#61 Free ChatGPT, Muse Code, Qwen3.8-Max and AI progress about to stall

We're live. Hello there. Hello there. Hello there. Um, do the old the classic volume. Oh, I haven't even set up my microphone. I've got to do the classic that I also put on my head. Do this. Do you reckon one day we'll do this in person? Yes. Command AI live. And alive. Alive. Um, let me let me go to YouTube. and do a little volume check. Mic check one, two. Mic check one, two. You're really good. Nice. Okay. Maybe maybe I'm a little bit louder than you, but I don't know. Uh anyway, let's let's crack on. Yeah. Let's crack on. I've got I've got a cool new tool to share with you guys. Um, you've heard of Notebook LM, right? Google Notebook LM. It's kind of like it's kind of like for those who don't know, if you're a student, you'll know this where you've got like so many sources. you you've got like a PDF, you've got like a YouTube video, you're learning something and you you want to kind of bring all of that stuff together and just talk about it, query about it, like create documents, create charts and diagrams and whatever. Well, Notebook LM is actually the tool for that. Maybe I'll even note Oh, I can't do this onehanded. Hang on. I should have had this prepared. on me. So, this is for students or you guys? I would I don't know. You you tell me. I've used it. I used to use it very very early doors when it came to like my experience using AI. Um, and I kind of just used it to just I would I would interrogate a transcript, actually, a multiple multi-page transcript. In hindsight, I probably did didn't need to to do that. But let me just log out because I'm not showing you my Oh, I can't. Bloody bloody Nora. Can I go? Can I sh Oh, no. I need to log in. Okay. Well, I'll show you notebook anyway, just cuz um yeah, I haven't got anything too too secret on there. Let's go. Oh, no. Jupiter and draft market research. Right. So, I've got all of these different inputs when it comes to like research and and whatever. And then in the center here, I've got a chat window that I can So, I was just doing a bunch of market research and I pulled together loads of resources. You can pull in like web, you can pull in drives, you can pull in um even YouTube videos and stuff like that. And then you can generate and what what was really nice is you can generate um a podcast that pulls together and understands all the information. Really cool tool uh even flashcards and stuff. For me, this is a great study tool, right? But it's Google and um if you're outside of the EU, you know, they're probably like sucking up all of your data. Well, I stumbled across this basically a free clone of this tool. It's pretty cool. It's called Open Notebook. Wonder if this actually has a Oh, it does have a a website. And it's essentially the same thing. You've got a podcast generator, AI powered notes, privacy control, and it does everything that you want from uh note. Oh, they've renamed it Gemini notebook. I've just seen there. Uh it's got everything you need. So, I thought I'd just show you guys what it's about. Get it all set up and probably I'll what I'll do as well, I'll show you how to get it set up with like a local AI because that'll make it super super private because this works with chat g uh open, anthropic, and all the rest of it, but what what about getting it set up with local AI. Um, so basically, you need Docker installed and you need it running, which I'm going to get in the background here. Do you know about Docker? Do you know what Docker is? Yep. Heard of it? Heard of it? It contains basically everything your app needs in one [snorts] place in one image if I'm not Exactly. Exactly. So you can update it in a go or like upload it to from a server to another. Right. H you can also just have um just your database just a database in here and then your front end app that isn't inside of Docker connects to a database inside of Docker. So that's quite a nice thing to do and you can mount it. So yeah, easiest way to do it is just to copy this and paste it into your um terminal which I will wait until it actually loads this time because it for some reason is it just me that it just takes ages to load. Uh, someone could you see that chat message? Yeah. Yeah, but we were like talking about something else. That's why I haven't mentioned it. But I can't see any I actually can't see any chat messages. I can only see that one. Well, there is just two. Yeah, I can't I can only see one anyway. Um, we'll get to it. We'll get to it. Cool, cool, cool. Um, so yeah, from here's I think this is the thing that's people sometimes people get a bit confused when you have to be inside of a folder in your terminal. Um, you don't necessarily need to in this one because you can also just run piece of code like that and it will just install it anywhere. And once Oh god, I hate all this bloody switching of Windows. Let me just double check where this is now because it's installed in Docker. I'm just double checking. Is this downloading? Yeah, it should be downloaded. Uh so for people watching asking uh this is an open-source and free version of uh the Google Gemini notebook or notebook LM uh has different names that the go-to studying tool for students and this is installing uh a free version of that free open source and local it is everything you want um on your computer. So people are asking what we are doing. Nice, nice, nice. Yeah. So this downloaded basically this downloaded the relevant stuff into docker. And then if I just do uh docker compose up then you'll see it should do all the business. Oh, there's an error. Found port is already allocated 8,000. Why does everyone use 8,000 all the bloody time? I I need to make an app that kind of like delegates the right ports to the right apps. There's so many I have this like everyone is using 3000 with me and maybe I should add it as a system prompt that check the freaking like ports because it like AI can check the ports, right? Um yeah, AI can check the ports as well. Yeah, let me in case you're not using it. Yeah, let's do this old school then. What I'm going to do is I'm going to CD. Um, let's go back. Let's go. Let's see if I've got it because I think I've already got it. This is the problem. I think I've already got it. Use um -p 2234. -p2234. What's that? I have no idea. Okay, there it is. Right. What I'm going to do, so use the command line prompt to see the the prompts to see the ports probably. Uh oh, no, it's okay. What I'm going to do is I'm going to get [laughter] clone. Yeah, Sam is the developer. I'm the guy who pretends who knows development. What I'm going to do is I'm just going to get clone the repo because this way we can actually set the Well, actually, this is the annoying thing as well. If you get clone the repo, it actually has a different port. So if we go to cd open notebook there and then we can just do Oh, sorry. I'm copying some scripts off screen like this. Docker compose up. Oh, come on. 8,000 already allocated. Um, so you can change this in the what was it called in the is it here the docker compose right what we're going to do we're going to just do oh my god this is really annoying okay we're going to just change it to 809 I