#50 GPT-5.5 Cyber, Apple $250 Million Class Action, Anthropic x SpaceX, Webflow Pricing

Oh, we are live. We are live. I wonder when you were going to join. How are you doing? I'm good. You're coming out my head my uh earphone now. So, yes, I am doing wonderful. Thank Thank you so much for asking. Thank you so much. I really appreciate that. Um, yeah, thank you. Thank you. H how's your how's your week been without me? Uh, missing you. Yeah, of course. Of course. Um, I was off gallivanting around. Have you ever been to MadiRaa? No. No. It's uh it's a volcanic island, Portuguese volcanic island off the pretty much the coast of Africa and it's like it's basically like a hiking, you know, hot spot. And it feels like Jurassic Park when you're there. So I was constantly every time something magical happened, I was like you you've seen uh Jurassic Park, right? Yeah. But no dinosaurs. No dinosaurs. No, just um just us, you know. Yeah. Um so yeah, it was uh but it was also quite rainy as well. I saw the the social posts, the the videos you made, the social what what am I saying? Like the the Instagram stories the the stories. Yeah. Yeah. Yeah. The snap the the Snapchats. the Snapchat. Mate, I ain't on Snapchat. Don't drag me down. Are you on? I am on Snapchat. Yeah. Are you? Yeah, I'm a daily user of Snapchat. So, I have my friends uh on Snapchat and that's how we know who's up to what. Uh it's just kind of like a quick check in, you know, what your friends are up to with your homies. Kind of fun. Yeah, it's kind of fun. So, so what what what have you been building then since I've been gone? What's been what's your agent been up to? Uh gosh, a lot. A lot. Uh one of them is I'm not sure if I want to share it now. It's like the one that I showed to you, but I'm not sure if I should share it online yet until maybe later on or I will anonymize the information. Uh we're planning a trip to Croatia and I made a website for it uh to share with with the boys. So so everybody knows what we have to pack and you know where we are going, how is the weather going to be. Yeah, you saw it. It's it's like it's cute. Packed with information and beautiful. It's really good design, isn't it? Um Well, I mean, maybe you can show it when we go over the the unreasonable effectiveness as HTML, but that basically is is uh exactly what it's for, right? Yeah. Yeah. It's it's quite nice to have it that visual. Yeah. You can't be sending your home boys markdown, can you? You can't be sending them markdown. No, they they don't read even the you know the big H1, let alone, you know, the you know the the hashtags in an market in a market. I don't know. I I've planned a trip in the past. I know exactly how lads are when it comes to or people even maybe maybe if you're a girl, you can you can chime in here if you've ever had to organize a trip. Getting anyone to do anything is like a a frigin slog. But yeah. Yeah. Yeah. Um like if I send them an an MD file, they will say this is code. That's Yeah. Yeah. Yeah. That that's that's the what they will tell me. So I um I mean I've got a video coming out on what I've what I've been building I'm quite excited about. And uh it was I I want to do more videos on like actually building stuff and showing what I'm building and how I build it and whatever. So it was really nice to get a sponsor who gave me something to work with and actually build something quite exciting. So yeah, it was all we revealed, but it was a drag and drop video thing that had subtitles to our videos and it was just it was just nice just I I like building stuff like I I would like to do more of it. So go to samuel.co.uk. UK and hire me. So I build for you. Yeah. Yeah. It's so fun. It's so fun. Uh what I'm sure I built something else. I mean I've been adding a lot of automations and stuff to n which has sometimes used AI sometimes hasn't. Um, just because again I think that's the that's the golden age we're in as like entrepreneurs that we can do so much more now rather than hiring a team of people or working longer hours. It's like you know um we can we can do all that ourselves a bit of investment and then every week then it compounds. Saying that like I had a mass I I recorded two videos yesterday. I did a bunch of updates to the uh N stuff that I've got going on. I actually finished my work like on time. Well, early. I finished work early. I'm like, you know what? Just going to stop working now because I was such a temptation. Such a temptation just to keep working into the evening or or get this done or get that done. And I was like, no, no, it's still light outside. I'm going to enjoy my evening. Yeah, just one more prompt. It is quite addictive uh because you build and then you immediately see the result and also there is this gambling aspect of it like um it can get like it can get you really good results and then you get stuck at like a small detail. Um, so yeah, you don't you don't want to you don't want to