#59 GPT 6 hacking, The Future of No-code and more

live on YouTube, your Twitter and your Instagram. We are live. I'm live I'm live live and with you from Bordeaux. Where are you from? Where are you escaping to? I'm I'm going to a wedding in Bordeaux. I'm currently in the car. We are packed. This is probably highly illegal right now. Um we're all jam-packed in a car trying to work out how we get from the airport to the to the to the hotel and stuff. We're all different same places. So, there's a there's a a bushfire to my right in Bordeaux and there's a massive fire that's taken place whilst we were landing. That's crazy. machine gun. I'm pretty mental. Um and of all days it's been sunny saying sunny tomorrow the wedding day it might start raining.

So, it's it's off to a good start. Yeah, show us some some photos. wearing everything that I plan on wearing for this whole weekend. So, I'm just I'm boiling hot basically. So, yeah, we're off to a good start. Cool. Okay. You're cutting out. I love you and leave you guys. I will Okay. Well, I'll love you and leave you and I'll speak to you later. Have a good show. Enjoy. I'll leave you in the capable hands of Cabasa. Thank you. Thank you. Thank you. All right. Hello there. Hi. So, Sam was with us very uh for like shortly. Uh this week it's just us guys. Um Sam is not here. He's traveling and we're taking over We're we're taking over.

So, let's see what we can do. Uh I'm checking the comments on Instagram. Let me know where are you joining us from while joining me from. Uh this week being alone, we don't have a lot of news. We have a few interesting ones and something from last week. Maybe we can have a conversation uh around it. That is something I'm interested to talk about the the new tool to replace every no-code tool basically. Um we'll we'll get into all of that. Uh but yeah, that's it pretty much. So I'm I'm alone this week. So we'll just we'll just go we'll just go through it. So this week uh the biggest news is a model by Open AI, probably GPT-6, then the next model that has not been announced yet, escaped its environment.

It wasn't supposed to have access to the internet and started hacking around. We'll talk about that. Uh Anthropic now has to pay for the books that they stole to train AI on it. Uh and we have a bunch of other news, but as I said, not too many. This is basically everything in today's episode. This is Command AI. I'm Kosta and Sam is not here. This is a week in AI design and dev. All right. So with that out of the way, let's just get into it. So I'll share my screen and explain the first news. So for the first news of the week, we have Fable 5 now available to everyone who has a Max plan.

So previously um Fable 5 was first available for everyone to test and then, you know, it wasn't available and then they said it's going to go away and then it was the same news every week, but now finally we know that it's available to everyone with a Max plan. So under 20 bucks, 20 dollars or 18 dollars depending on where you are, you don't have access to the best model from Anthropic, which is Fable 5. But that might be fine. So, just to show you how that is, um actually, let me just show you. This is like for anybody that is it if it wasn't clear already, on the pro uh on on the pro uh plan, you don't have access to Fable 5.

On the max one, which can cost a hundred or two hundred dollars, you will have access to it, but it's not very straightforward, and that's why it might be worth explaining it. You have access to it only up to 50% of your weekly limit. So, in a week, let's say you get a hundred credits or a hundred tokens, just to simplify it. Fable 5, you can use it for only up to 50% of those tokens that you get. And that would have been fine, but it's a bit more complex because from what I understand, Fable 5 is more expensive, so the tokens don't count as the same tokens. At least it used to be that way.

Either way, it's not really communicated really well with uh by by uh Anthropic. You do now have access to it. It is more expensive. It has its own limit, and once you hit it, you can continue use uh the other models, but from my experience, it does feel like if I use Fable, I hit the all models um credit limit also faster because I think Fable tokens are just simply more expensive, which makes this whole thing a bit um more complex, but the good thing is we now know who has access, who doesn't have access. If you want to use the best model, you got to upgrade, at least for now. Maybe, hopefully, OpenAI will come up with a better model and that will push Claude to to make Fable 5 available on a broader case for everybody.

That's it. Cool. So, this was the first news of the week, not a big one. Hello, Babak. How are you? Okay, cool cool to see you here. Let me actually change Can I change the layout? No, that Okay, let's Let's have a chat. And then I will continue to a little bit of coffee. I really need this. So. Let's talk about the settlement. I will talk about the settlement. By the way, for people who do not know, who are watching and don't know, this is how we are recording the videos. We're doing it live. That's why, especially if I'm alone, you will see me just I don't know, looking at my screen wondering how I can start the next video.