don't know what is using port 8000 but we're just going to change you could also write a command kill that one probably. Yeah, but I don't know what it is. I don't know what 8,000 is. Yeah, you don't want to. So, we have a developer in the chat. Yeah. Uh OG Ashot Gupta. Uh I hope I didn't pronounce it wrong. Join us on YouTube. Like he's uh on Instagram. Instagram like you see a tiny window. Join us on YouTube. Uh just search command AI and you will find us there for better quality. Uh yeah, he's a developer and he Well, he's a developer. He's helping also with the port. Uh what is he saying? So I wasn't listening. I wasn't listening. What is he saying? Well, the the the dash the command dashp and the port I think P not D. No. What you're saying here? Yeah. Yeah. Yeah. Yeah. Uh thing is [snorts] now. Anyway, oh I know why. I actually know why this will get fixed in the edit for sure. It's it's surreal DB, right? Which okay, I've already like I need to delete Surreal DB because it's already running. Ah, okay. It's it was its own service that hadn't been deleted that I uh you obviously wouldn't have this issue because you wouldn't have already you we you won't have already installed it. So if I just do compose up again and then if I double check the ports and this is the part where I do claude dash dash dangerously skip all permissions and then tell it to run the server. Uh, Fable Five on Extra, but I feel like you get you just get lost then. Yeah. No, I'm kidding. Like I'm I'm getting to do more and more myself, but obviously sometimes it just I just ask to do it. Fair enough. Fair enough. It's fun to learn. Like I I started to push also like the updates to Versal myself. I used to just tell the agent to do it. It's kind of fun. like the agent is working and I'm like doing the last uh command and it feels like I'm doing the work which is totally untrue. [laughter] Well, annoyingly this will get fixed in the edit, right? Obviously, but like if the what what what was happening right there is that because I had already tested this obviously I just hadn't deleted a Docker thing. So, it was already running the service it needed. I just needed to delete that. And you should just be able to get clone. Uh no, you should you should even just be able to run Docker that Docker script that we used straight away. Uh, let me just really quickly show you. You could just you should just be able to run that, but it's because I was already running it that was Okay, so cut to the edit. Nothing happens. I just run that code and then the edit will just show you me coming to here and it's essentially everything that you want. You basically can uh set up a bunch of sources. You can create a new notebook. Let's call it, I don't know, command AI uh re research. And then in here, I can add a bunch of sources. And these sources could be a YouTube video, it could be some PDFs, it could be I could handype it, do whatever. Um, let's actually just add the website. And that's just processing the website then pulling all the information down off the website. If it was a YouTube video, it actually download the transcript as well. And what's quite cool then if we go to models here, you can then choose a model to or at least a provider to be able to configure. So I'm going to add this is how to get something local running up. I've got OML X and 1 2 3 4 is the API key. OMLX is if I start the server running on in fact I'll copy paste it um 806 add that and this is running on your laptop. Yeah. So let me load up a model. Now you'd need obviously like a way I don't know whether you need to see what I'm showing you right now. This is OMLX and you basically search your local models. If I Let's do inkling small. Um how do I even load this now? Why can't I just load that? That's really annoying. It's kind of fun you're figuring this out and I'm just chatting with our audience. [clears throat] What am I not seeing here? This is Why can't I just load the model there? I'm going to do it a different way. What is OMX? OMLX is is a tool for downloading and running local models. Ah, okay. It does everything like for you contained. Yeah, that's nice. Yeah. Um I'm wondering why I can't just Oh, okay. You have to do it from here. That's a bit of a UI thing. Uh so what I've done inkling small. If I load that, which is already loaded um in open, I should be able to add that now. and then select it as a model. So when I go into notebooks command AI now I can just chat with it and ask it about like I don't know um what's the podcast called and this is could not connect today provider check your network and connection provider URL are we Okay, this is the magic of live demos. Yeah, I I have a feeling and I wonder if that's just not loaded. What do I think does not fit under dynamic memory ceiling? Okay, maybe inkling small is a bit too extra. Should what would be a really small This will do. Loaded. There we go. Okay. I'm going to I'm going to add that one into the into the memory into the thing. So if I go here models, add should just be able to do that. And then notebooks. Cool. And now say what's this show about? [clears throat] Oh, come on. [laughter] [gasps] Oh, I did this earlier and it was uh fine, but I'm sure I'm just rushing through something. Let me just double check that we've loaded it. We might have to bail on it. Yeah, it's loaded. might have to b this boys and girls. You might have to figure it out yourself. But this is local model anyway. Like if if I was to sign in through chat GBT um it would be it would be fine. I'm just trying to get this loaded. All right. Detail. Okay. So, actually that's true. It's not, for some reason, it's not picking up my um local server. Okay. No, no, it is. It is. Test. Cannot cannot code server. Is URL correct? That looks that looks right. Do I need to do that? Okay. Maybe it's that which I'm pretty sure I put in. Okay, we're going to screw it. We're going to bail on that one. That's annoying. But add your OpenAI API key or whatever. Go into your notebooks. Add all of your sources. You can create notes. you can query it. Really, really cool study tool and it's free. So, I don't know. That's my tool of the week. I'll probably do a more in-depth uh tutorial on my own channel, but like I think this is a really cool way to get a pretty good tool just open source running on your own computer. Cool. On with the news. Yeah. Uh, let's check what else do we have? Well, well, you do you want to do the intro? Well, kind of just started. That was kind of funny. Okay, cool. Yeah. Yeah, we we just started and maybe that worked in our favor. We We have some viewers. So, let's let's continue with the EU one. Do you do we have a question? Sorry. So, M Mc Men was saying that are we reading chat? What was there a question? Um, we did have some questions. Yes, but I don't think anymore. Uh, we have them anymore. Uh, let me give me a second. I was preparing the EU thing. Oh, you're talking about this question here. Is your channel related to Command Code or something like that? I think it is. I don't know what what's Command Code. I have no idea. Okay, you might want to rephrase your question. I don't know. Claw like Claude code. Is that what you're talking about? I