try and have to pick up where you left off the day before, you know? You want to try and round it off like nicely, but then yeah, that can of worm gets opened up and that can of worms gets opened up and and this and the other. Anyway, I've got a hard stop, so I think we should uh we should try and kick the show off if we can. Yeah. Yeah, definitely. Um, I haven't seen an intro from you, so you're going to have to wing it. Yeah. Yeah, we True, true. Uh, we were like about talking about um Okay, we'll we'll just go through it. And are we going to talk about the Open AI stuff, right? As well. Yeah. Yeah. Yeah. Why not? Okay. Yeah, why not? What are What are you drinking? What is that in your hand? As always, sparkling water. No, no, that's not what I'm asking. What are you asking? What is that in your hand? Well, I've got a microphone and I've got a bottle. No, no, no. Yeah, that one. Bottle of water. Leave. That's uh that's racism. Yeah. Uh, by the way, before we start, your video is getting stuck quite often. I don't know why. Yeah, it's like lagging and like it it freezes and I'm I'm not sure, but I think it also your voice that freezes as well. Can you check on YouTube? Maybe. I'm doing that. I will check as well. Your voice that freezes as well. Let's rewind. Just sometimes. So when you had like the bottle up in a questionable position or in my mouth. Yeah. Yeah. This looks like you're on YouTube. Maybe people can pitch in wherever you are. So maybe it's just me having too many instance of cloud code and terminal. I probably have like it is there's no one in my house right now. So that's why I was so surprised. It's like I'm I'm the one eating up all the bandwidth right now. So what it's worth for who who's on YouTube, give us a give us a like. Give us a little like. Um anyway, just in the process. Yeah. Yeah. Yeah. We have a lot to cover, so let's go. All right. So, give me um So, this week we have Open AI's new security model. We have Apple was going to pay 250 million to settle the AI lies that they they've been telling everyone. Uh we'll talk about that. Anthropic. We have actually quite a few news from Anthropic, especially a deal with SpaceX. That That's a big one. Uh, Proplexity's new Mac app. Uh, we have a a new model with a new technology. I'm not sure how to describe it. That's above my paid uh grade, but sub quadratic uh sub like first sub quadratic LLM with 12 million token context. Crazy stuff. uh more anthropic news and we have uh updates to web flow plans and pricing along with something from lovable uh which we will talk about all of that and more the HTML versus MD file one uh more in today's uh episode. This is command AI. I'm Kavarzo and I am Samuel Gregory and this is a bottle of water. So let's get go uh through the the news. We have quite a lot especially because we weren't here last week. So yeah. So last week quietly announced I think quietly announced was a new cyber security model from open AI this time a challenger to the great mythos. And we all laughed when they said no no no you can't have that. But now open have practically done the same thing. Um they have released a a cyber security model called GPT 5.5 cyber. Let me share the article with you. Um and they have only released it to the privileged few. Open hour is preparing to launch a new frontier cyber security model GBT 5.5 cyber. Uh CEO Sam Olman said the model will not be available to the general public but will be first to roll out to a select group of trusted cyber defenders in order for institutions to shore up their cyber defenses. Now, it's really interesting because like I say, we kind of sort of laughed and it was a bit the primigan said something along the lines of he he phrased it in such a way, but there's a bit of like a snobbery when it comes to anthropic and holier than thou is the is the phrase he used and it was sort of exemplified when mythos first came out. But clearly, uh, OpenAI have done the same thing. And I do, you know, I do wonder why it is, it it's gone that way. I mean, we've seen the the MPM worm that got released, the shard habude something or other. Um, which is quite worrying actually. I think I think the best thing to do right now is to kind of shelter in place almost, not like mpm install anything because tan stack got infected by it. I don't really know the updates on it. Maybe some people can chime in on that those updates. But cyber security is becoming a big deal and obviously I think this will get released I thought about it earlier. I think this will get released to the public when they have a better model. Do you know what I mean? So that these cyber attackers will never have the best model sort of thing. But anyway, I'm I'm I'm digressing here. The limited rollout will take place in the next few days. Sam Alman said on X. who will work with with the entire ecosystem and the government to figure out our trusted access for cyber. Um it's not clear who will get