cases, you don't need Fable. You need Fable for the planning, right? Like it's like a a company. The CEO cannot do everything. The CEO shouldn't be the janitor, right? But like if there's an issue with the janitors and like hiring the company that's providing them is bad, maybe the CEO should create a plan for this, right? That's the same thing. It's like running a company. People need to think about that when you're managing these agents. You are like a like a chairman of the board and you need to architecture the company in a smart way. So if you need the most powerful model which right now is fable, ask it for like how it would architecture and organize and orchestrate the other agents. Ask it for long-term plans and vision and you know analyzing competitors and what features and like product decisions, things that really matter. Don't have it do stupid stuff and you know waste your limits on like fixing some front end button. You can do that with KimK 2.7 for like 25x cheaper than Fable. >> So if these models keep getting banned and unbanned, how do you build a business on top of them? Or do you? >> Well, you build a business by using these models to build a business, but the underlying models should be open source cuz like nobody can ban GLM 5.2. You know it if you just go on open router, it has like 30 different providers. So if somebody like takes down fireworks or together AI or you know GMI cloud it doesn't matter nobody will notice you know people will just route their traffic to a different provider. So if you're building your own business you should definitely invest into self-hosting and you know spin up a local cluster and host a powerful model locally for your employees. Also your data will not be leaking right. So your data will not be sent to Dario Amod or Samman. Nobody trusts these guys. Let's be honest. What's important is that you use the right model for the job. You don't use small models like you know 20B model or 30B. You need to be using large models and for that you need large GPUs. So you're going to invest the money either way. I would say that you know you're going to pay it either way. Either you get out competed by somebody who uses more AI and then you're going to lose your competitive advantage in business or you're going to spend it on cloud. You know whether that is open subscription, cloud subscription or open router credits. You're going to spend that money either way on tokens. So might as well just invest in your own GPUs. Buy Nvidia Spark. buy a better, you know, MacBook computer, more RAM, more VRAM, better GPUs, and make that investment into owning your own intelligence because in the future, it's going to be the most important resource on Earth. >> So, last question, would you pay more for Fable if the price doubled tomorrow? >> Yes, because it's a step change, but then again, GBD 5.6 is coming soon, so we'll see how long Enropic can do these prices. The the beauty of competition is that it drives everybody to, you know, do better. If there was no open AAI and Froic would just have a monopoly, but GBD 5.6 is coming. Everybody knows it. It already beat uh Fable and Mythos on many different benchmarks. We'll see how easy it is to use. You know, that's the main thing. Like benchmarks are are just one thing. They matter. A lot of people like hate on benchmarks way too much. Benchmarks and EVAs absolutely matter, but what matters is how good the model is. You know, it's it's like a hiring an employee. If they have the best application, the best CV, but then working with them in person is kind of weird and, you know, unproductive, you're probably not going to hire that person. It's the same with the model. So, we'll see how good GPT 5.6 Soul Ultra is. Whoever is naming models at OpenAI should get fired on this list. It's terrible. But yeah, um I would pay double. >> So, where can people find the fable safe prompt skill? >> Yeah. So, again, we're going to put it as the second link below video. It's the Fable save prompt. Any prompts that just use this on your prompts and they won't get rejected by Fable. I built it the first time Fable was released and again you can get it completely for free. Second link below.