don't know. I'm sure someone in my comment section asked if if I was the command code guy, and I was like, what's command code? [laughter] I I I have no idea what command code is. Um, we have copycats already. Have you tried Deep Seek V4 Flash? Yes. On my channel, I've literally just done a whole series on playing around with it, including like the new version using DS Dwarf Star. So, my channel's 0x50 or you can just search my name Samuel Gregory. I've looked into Deep Seek all all about that. Um, yeah. Uh what is free here? Um he's referring to the tool that you were showing. It is open source and yes you you run you can use models that are running on your own computer. So it's both the software is free and the model that you use uh that can be also free depending on Yeah. If you got if you got good hardware. Yeah. Yeah. this super computer, but like if you mini, the model that I just ran, I think it's like 6 gig. So, if you've got at least 8 gig of RAM or 16 gig of RAM, you can run that model. So, put it that way. You don't need a superco computer. But, um, yeah, otherwise you can just link it to like a paid API thing. So, that's pretty much everything that's free. Cool. Should we talk about Quen? Well, we have the the news the EU. Go on then. Go for it. Yeah. So, starting now, if you are posting an image with made with AI or edited with AI, EU um asks you to put a label on it, a big AI modified label, very clearly, very visible on the image. [laughter] It's just so dumb and it just gets worse. From uh August 2nd, EU rules required disclosure of professional professional we'll talk about that professional use of generative AI if content can be mistaken for real people, places or events. So this is essentially images that are uh realistic, but I mean grandmas might see it and think it's realistic. So that's not really defined well at all. Text on matters of public interest. What the does that even mean? I have no idea. Like yeah, sure. Okay. What? But anyway, um had no human review. So if your text didn't have human review on it like you just generate text and don't read it and just post it has to be labeled as AI uh made and user is interacting with a chatbot. I mean chat bots have always been chat bots like yeah but now you have to mention that it's AI whatever. Uh and this is an example. Uh we will go through the the writing but this tweet was quite funny. If AI agent uh computer use uses my Photoshop software, is it technically AI made image? Uh what if I use Photoshop to create a very good good deep fake but it's not AI? I don't have to label it. Got it. Which is like yeah like the whole point is just so dumb. I understand the good intent of EU trying to you know regulate this but I don't think this is really doable uh or at all. It's worse than the the whole story with cookies. They are trying to regulate the out of AI but I don't I don't think it it can be done. And it's quite funny like you can come here and read um like the whole thing and how there is a basic icon fully AI generated or partially AI modified. I don't get this like imagine I I take a picture of myself or anything and then there is a trash can. I want to remove it. If I remove it with Photoshop with the stamp clone thing, that's fine. But if I do it with AI, that's not fine. like where is the line? And and then we have the question with who's going to enforce it? Like like how the do you want to enforce like million potentially literally like actually in billions of images soon like you know the dead internet theory which you know we don't like that but it's probably going to happen. It's happening at some point maybe already like the the viewers we have on Twitter are bots. Uh the the people posting on like many like different types of social media. We have a lot of bots. Uh already bots posting bots reading them and now EU trying to regulate AI generated images and text. I don't know if they can do it. That's that's that's basically what I'm trying to say here. And I don't think it's making any sense. I understand the intent, but still like you you you cannot you cannot like have a image made with Photoshop being fine, but some made with AI not being f I don't get it. Like that's just doesn't make sense. It's like what's the what's the intent? What's the like everyone's got their own different reasons for disliking AI like I just don't I don't know. So the intent is um well if you have people who might mistake a re a fake image for a real thing you want to prevent that and AI is making that easy to be done easier than ever. So we had we had a festival here in my city. It's like the biggest yearly festival we have uh called Anna Fest in my city and it's like I I think like half a million people visited every year. I think it's something like around these numbers and there are like many of these like smaller shops or like mobile shops like I don't know what they are called like when you go and eat something and you buy something and they're like these kind of shops. Pretty much every single one of them had AI made images. like of the food and I can tell because I can tell the you know the chat GPT very like unique style of like making up the the pixels I could like easily tell this is AI generated and it's kind of ugly but maybe it is just to me ugly because I'm you know I have I'm a designer I do design I care about details but maybe for people like my parents seeing it and they they see it and they're like oh That's that's like the list of food that they have. It's better than not having an image. So all of that was AI generated. So does that mean that these people have to put AI modified on those images printed on their shop? That's e the new EU law. So it's kind of dumb. Uh it's kind of dumb when you think about it how much AI is being used and how much it will be used in all formats like books uh prints digital non-digital like yeah how do you want to enforce [snorts] this is and and you're also asking people like the people who are the people that this most uh mostly applies to are those intentionally trying to be deceptive and you think you're going to be able to tell those people who are intentionally going to be deceptive to Yeah. follow along. No. People, if there's one thing I've learned, especially in recent months, maybe years, like people online will do anything to deceive you, right? They will try and bypass any kind of rules or terms and service. Like, no one's going to listen to this. Absolutely no one. Plus, it ruins any sort of design guidelines or anything like that. like and and you know you might have photoshopped the actual image or hand drew the image but maybe the research that you you did to come up with the image maybe the initial sketches was AI model like do you know what I mean like where does it stop yes this is uh what you are talking about is exactly the conversation about Hank Green and there was um a conversation on Colleen and Samier podcast about that as well and they talked some Gen Z creators they have on their team. Um, and one of the guys mentioned maybe the line is where the user doesn't notice like where when is it okay to use AI and when is it not like you know if we use it with good intention and create