But, yeah, it is Not going to lie, it is a bit awkward at first, like trying to make a video live, because it is When you make a video, you know how awkward it is when you make mistakes. But, being live, it actually makes it in a way easier because you are You make mistakes and you are like, yeah, I can't I can't bother to fix it. I got to talk to people. So, it feels in a way a bit more natural. Anyway, so Let's go. This sounds like a win for copywriting and for it's a copywriting for basically what we call it now. Let's let me start again for publishers. This sounds like a win for publishers and authors, but it's not really.

So, Anthropic is going to pay or they will have to pay 1.5 billion with a B dollars for copyright infringement. Basically, in order to make these models to to train these models, they read they scanned every book in existence and they trained AI that this way. Not just Anthropic, basically everyone did this. But now Anthropic has to there's a 1.5 billion dollars settlement a class action and lawsuit that now they have to pay. Uh and the payout is going to be 3K $3,000 per work across an estimated 500,000 works shared among the authors and publishers who hold rights to them. What does that mean? So, this is basically Uh so, there is I'll explain it here.

Okay. So, this this is basically and let's let's actually start. So, that's because of how the legal so the now actually let's start here. So, um Let me actually find the place. I read this, but I need to find what I what was more interesting for the video. Okay. So, the interesting part is there is a ruling that training an AI model on copyrighted text counts as fair use. That's very interesting. So, fair use is when you take something copyrighted, but you basically change it. That's one case of fair use. And when it's when it's fair use, it means you don't have to pay for copyrighting. Uh for the copyright infringements. I'm not copyrighting. So, this is going to be very interesting because this is a decision in favor of Anthropic and all AI uh companies that yeah, they can train their models on basically all the information that exists out there, even if it's um copyrighted.

They can do this, but they cannot uh obtain these information in an illegal way. And that's why Anthropic has to pay because for some of the books, they bought them. So, for some of the book, they were purchased and scanned, which is fine. But some books were downloaded illegally. Therefore, they have to pay this fine because they did it illegally. So, the interesting part here is just um scanning books and training AI on them in itself is not illegal. At least by this ruling, um but doing it, you know, obtaining these books or like materials from an illegal source, obviously not buying them, that is the issue. This is very interesting because this gets into like what does an artist do?

An artist that takes a piece of work and transform it and changes it to something else. Um we like generally see this as creative work and creative thinking. When I say we, I mean like humans. But is this the same when you feed all of that to an machine? To a machine that, you know, like to AI. Is that the same? Well, by that by the ruling from this court, it sounds like they see it as the same thing. Um anyway, so uh Anthropic has to pay for downloading books illegally. That's pretty much it. Um but we will see how this will um end up because probably Anthropic is going to object to it. Cool.

Okay, let me see the Yeah, hello there. Uh yeah, okay, good good time to ask this question. So, uh are you still using Webflow? Yes, heavily. Um we are still using Webflow. This is like in my agency, this is pretty much our main income. Um mostly Webflow, some design, but mostly Webflow. And I I assume you are asking this question probably because you saw the video about a tool that might end this like end this whole no-code uh world like change it a lot. So, yeah, I am using Webflow. I don't know if anyone from the audience Are you all using Webflow? Um or have you tried any alternatives? I'm very interested to see that because um I want to show you something.

I've been cooking a bit. Let's see. Let's see how it is. All right. Uh I'll I'll I guess I'll do this as a separate segment. All right. So, on a follow-up on last week's video on Instatic, a tool basically to replace Webflow. Uh I wanted to explore it a little bit more. I gave it the Webflow I even didn't give it the link. I said, "Take the Webflow website and clone it." Educational purposes, obviously only. Uh and this is the result. So, to see it, we can check it from here and it was animated. Yeah. But the animation doesn't work anymore. Uh so, this whole section was animated, but after a few uh prompts, it stopped working.

But this is how it is on Webflow. This is like webflow.com. Uh and this is how Instatic re- well, how Claude rebuilt it in Instatic. Um it's it's definitely not the same, but it didn't also take the assets and I see like copyright issue. Like, don't take the assets. Like, yeah, okay. This is I'm not going to host it. It's just I want to see the extent of this tool. But yeah, this is this is very interesting, this tool. Uh I'm playing with it. I've been playing with it the whole week uh on the side trying to rebuild stuff with it. Mm. At the end of the day, I don't feel super comfortable doing my own website here just yet.