access to the model first though previous trusted access schemes involve vetted professionals institution. Okay, so they've not actually announced who who's going to get this thing yet. It must be like a this must be like a pre pre-released um yeah thing. Let's let's have a look at Sam Alman's tweet. We're starting to roll out GBT 5.5 cyber frontier cyber security model to critical cyber defenders in the next few days. We will work with uh the entire ecosystem and the government to figure out trusted access. That was on the 30th. So it was actually two weeks ago. Oh, we're really behind on this. I thought it was just because I was away on holiday, but you know this I wasn't paying attention clearly. Um but I think that's kind of it. Um let me check my notes here as well. Um, to roll out, no technical details. Wirehouse is separately pushing back on anthropics plans to expand Mythos access. Um, yeah, growing industry trend. Frontier models being branded too dangerous for mass release. So, I think that's the kind of headline thing here is that it's it's becoming a yeah, either, you know, it is a holier than thou thing or it's a privileged few. It's a payto-play kind of thing. Um or they know some stuff that we don't. Yeah, with methods we had some proof like not exactly but they release more, right? We had the they they said like these are bugs that we have found and that's why it was so impressive with open AI not so much. But on a similar note, I learned something. I don't know where I read it or a video I watched. uh hackers are using the the current models like just uh 4.7 4.6 whatever and they are using other openw weight models to to take on a security task and break it down into pieces with a with a dumber like open model uh open weight model. They are breaking down the the pieces of the task like how to hack something into chunks of prompts essentially or like smaller tasks that do not sound like hacking something. Does that make sense? Yeah. So they're sort of just disguising their actions. Is that right? Exactly. they are disguising their actions, making it in small chunks, giving it to a smart mo model to to do it and then the results are or like the other way around. Uh I'm not sure but yeah, they're essentially um making it like a big a big task, they are breaking it down to smaller ones to to do it. So maybe you don't necessarily the point here is that maybe you don't necessarily need a big model to to do this for you. Uh maybe all it takes is like these security experts if you know what you are doing. I suspect. Yeah. You know how to prompt your way around doing something maybe without even saying it. Like I would go there and be like hack marker Mark Zuckerberg, but a security expert might be, you know, much more delicate than that. I did see an article somewhere. I'm just trying to search for it, but a a company was notified of a security bug with which they patched. And I don't patched and I don't don't know whether it was a cyber security firm. I don't know whe they were using Mythos. I don't know anything really much more than I'm sure companies are starting to find out about um about cyber or vulnerabilities more so than anything. But to your point, yeah, like I think with that comes like a deep knowledge of the systems and just being clever about it. Thinking outside the box, I think is the is the the term to summarize what you're saying. Like they don't need the smartest model if they can think outside the box with with dumber models. But um with the MPN thing that happened recently that was that was pro like again I think this was the primigen he said like there's so many safeguards in some of the frontier models that it was most probably done with an open-source model and to to sort of bypass um some of the the the uh nefarious activity that they were getting it to do. And again, this is a really um uh a destructive thing that happened. Really destructive thing. And this would have been done with a dumber open-source model. So yeah, it's a bit of a bit of a wild west out there. And um I think there were all this this battle will rage on though. This battle has raged on for centuries. It seems like where the hackers you kind of need them. You know, you get the what do they call them? The white hat hackers who hack things and then tell you Yeah. like there's a vulnerability here. Um I I do think goodness will prevail. However, it's an ongoing battle. There will always be cyber security hacks. Um but yeah, let's uh let's move on because uh time is of the essence and that's about all I've got. Unless you got anything else to say? No, that that's it. Cool. Uh, let me move some stuff down because I'm I've zoomed in because I've got a demonstration on something for later on and everything's all massive big thing. So, the Apple are to pay $250 million to settle a Siri class action lawsuit and I've just set off my devices. So, you remember the iPhone 16? You remember Apple