something good with it and like I know you could go into that like subjective but if you even like define it like between us that it's good um and the user doesn't notice it is it fine fine, but if it's slop and they notice it, it's cancelellable. It's kind of dumb. Yeah. Uh you cannot prevent AI. That's my point. And obviously we're advocating the right use of it. And you know, we we are you're a developer. You're all for quality. I'm a designer all for quality. And we use AI both. It just doesn't make sense. Try to regulate the out of it and do this. This is like a really good [laughter] comment. Like this is fixed so make it very clear. Yeah. Yeah. I don't know. I think it's ridiculous. I think it's absolutely ridiculous. Good luck. I mean, it is annoying. Don't get me wrong. I I sympathize with the ambitions, but I just it's just uninforceable at this point. people people are going to use AI for all as long as they can get some sort of gain or you know financial gain or whatever they're going to use the shortest quickest path to it. So I don't know. Yeah. And same with this is the exactly the same thing with cookie banners. They ruined the internet. They ruined every website for what? Did it prevent Facebook from like stealing? Did it prevent AI from stealing all the books from, you know, every website ever? Not really. It's Yeah, it's kind of dumb. Oh well, can't blame him for trying. Yeah, [laughter] you got to do what EU got to do. Yeah. Cool. Next. Cool. Uh, next on the list we have. Wow. Yeah, you got that. Yeah. Is that command code? So, I've just been doing a little bit of Googling cuz I'm curious. Is the keep hearing this command code thing. Is this what you were referring to? Wait. Command code. It's an agent. It's like Pi or like Hermes or something like that. It's it's an agent that runs on your machine. If this is what you're referring to, no, no, we we are not. We got to buy buy them. We got to buy. But the but first of all, it's completely spelt out. It's like, you know, we we use the shortcut and we don't know. There's no code in our name or whatever. I don't know how people are how people are getting confused, but no. If this is what you're referring to, no. Otherwise, I have no idea what command code is. So Kimmy K3 has been dethroned. Shock horror. Um by Quen 3.8 max and all 2.4 trillion parameters of it. Uh here we go. So Quen is Alibaba's uh model. They've released their Max version. I think they had a 3.6 six max. I don't know. But this is balls to the wall parameters. Um, it's what? Two [laughter] balls to the wall. You ever heard of balls to the wall? No. Everything thrown at it, including the kitchen sink. 2.4 trillion parameters. 95 act billion active parameters when it's being used because it's a mixture of expert model. And it's another one going toe-to-toe with I think there's a Fable. Where's the color? Oh, here we go. Uh, no. Oh, okay. There there is Fable. There is Fable. Yeah, the light the light lighter one is Fable. So, it's it's literally really pushing against these amazing, you know, interesting. Yeah. Multi- trillion parameter closed source models. Um, and they're going to be releasing the first time we will release the weights of a max class model and they're going to be released next week. I think this was this week, so uh 2 uh 3rd. So yeah, still a few more days left, but yeah, 2.4 trillion parameters. And they asked it to build a self-evolving harness. So basically they asked it to create all my CLI um and over 10 days long horizon autonomous uh coding run built this harness and I'm guessing that this is a coding harness or or even command code similar thing to that. It built that um under a loop engineering setup. So, it's a it's a pretty capable model and it kind of goes into what I was talking about last week of like what happened like Kim K3 was like an amazing model but it had a super like it was a lot of parameters. I actually think it's bigger than this one. I think I think in fact I've got it here. Um Kimmy is 2.8 8 billion whereas this is uh 2 4 billion so 400,000 4.4 million I don't know um parameters more that reaches the sort of fable level thing and this is crazy with what we're going to talk about in a little bit but yeah it really does come down to the more parameters you throw at something the better the model is going to be. So and Quinn have a series of open- source models, right? They do have like a 12 billion. In fact, here working on its own for about 5 days, 125 hours of continuous effort, Quen 3.8 Max wrote roughly se 7,600 lines of code over uh 100 1,100 actions and ran 33 rounds of GPU training. It's it first spent 37 hours rebuilding the paper. So basically this was it reproduced a research paper and improved it and it wrote a bunch of code to be able to do that and what it did it finetuned their 8 billion parameter model to be able to improve the paper basically. So it tuned its own model well it tuned another model to be able to to do it. So they have all of these smaller models that you can do more fine tuning with and this and that but again this is the balls to the wall model where they've just thrown everything at it. So, pretty cool. We're still waiting on the weights. They've done the same thing as Kimmy where they're like um you know the we'll release you the the sort of closed source version first and then we'll release the the open weight. So, for now, you can run it in Quen Cloud, which uh I don't I don't know whether you're interested in this sort of stuff, but Alibaba had a token plan which was $10 for Quen, um Miniax, GLM, Deepseek. They had a really, really good deal. They've recently increased the price and actually stopped signing up to it. And then they dropped this Quen cloud plan which check this. They've kind of replaced it. Multiple models, one plan where you get Quen. I don't know if you can see that. Quen, GLM, Deepseek, and WAN, which is an image generation model for $6 a month. So insane. I'll put a link down below which I think gives you a discount or something. It will be a referral link. I'll I'll drop that somewhere. um for $6 a month for all of that. I doubt it will last you very long. And it looks like they're already trying to they're already shutting down some of the you know. Yeah. Um I don't know what this access is. 3,000 credits every 5 hours. Oh, it's the maybe No, maybe I think this is good. I think this is a good thing. It's the It's the It's the five hourly limit. they just given you a week to. So that's really good because sometimes you're on really heavy coding days, sometimes you're not. And you know to be restricted by those 5hour limits can be really frustrating. So this gives you more autonomy in in balancing how you like to work by only limiting you every seven days. So this is a good thing. Open a Open AI is doing that if I'm not mistaken. Maybe. I'm not too sure because I'm checking my usage uh and I see just weekly. I don't see an hourly. Oh, there we go then. Yeah, exactly. There you go. They I know that they did this, but apparently they have stuck with it. And that that that's a good thing. That's a good thing. I like it. I think there is a downside though. If you prompt something wrong or maybe something