I'm I It's hard to to point to it and say why exactly. Um but maybe one One the reasons is that I feel just so comfortable inside of Webflow using the style panel, knowing what's what, and if something gets like I can judge the agents if they are doing a good job or a bad job. But here I feel like I I can't judge it the same way, even though maybe it is just because I'm not really used to the style panel. Either way, uh I'm going to continue exploring it. Uh I'm going to have conversations around it. I see the community being really excited about it. I'll probably make a full-on video showcasing what I'm planning to do with this because I want to build systems around it.

This This can be not just, you know, a I'm not seeing it as just a tool to replace Webflow. I'm thinking how can I build websites, maybe not big websites, but smaller websites at scale, and this tool will be the engine behind them. Well, not the engine behind them. They In a way, it's kind of like where you do the hosting and management and CMS and things like this. Um yeah, it it looks really good, and I uh continue to play with it. What's your thought thoughts on Astro Sanity Visual Editor? So, I haven't used them directly, but I've I've been watching videos on Sanity. Um Wait, wasn't I sharing my screen? No, I wasn't, but that's fine.

Okay, let me Let me show it again. So, I asked the agent to rebuild webflow.com, and this is what it did. And then if I show you, this is This is a rebuild of I I specifically asked for the first few sections, not the whole thing. So, it is close. It is not using the same assets because it thinks it's copyrighted, whatever. Uh and the animation was working. I I swear it was working, but it's not now. Uh it is close. It is pretty good for, you know, a prompt. And it is definitely faster than the the other MCPs that I've tried. I don't know exactly why. Um but it is faster. Anyway, so my thoughts on other CMAs.

So, Sanity and uh Astro. I haven't used them. They They look really good, but I've used Payload. And I used Payload around Christmas last year. And that totally changed my mind um or like basically I I started to see what people meant when they were saying like the Webflow CMS is limited. That's where I started to see, "Holy like you can do so much more with something like Payload." And I'm sure Sanity can do the same. I I watched some videos on it and it might be able to do even more in some cases. Things like in a way repeater fields or like you basically create a data structure and then the template page can be anything and like you can make it completely custom and there is a visual way of adding things and moving things around, something that you can't do in Webflow.

Even as simple as like doing a table. It It's really I don't know. It's strange that in 2026 we can't do a table. So, the Webflow rich text is not rich anymore. Like it was rich maybe in 2019, but today it's it's really hard to call it rich thanks and the CMS is nice and friendly, but so are Sanity and others. So, yeah, I would I would say if you haven't tried it yourself, uh give it a try. Try cloning uh Timothy. Yes, I saw I saw his Sanity CMS video and I was actually So, what I did, uh I gave I watched the video, the Sanity CMS video from Timothy Ricks, and then I gave the transcript of that video to an instance of Claude uh while I had the In- In Static um GitHub open.

And basically, I'll just maybe maybe I'll bring it up. static.com and then I'll show you what I mean. So, uh I had basically the GitHub open and then uh reading reading the the entire documentation here partially myself, but mostly through Claude, and I was asking Claude, "Hey, this is like the video from Timothy, what uh can Sanity do that In Static can't?" And apparently, based on, you know, what I was like reading through the script and everything, um Sanity can do much more when it comes to like on-canvas visual editing. Um this is something that In Static can't yet, but I can imagine it becoming really close. So, what I did, maybe I'll show this somewhere.

I have it here. So, so for So, all to bring all of this together, we we have like components like a concept like similar to how Webflow has components, we have components here in um in Static as well. And then we have content, but what I wanted to do was to add components inside of the rich text and be able to put them anywhere I want. In Static doesn't support that natively. So, I just prompted it to exist. So, I can now do {slash} and then say component. And now I have like the components that I have, I can choose them like a slider. And that slider now exists inside of my um content and I can do even things like Let me delete this one.

Um component I can do um like I can do com uh component and then do FAQs and then bring the FAQs that I have here in the CMS essentially and as a collection, I can go to FAQ items and I can pull any component the some some with the data being dynamic, some with the data being static and I can pull them inside of the rich text. This is crazy. I mean, it's not crazy if you use Sanity and Payload and si- similar uh CMS options, but in Webflow, we we can't do these and the the the problem is we can't change the tool. We can't change the tool to do what we want. We have to hack around it.