Intelligence that got released and it was going to be the coming of Christ and it was going to know exactly what you're doing at all times and and give you context awareness and do this and that. And you had that girl uh if you can I don't know whether you're allowed to call him a girl to be honest, but you've got that person, let me just share my tab here, stout uh running into to someone um and not knowing when they knew them from. Um, oh, I am sharing my screen. Sorry. Uh, this this this person here, you remember all this? Well, they sold you a phone. They sold you a phone, right? Yes. Yes. On the premise that you will have Apple intelligence, but you'd never got it. Still haven't seen it. Still haven't seen it. Um, I've got the 15 just to say, and this was the first phone to have Apple intelligence. I'm not sure it was sold under the premise of Apple Intelligence, but I think they determined that this chip was fast enough. But yeah, so now you may well get uh $25 rebate. So I'm not getting $250 million. You're No. No. Unfortunately not. No, unfortunately not. The lawyers are getting most of that probably, but you as a consumer Yeah. I don't know how like in the notes we have it that um allegible devices will get $25 to $95. Oh, I see. 95. That's quite good. Yeah. I mean, if I get a hundred bucks because Apple lied to me, that that'd be fine. I I' I'd be happy with that. Yeah. Well, I mean apparently uh according to my notes here, the Can you zoom in a bit, please? Sure. Yeah, we need the Apple have have how do you say they have denied, you know, they've not they've not admitted or they they've they're not necessarily agreeing to it or something. I don't think they've been found guilty, but I'm not trusting my notes there. That that's quite ridiculous because Apple intelligence for all like seriousness and for all purposes it never happened. Like I I can't summarize a thing uh with that thing with the emojis. What's it called? Like the emoji gen whatever it was. Genoji. Yeah. Genoji. I tried it like 20 times in the past two years. I have like zero results. It It just fails. that says it can't do it. Yeah. Yeah. But I never even made one because it just never worked. So I was sold a phone on the premise that it can do something. They shipped it. The UI exists and I click it. It it does the spinning whatever but it never gives me a result. So yeah. Yeah. I want my 100 bucks. Apple. There you go. Um, yeah, I I added some results, but it never worked the way I intended to. I thought it was literally going to fuse two emojis together and this and that, but it never worked the way. Anyway, that being said, I did just notice iPhone 15 Pro and 15 Pro Max. So, my my phone is covered, which is great. So, I I had the 15 Pro Max and the 16 Pro Max. So, I you should receive double. Hopefully, you get 200 bucks. set has received preliminary approval and notices to those eligible to make a claim will start to receive email notices no more than 45 days from today which this was announced probably last week now isn't it May 5th so 10 days ago so 30 35 days you'll hopefully get an email okay there you go cool I'll let you know I will let you know because I've got 15 I've got the 15 Yeah. Um, yeah. So, there we go. Um, move on. Let's move on. Oh god, it's another another anthropic one. So, so with the anthropic, we wanted to kind of like say them together or like merge them or Yeah, we we go I think I think we'll do an intro for for each of them, but yeah, we'll bundle them all together. Um, go on. Yeah, I want to say like you could also do an intro for for all for all three and then do an intro per news and then later on in post you can decide to merge them into one video or have separate videos. Okay, to be fair, there's not it's not crazy. Uh, I'm just moving things around so I can see them all together. Yeah, really. It's only two. It's only two. Wait, wait. There was Oh, the other one we didn't add or like so the there is this the deal sign the the agent view, but we also have the uh the the recent limitation. Oh, the uh Yeah. Yeah. Yeah. So what's it called? The Yeah, the the programmatic use programmatic use of of chlor code. Yeah, I don't have anything that was kind of like disguised as a as a treat, but actually it's it's it's a massive downgrade. Um we'll just talk about the programmatic use with the deal with SpaceX. I think they are very relevant. Um yes because because we got the it's compute um and it's also this idea that there is they we got given like increased usage but then all of a sudden a week later it gets taken away which I I I was I'm pro and I I felt like my usage got used up so much quicker whilst these reduced usages. I was like surely anyway whatever. Um, Anthropic have been busy boys this week, pissing off the community as well as delighting them in various different ways. It's been an absolute roller coaster to be honest. So, first of all, SpaceX uh/xai signed a deal with