goes wrong and the model just gets crazy about trying to get what you want to do, but it can't do it and it just gets in a self loop trying to do it. You can potentially run through your whole week week's worth of um credits in one prompt. Like this could happen. Yeah, it's a bit Yeah, it's difficult, man, because it gives you autonomy. Maybe there should be an option to like turn on the 5hour limit or not. It should be like there should be an limit where you can set like if it hits like 20% stop the agent or like pause it. I don't know if it can how that technically would work because your with your chat and with the history and everything. I don't know how that would work like hitting a pause. Don't know. But anyway, but it it's cool that these Chinese models are getting this capable. And as you were like showing me these, I was thinking like, do we need something significantly more powerful than Fable 5? Like Fable is awesome, right? But do we need anything way more powerful than Fable? I mean, we'll take it obviously, but like realistically, what can it do? I mean, it's just better. Like, Faber will still make mistakes. Faber will still hallucinate. Faber will still um, you know, Yeah, hallucinate. I think it's the worst one there, really. So, I think it's just better is not necessarily more intelligent, but better could be maybe it's faster. Maybe you talk about it being faster. Maybe you talk about it being less hallucinogenic or syncopant, you call it. Isn't it like the the model being faster based on like the servers and the the hardware that it yes access to? Yes and no. like there there's only so much physics that you can throw at a model to make it faster and I think yeah I think maybe the maybe someone will come up with something maybe it's the architecture itself like we talk about um you know obviously mixture of experts was something that wasn't available in first instance but we found out that you can have these massive massive models but you can defer the task to um uh Google released Gemma released two models that this is their open- source models called E2B and E4B. And they did a little video about what the E means and it's an it's a a method of embedding. It's called it's E is for embedding, right? And it's just so that they can have the same token exist on many layers. They embed the same token on many layers. And again, this is all way beyond our pay grade, but like the idea is that people are coming up with interesting ways to handle and deal with AI, right? So, yeah, speed can absolutely be achieved through it might be the same intelligence, the base intelligence, but the method in which it retrieves that intelligence could be improved on. Um, and then fine-tuning it with the hardware. We'll talk about Muse in a bit which they talk about um tuning the model with the harness. So there's a synergy there. Again maybe actually they it's been found out that Claude code doesn't isn't the best thing for Claude which is interesting. You'd think that they would be tuned together but that idea similar to Apple I guess the software is apparently tuned to the hardware blah blah blah blah blah but vertical integration. vertical integration the this idea that there yeah basically that there are ways to make it faster without necessarily putting it on cerebrous hardware um but yeah 3.8 Hey, it's hopefully the weights are coming next week. We'll see what interesting models. We covered Kim K3 last week talking about how open weights means other people can serve it. Faster servers can serve it. Um cheaper people can offer it cheaper, people can fine-tune it. All of that stuff comes when the weights get released next week. So, looking forward to that. But speaking of pricing and open AI, uh Chach is free. Did you hear like free free free free unlimited free unlimited free and soul got a little bit better so they done some some magic to improve the prove the model so plus and pro users were updating GPT 5.6 six soul in the chat to be more reliable with facts and providing more focus answers and free users. We're updating the default model to be Luna. So, it's got, if you remember right, you got Terra, you've got Luna, you've got Terror, and you've got Soul. Terra, Luna is their smallest model. Yeah. Um, and expanding access with unlimited text chats. I don't know. We'll talk about But with Luna, with the smallest model. With the smallest model. Yeah. which is good for like basic basic chatting, you know, not for heavy duty tasks. Okay, so but this was an interesting stat. Every week one billion people turn to chat GPT. That is insane. That's insane. Absolutely insane. So here they are showing you know how it's this is GPT 5.6 getting a little bit better. GPT 5.6 soul is designed to make fewer mistakes especially when answers depend on dates, numbers, sources, rules or assumptions. So, it's weird how this isn't GBT 5.7 soul. Do you know what I mean? Um, I thought like if you would improve improving on a model, surely that deserves a little wee bump there. But maybe I don't know. I don't know. But it's still not um like the the fable level. Um, well that raises an interesting point. Well, FA Soul was their No. Yeah, you're right because then they then Opus 5 got released and actually five Opus 5 point uh Opus 5 on um medium effort was the same as GPT soul on high. I I think based on my experience, Soul is between Opus uh five and um Fable 5. It's in between them, but it's better than Opus. Maybe because Opus just talks way too much. Yeah, I know. Tells me like it's just too verbal. Shut up, man. It's like nagging at your ear, you know? Um so thinking about what you just asked makes fewer mistakes. I think it's still the same intelligence. It's still going to score the same. But maybe this is the sinker fancy. I forgot what they call it. Um, it's just going to make fewer mistakes. Whoa, big launch today. Have you increased the load? No, [laughter] that was a big one. Yeah, that's kind of funny. Yeah, maybe it's just the reliability, which I don't know. I don't think it's necessarily more intelligent or more capable, but whatever. [snorts] So, yeah. Um, and then you get a new slider. Um, how to choose how much thought chatbt puts into an answer. Here it is there, which I don't know you. I thought they've always had this. Yeah, I I checked it and I have that as well, but I don't use it because there is an advanced option uh with which you can choose the model, choose the speed and choose the the effort level independently. Got it. Where this is a simpler version of that. Got it. Okay. Yeah. You're you're just you're just too too advanced for this. Yeah. Yeah. Anyway, and free expanding access for free users. We're expanding access to our latest models for free users with unlimited text chats using GBC 5.6 Luna, plus a new think button for harder questions. Now, I couldn't find this said think button. I updated to the latest version, but I think they're rolling it out. I think they're rolling it out. So, uh we'll wait and see for that. But um yeah, that's that's insane though, giving free access to Chachi BT. And this just further exemplifies this big bubble where it's like, yep, free, free, free, free. It money's got to come