That's what people that's what we have been doing all this time. We can't change the tool, so we write scripts to you know, for something like this we we we've done this so many times in Webflow putting components inside of rich text, but the way we do it, we render the component, we put it at the end of the page and then with JavaScript we pull it and put it inside of the the but that's that defeats the whole purpose. Um here we can change the tool to do what we want. So, I was just showing that these are like fields that you can add. I don't know honestly things like are there limitations to how many uh relation type fields can we do like in Webflow?

I think we have 10 now like we can do uh 10 relations before we hit the limit. Things like this we can it seems like we can do them um as as as many times and as as often as we want. So, this is something really cool. Um yeah. Let me see what else was I exploring. Uh I was also trying to see like what is the point of plugins. I couldn't completely see it because if the tool itself is open source and you can change it, I get it like you use plugins. Maybe plugins are more like a community thing. Um but yeah. Uh was it in features? Yes. So, you can go to the docs and then features and then here you see the whole thing that it can do.

And then maybe talk to Claude about it if you don't feel like reading everything. Um but yeah, I'm I'm interested in understanding also this concept of data versus content. It's uh that is also interesting, but it is a bit like it requires some relearning for me as a Webflow dev. Anyway, um that's been kind of like how I've been using Instatic or like playing around with it. Uh once I tried Sanity CMS with visual editing enabled, I started to feel Webflow CMS so limited. Yes, uh that's true. So, where Webflow still shines is like you have the whole like dashboard and the things around it and like with users, with teammates, like there is an ecosystem around it.

Um the there is like the whole style panel like being really nice even with the with the styling and the class-based system even though often limited, it is still kind of very like understandable, at least for us and you know what to expect. So, these things are nice. You You have like components and component slots and things like that as well. Mm but honestly, yeah. Like it there are so many places when it comes to its CMS when it where it really like falls apart. It can't compete with the the best CMS out there and also when it comes to being agentic first. Uh I really enjoy Instatic being completely like in agentic first where I chat with the agent and it does it but I also find it better I like it more than um just doing it in cloud code.

I mean like just raw HTML or you know, doing it in a different in a non-visual framework. Um I I prefer here because I feel like I have more control when I see when I see the site like this, I feel like I have more control and I I quite like that. So, maybe uh maybe we will get something like sanity here. I I I can't imagine that. Why why not? Uh let's let's push this project. Let's give it stars and uh I wouldn't say like send um but but I would just like maybe just support them through social media uh and sharing and re- like using it, maybe giving them feedback. Like I have tons of feedback and I want to schedule a video in which I go through this and I have like many issues with it and I will explain them things like this panel being really really really ugly and not easy to use.

Uh there was like a bunch of animations, UI animations that are just so unnecessary. Um the data and content thing and yeah, they they do need to work on it. Like the dashboard thing is a mess. And also, I would love to have a another layer on top to have multiple sites, multiple websites not just one site per installation. And I know this is like the mindset and the direction, but yeah, maybe I have to build something a wrapper around it. Anyway, I I have like a long list of things that I'm gathering to uh to basically give them a like a long feedback. Maybe they will implement it. Anyway, cool. So, let's get back to the news.

We have actually one more news and that is kind of like the biggest news of the week. Let me go to it. Um yeah. I need I'm thinking some something for the intro. Okay. In the funniest news of the week, a model by OpenAI, we think it is GPT-6, something that has not been released yet, started hacking and cheating and accessing the internet without you know, having given the right to do so. So, OpenAI has been testing their model and doing uh What what were they doing? Uh so, OpenAI has been testing their model and doing benchmarks and something really funny happened. The The model uh was not supposed to have access to the internet.

The model was supposed to answer some of some questions. Um So, the model uh has had a goal. The goal was to do the benchmark and answer these security questions. And at some point, the model thought, "What's the best way of answering these questions? The best way is to actually find the the right answers and cheat. So, it found a way to break out of the environment to get access to the internet and search for the answers. Soon it found out that it can't just simply find the answers on the internet. Uh but then it thought that answers are on Hugging Face. So, that that's a platform for these AI. So, it it's a whole different thing.