from somewhere, boys and girls. You know, or if you're if if you're not paying for the product, then you are the product. So maybe there's a small maybe read up on the terms and conditions before you start using the free version which to be fair I you are on the free version right you can yeah but but then I don't really I mean they're offering me a again Sam Alman has just given away they are they are um sorry let me just add add to the stage there uh they are giving me GBT plus for free another like another free trial which is just insane. You should you should try it and you have like you know the tokens you can do the planning for some of your apps with Fable and give it to Soul to do it for you. It it's pretty good at that this kind of stuff and it and I like the new GPG app the the new codeex app. It's it's quite nice with the browser uh being in line, you know, next to your chat. It works pretty well and it's really good at understanding it its own environment. You can tell it to run the server and open it in the internal browser and it does and uses like the internal browser rather than try to use your computer separately. I was going to mention your your microphone is stealing the thunder there. Yeah. Yeah. I mean like I I'll give it a go. I'll play with it a little bit. I'm struggling to like I've got so many models that I'm playing around with right now. Um but yeah, Sam Alman has apparently got infinite money, but I think it's going to come. I don't I don't want and I know I don't know I'm in a very privileged position to say this, but like I don't want Anthropic to join Claude in just giving away all of this stuff. like Chachi BT one of the only frontier from my perspective, one of the only frontier models giving away massive intelligence for for like next to nothing. Do you know what I mean? Like I think all the others are competing on speed and price, but their models aren't as intelligent as Chachi BT, so they the price checks out. So, I don't know. I don't know. Anyway, let's enjoy the bubble while it's here, boys and girls. Let's enjoy the bubble while it's here. Um, I think we should do Pac in the Frontier next. Did you want to cover this one? Pacing the Frontier. Not really. Okay. Like I read on it, but yeah. I mean, I don't I don't I did a little bit of reading on it, but like basically go talking about all of this intelligence and all the rest of it in a weird twist of fate, anthropic and open AI both think along with many others that we need to stop developing AI and give the world a chance to catch up, which is just bonkers if you think about it. So, there's this thing pacing the frontier, a statement from 1,067 employees of open uh frontier AI companies. You got John Shman uh chief scientist of Thinking Machines, Jacob Pachi, chief scientist Open AI, Jared Kaplan, co-founder and chief science officer of Anthropic. You've even got Dario Amade signing this thing to say ultimately we we need to pause. And can we read that together? Like what what is the the statement? [cough and clears throat] AI could help create a dramatically better future, but that outcome is not guaranteed. The world's leading AI companies believe they could be close to automating AI research. It is hard to predict exactly how much this will accelerate I AI progress but there is a real risk that the capability that capability development rapidly accelerates beyond our ability to understand or control the resulting systems. Now Quen which we just covered that spoke about the model itself building its own harness building it training its own models and all the rest of it to realize AI's potential industry government and society at large may need the option to buy time to address the emerging risks develop security measures and strengthen oversight but each company and country is under intense competitive pressure not to ununilaterally slow that acceleration. And today, the world lacks the technical and governance tools to deliberately pace the Frontierwide progress. Building on work already underway to monitor Frontier model releases, we request that the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development. Now, if you remember last week or maybe even the week before, hugging face was hacked by OpenAI's next leading model, right? And it wasn't like it basically was asked it was running a benchmark and it was asked to find the well just answer the questions. So instead of answering the questions with its own knowledge, it thought how can I get these answers to these questions? So it broke out of its container, this sandbox container, found a vulnerability in hugging face, exploited that vulnerability to find the answers to the questions. So God knows what we really went, it really went on hard mode there, just it was like, you know, instead of just like working out the answer, it just hacked a system which then triggered hugging face to patch those systems, right? It wasn't like a control. It wasn't like a planned thing. It literally just broke out of its containerization. We didn't report on this, but actually Anthropic also reported the same thing. It also broke out of its like these things are breaking out of their training harnesses and um you know find running a mockery of all of these companies with their vulnerabilities. It's crazy. Yeah. Um, and then of course it comes back to again just they're they're slowly we do you remember an article we looked at with uh what's it called? I think it's called something like self-improving agents with anthropic. Let me just do a quick let me just do a quick Google. Oh, here we go. when AI builds itself. And here it is here. This is where it started. The person talks to the computer which talks to the AI and gives an answer. Then we get into um chat bots. So this was like pre- chatbot. This is pre- chat GBT here. Then we get to G chat GBT which is like yeah I guess I guess a chatbot talking to the agent instead of just the computer. Then we get the agents that um can sort of self-run and then we get workers where the agent can spin up more work uh more sub agents and then we get into the place of closing the loop where the agent itself spins up his own workers and it's just in this infinite loop. This is what Anthropic reported on a few probably like a few months ago now. But when AI builds itself, so we're slowly getting to a point where it's recursive self-improvement. And this is a this is where we're at right now where they've noticed that before we make this leap, the world needs to catch up because their systems, these hugging face is massive. They are not able to lock down their own software. These AIs are able to exploit vulnerabilities that they didn't even know existed. And it's embarrassing, don't you think? Slightly embarrassing that these these companies are just Yeah. being exploited. So, I think that's where this comes from. The other weird thing about this is that Chinese companies aren't able to sign. And this is basically why that doesn't make sense. I don't know. I think it's because it's only a signature at the end of the day. and it's the signature to get the US to lead the charge on this movement, right? It's an international endeavor. So maybe they just want like I I just don't know. I just don't know why they But it could be a worldwide effort that the US charges and it could just be them trying to protect it all. Do you know what I mean? Yeah. But it it doesn't make sense if you think about it because if we stop if UK, US, Europe, we all stop based on the fact that we want the rest of the world to catch up. That means the Chinese models will catch up because as you saw with Quen and Kim K3, they're already pretty close. Um that that's yeah you stop for one second and the the rest will catch up and will be better eventually or maybe all of this has to do with the bubble like if if they are like um promising so much uh revenue like in the future some somewhere in the future we will make so much revenue but uh they actually see they are not making that type of revenue anytime soon. Wouldn't it make sense to kind of like pace themsel and be like actually we don't want to make the revenue just yet. We need to slow down because otherwise we will make so much revenue that we will destroy the economy itself. So let's actually slow down. uh and that might be I don't understand uh economics enough to to say this but this might be just a mask uh for all the spending uh they are prom that they are doing and all the revenue that they are promising that we know it's not going to happen even some altman mentioned recently that he's happy that he was wrong that the you know the jobs did didn't they didn't destroy all the jobs and and we seeing the AI like even fable I don't know even if fable can actually like fully released cannot destroy all the jobs or something like that. Uh so maybe it's about the revenue that they are promising and they know they are not getting it. But the thing is this is I mean they're obviously going to put the big dogs at the top of this signatures um thing, but this is from 1,367 employees, right? This could be Jane and HR. Do you know what I mean? Like this is Yeah. These are people who don't directly benefit from any kind of revenue. Do you know what I mean? So I don't know. [snorts] I don't know. It's a weird It's a weird one. And I [snorts] it's almost like saying to the world like you need to catch up because we're on the cusp of something mental, which they are. They know they are, but I don't like we we just we're just speculating right now, but they might literally be they might have it in their hands. They just haven't released it yet because it's not, you know, they just Yeah, maybe they are thinking about revenue. I don't know. But either way, it's interesting. It's interesting. And and like I say, this has been going on for a little while. This idea that like I think it was Anthropic that said it like we need to we need to stop. It was it was Dario actually. It was Dario Made. He said like we need to stop. That's how they started. Yeah. And uh about this self-improvement thing. uh something very important is that none of these AI labs are making the claim that this is inevitable and that if that it will happen. They are all saying it [snorts] looks like it will happen, it might happen but it might not. So that means we they they're not sure if we get AI so good that it can create like monstrous uh type of like good AI. It might never happen. We might get stuck at something better than Fable with much better tooling maybe, but not the AGI that they promised, which is AGI not just, you know, a a super um kind of like super smart typing machine, but that doesn't understand the world. Actually, that's not AGI. Like the definition for AGI was like human level, like being able to replace a human. We're just nowhere near that. [snorts] I wonder how much hunger there is for that at the moment. Do you know what I mean? It was a good It was a lofty goal, but like again, you just said like I know I know I know you know you meant well with the with the Fable question, like do we need something better than Fable? And like I'll take it, but like you know, we're already it's already great. things are great. Yeah. You know, I I it can do so much. I would want a fable, as you said, you you made a really good point. I want a fable that makes less mistakes, a fable um that do the things that I ask it to do basically. Um a fable that is much faster. I don't like if I want to create a prototype sometimes I have to run it for two hours, right? So the tooling around the model, these are the tooling around the model, not the model itself. But hey, maybe we will have also a better model. Uh well, we sure for sure we will have better models. But how better? I'm not sure if we will have how Fable was to let's say to Sonnet, not even to Opus. How Fable was to Sonnet. Will we have something like that? Marginally better than Fable maybe. Hopefully. Not sure if it will happen by the end of this the year. Would be insane though. I mean, it's got to, isn't it? Really? Yeah, it would be cool. Like, we we would we would we would take it. Um, but yeah. Um, that's that. Let's see if we can wrap up by half past because this is only a quick one. I haven't got much to say on it other than notes up that Meta is making a really unexpected comeback in coding AI coding. They released Meta AI. What which which layout we going for? We going for this one? We going for this the custom one? Yeah, this one. Okay. Probably need to be Oh, no. Ah, there you go. No, this is good because now on Instagram, uh, you see the vertical one. As you were like talking, I edited that. I didn't know I can do this. Uh, now people can see us more clearly. Great. Anyway, so they introduced Muse Spark 1.2. Um, I think there was 1.1 recent. Obviously, it was 1.1 recently, but then they also released Muse Code, which is a nice wee little harness, command code that you can install with this one thing, and it does everything you expect. And, you know, I mean, this is more muse, more of a testament to Muse Spark. They built this little thing. I don't know how much we want to dwell on the actual um capabilities of the model, but it has everything you expect. It it takes your skills and I've got it loaded up. It understood all my clawed skills which is really nice. Like it's nice that I mean annoyingly they elbowed their way. Okay, I see you're disconnected. Or maybe I am. Uh got to test this. But he did get cut off. Yeah. Okay. So, I will try to see where he is. Uh that's Yep. That's on that side. Yeah. Uh, okay. So, hello everyone. [laughter] So, apparently, uh, Sam has some connection issues. So, I'm trying to to message him and see where he's at. Uh, right. This is the art of going live. Okay. Yeah. You got disconnected. He's coming back. He's coming back. Where did Where did you guys go? You ran away from me. We missed you. [laughter] [gasps] Um, what was I even saying? Well, Muse code, you were talking about how you actually opened it and ah, you said it loaded your, uh, skills and Oh, yes. Yeah. Yeah. And you said something about annoying. I don't know. Yeah. So, um, Anthropic and Claude are kind of like the apple of the AI world in which they've got their own ways of doing things. It's a clawed MD file. They've got the clawed folder, whereas all of everyone else uses the agents folder and they use agents MD and this and