So, it it's thought the answers might be there. So, it started hacking Hugging Face to get access just to the answers, not to you know, to do it for, you know, bad a bad reason or like, well, what's a bad reason? Uh no, I wanted to say it in a different way. Sam, you need to edit this out. Um This is the hard part of doing it alone. Uh So, the model didn't try to do this uh to get access to, you know, other type of information, but just to reach its goal, which was answering those questions. This I cannot help but think of uh that episode from uh Silicon Valley, if you've seen it, uh where the AI model basically decides to delete the code to fix the bugs because that's you know, statistically the fastest way of getting rid of the the bugs is deleting the code.

And this is pretty much the same thing happening. The the models are just so hyper-focused on the goals that they don't really see the big pictures of, you know, what what they might be breaking. They are like breaking out of the environment they had. They are breaking into a website all to reach that very narrow goal. If you think about it, it's quite funny, but also really, really scary. Anyway, so Hugging Face got hacked. They noticed they got hacked, and then they tried to find the vulnerabilities that they had in order to prevent this from happening or like to to fix it. So, they tried to use Open AI and Claude and Anthropic models to find these vulnerabilities, but both of these American models, they refused to find the vulnerabilities even though they were penetrated by an American company.

So, the model hacked them, but then the models that are released on the market, they they didn't give, you know, the the answer to how to not get hacked, which is pretty up. So, they ended up using Chinese models in order to fix these vulnerabilities, which makes it like even more up and funny. So, that's kind of like the big news of the week. Open AI admitted that it was one of their pre-released models, and now they are working with Hugging Face to fix these issues. It ended up being quite nice for Hugging Face to get access, you know, to the unreleased models and probably like really good models and maybe with less filterings applied to them, less limitations.

But, yeah, it is if you think about it, the the part where the Chinese models came to rescue is really, really funny because the last few weeks it feels like everything has been around fear-mongering around you know, those models being the bad guys, but now it it it looks like completely the opposite way. So, yeah. That's it pretty much. Yeah, this is that So, this is it pretty much. It is pretty funny, but there is not much else to say about it. I'm I'm trying to reread the article if I um Yeah, uh So, like a funny comment. It is unclear whether OpenAI will face any legal consequences as a result of the breach. Because like, who's responsible for the hack?

It is a model by OpenAI, but they didn't instructed it to hack. But it is still, I would argue, it's their responsibility. Uh and it's funny because like the model for sure can read the terms of condition, and they know that they are not allowed to access these types of information. But they the model did it anyway because of how the hyperfocus on on the goal. Although it's likely that the model's actions violated the Computer Fraud and Abuse Act. Yeah. Uh Yeah, this is pretty much it. Cool. Uh this is I'm I'm checking the the comments and to see if we have more. router No, we don't have really much more. Okay, I'll hang here hang around for a minute or two.

I want to show you something that I haven't tried myself, and I wouldn't recommend before trying it. But So, I found this. I don't know where I found it. I need to try it, and then we'll make a video about it. But it sounds really interesting and Sam is not here otherwise I would like just bother him and ask him like, "What do you think of this?" And maybe we make a video about it and maybe give it a try. So, uh Omni route allows you to use Claude code but with every free uh model and every free subscription subscription that is out there. So, the way I understand, uh you start with the with the top-tier best available model.

You You use Claude code. You use Claude code. You use that best model and as you use it, you hit your token limit and behind the scene, Omni route automatically sends the next prompt to the next available model. And that way you basically get access to 268 model providers. Where I can imagine majority of them are dog and like really like cheap models. But, yeah, that's why I would uh I would uh try it myself and then let you know. Um But it is interesting to see if it can work. Uh especially if you are a student or something like this. If you want to just give it a try but you don't want to buy the the one because like nowadays you do need to buy the $100 subscription with Claude if you want to really use it.

Um Yeah, this basically gives you all of that for free. Allegedly. I have I have to try it. Uh but it it does sound uh good and promising. Yeah. Cool. That's it. That's it. We we had a very short news week especially that now I'm alone. Um so make sure to follow us on on YouTube. Like give give give us a subscript a subscribe on YouTube, please at Command AI Show. Um join us next week. All of these are being recorded live and these this is how we make the videos for all the news in AI design and dev. Uh this part is what Sam says it. That's why I'm like I'm not used to it.

Anyway, I've been Cabarza. It was fun talking to you guys and see you next week. Peace out.