that. Um, point is the Muse code uh uh CLI TUI uh just kind of picked up um all of my Claude skills and whatever. So, uh quite nice. Um and what what's really interesting I don't know if I can show it. Yeah, I can't show it, but you can see that I've got the the model selected here, and there's the pricing, FYI. 15 cents in. Um, actually, I think it's $125 in, 15 uh 15 cents for a cash hit in. So, it cashes your previous conversation stuff and then 425 out. Um but the they have a really interesting pricing model which is uh present a new screen here. Um well actually here's the here's the benchmark. So it's not performing amazing but it's it's fine. Here's what I was saying earlier about co-raining Musepark with Musecode. So, it's work. They work good together. But the thing I wanted to show you was here it is. They've got this standard tier, which is what we just showed you. But they've got a cont they've got a contributor tier too, which is 0.002 cents cashed input. Input is 10 cents and output is 20 cents. So they are heavily, as they say, heavily discounting the token prices. But what does that mean? What does it mean? They're they're stealing your data. Okay. So if you're happy for them to train your data, train on your data, train on your code, then you can get huge discounts. So they're really really pushing the um pushing it. But as you saw there on my on my my terminal, I couldn't access this this contributor tier. So I wonder if there is a GDPR thing. Could be. Maybe you have to be in the US to be able to access this tier. I don't know. Um but either way, I've used it really really briefly and it's freaking fast. It's really fast and again, not the most intelligent, but speed is definitely on your side. So, uh I don't know if you saw my videos this week, but I played with Deepseek V4, and I I created a really in-depth plan with Fable and gave it to Deepseek. And even though Deep Seek is still quite intelligent, it's not Frontier level. It was running quanti a quantiz 2 version was running on my local machine. It got the work done. So that's nice with good with good direction. These lesser models, especially if they're fast, it might be the might be the game plan. How long did it take for it to finish the task? Two and a half hours. Okay, that's not too bad. It's not bad. When you were saying earlier Yeah. When you were saying earlier, like I think I think Opus would have taken about 45 minutes to an hour for sure. Yeah. So, not mental, but still a long time. Um, but I guess my point is not necessarily about local models, but more so about less intelligent models being able to follow direction. Yeah. Do you know what I mean? So if you use Grock, Grock's another great fast, pretty good model. Super fast. Yeah, super fast model. So, you know, the game plan might be leaning more towards getting the plan set up, you know, 10-minute plan, 12-minute plan, and then chucking it to a chucking it in an MD file and then getting another model who's super fast to actually implement the plan. So, yeah. Yeah. I don't know. Yeah. So, yeah, that's kind of it for this week. Any anything else you wanted to wanted to mention? Not really. Uh, no. Well, hopefully next week we'll have some more interesting news because it was pretty light this week. Uh, have you checked Twitter since we've been online and and probably Astro dropped or something like that. We got we got five minutes if uh if um we we look to finish our past. Um, anyone else got any news? Let us know in the chat. But yeah, I think that'll do it. When's Web Flow Comp? Somewhere in September was it? Good point. Um I should know. You should It should just be like in your training data. True. Um, so well well they used to have like a full-on website for it, webflowconf.com. I don't know if they have updated it this year. I think they have. Yeah, September 1st. So soon, very soon. Let me add this to stage. Pretty nice website by the way. Oh, interesting. There is I think a WebGL a WebGL effect uh going the going on in the background and there is this Oh, this is really nice. Uh yeah, the details here but it it's not working right. Yeah. Oh, no it is. It is the days for Yeah. Uh this is this is quite nice. I don't know if these are WebGL or just some GSAP animated uh things. Yeah, probably just Gap, but very nice. So, we will be covering this. We will see what will happen in web flow conf. uh we already know uh there will be a big big big release but we don't know exactly what but they are probably going to do web flow 2.0 O with potentially opening up the whole platform that that's [snorts] what I want to see like opening up the platform and maybe they become I don't know they they go the direction of like um like some I don't know like Cloudflare would that make even sense maybe for websites I don't know maybe they I mean they can't stay [snorts] as what they are. So they I think they should go the direction like to eat up um what lovable has gained u like in terms of like users and whatnot. So to be web flow and lovable at the same time and replet they could go that direction. they have the brand for it and I think what matters today is the brand and not so much of the tech that they already have because that tech is ancient ancient and probably not as good like I would really love to some somebody to correct me if I'm wrong there but I don't think their tech is like especially like the jQuery thing that they have like probably the value that they have right now is the community like the people who use the product that's what I mean by the community and the brand the image they have the clients already paying for the hosting and things like that that's their power so they should build around that which is to me a new web flow that enables the people who are already there to do to do way more with web flow like lovable you can export your lovable project no can you I think yeah Yeah. Yeah, you can. Yeah. So, what is the value of lovable if you can build with them and export export it? The value is like I guess exporting is just one option, but like most people just want to know that they can build and host. Similar deal with web flow. They just want to know that it can build and host and everything's taken care of on on the platform. Otherwise, the moment you hit that export button, you take responsibility. I I think it is a like a hard ego thing to say like web flow becoming what lovable is. I recognize this to be a an ego issue, but that's the truth where we are like Web Flow has amazing UI. Imagine Web Flow becoming lovable and the UI becoming something fluid that would sit on top of anything you build. Not just a landing page, but anything that would be amazing. Yeah. Cool. Well, it's yet to be seen. Let's um let's wrap it up there. Let's get on with our days. Can't talk about AI forever. Although, I'm probably going to end up at dinner tonight talking about AI, but you know. Um if you these clips will be released on YouTube next week, so if you're over on Twitter, go check us out commandioshow.com. that'll have links to the YouTube and all the rest of it. Um, what's he doing? Uh, I'm waiting for you to say that's it. Sam, I've been Sam and I've been Kava. Keep on vibing. Peace