Anthropic to use the entirety of their Colossus Supercomputer. Uh, Anthropic also launched Claude Code agent view, which is an amazing one. And they also delighted us with increased usage or um uh stopping with the rate limiting and things which I guess is a byproduct completely of all of this supercomputed nonsense and their deals that they've done with um Amazon as well. But then just pissed everyone off by undermining it all by saying no more programmatic usage beyond anything outside of the ecosystem. So let's get into them all. Um I my my monitor is doing this thing where it's where I need a new monitor. It's flashing and it I've re Do you know I've got a a really uh what's it called when you fear something for absolutely no reason. Um I always forget the term for it. I just think my machine is my computer is just going to explode and a bit of glass is going to go my way. What is it called? Yeah, but but like I don't know what it's called but you have glasses so you're safe. I'm protect I'm fine. I'm fine. Yeah, no big deal. What am I worried about? I've got glasses and a hat. Anyway, that's my That's my way of working when I work with a chainsaw. I have glasses. I'm fine. Yeah. Yeah. Stops a chainsaw. Yeah. Um, so Anthropic gets in bed with Space X as AI raced turns weird. And this is weird actually because Elon Musk has never been a massive fan of anthropic. And I might actually share this first. Where am I? Quite a lot of drama. Frankly, I don't think there anything you can do to escape the inevitable irony of anthropic ending up being misanthropic. Oh, good one there. You were doomed to this uh to this fate when you I can't bo anymore. Anthropic is guilty of stealing training data at massive scale. Uh where do we go? Antropic is myanthropic. God damn it. Um so so this is all Elon Musk making fun of anthropic and now they are buddies. By way of background for those who care, I spent a lot of time this week with senior members of the anthropic team to understand what they do and to ensure Claude is good for humanity and was impressed. Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector. Elon Musk has an Elon detector all this time. He's been hiding it from us. Gosh. So, as long as they engage in critical self-examination, exam examination, Claude will probably be good. What a what a what a new leaf you've turned over there. So, yeah. Um, so basically Elon Musk is now good with Anthropic because they are selling them their compute and Anthropic is growing like crazy. they are like limiting the compute that we get and a lot of like um programmatic users get in order to serve the enterprise more and they are doing that partially with SpaceX. Yeah. Yeah. I don't think it I I don't quote me on this. I don't think it's fair to say that they're just going to give all of this to government and whatever. I I haven't read anything that that suggests that. However, I'm not going to, you know, I'm not going to deny it necessarily, but since signing this deal with uh with SpaceX who /X AI, I don't know. Um they they signed up to all of the I think it's like 300 terowatts or 300 million or something. Let me It's 300. Let me research 300 megawws of capacity and they've they've got all of it. They've got all of it. I forgotten where exactly it said it, but trust me, bro. And um and I think the reason why they've given them all of it is because they're XAI SpaceX are working on superconductor or cluster supercomputer 2. They're working on a second one. So they've just given them all um of that and we've we've known that Anthropic have been constrained with their compute for a long time. It was a bet that Dario made very early on that actually long story short, the models will get a lot better, but actually it's gone the reverse. We they he needs more computers. So it was a bad business bet. Yeah. That he made very early on in Anthropic's um career, life, whatever you want to call it. So now he's in a big bad bum rush to try and get as much comput as he can. This is not the only deal he struck up. I think he struck up deals with uh Google to get some TPUs and he struck up some deals with uh AWS as well to get some of their comput. So I do think they are they're making up for it. However, yeah, just before this deal, go on. Yeah. Yeah. Yeah. Go on first. Before this deal, we are actually constrained. They were putting usage limits on. reported on that where there was a website telling you that you could use it um with this s like they were trying to encourage it during downtime and XY Z. Then they got this deal and actually now they've or at least last week we got basically increased usage from um our Claude Max subscriptions and all the rest of it. So there's definitely we're definitely immediately seeing the um the the benefits of this deal that they've managed to to secure. So um that's why I said I don't I don't know whether your government thing is totally true because we just we just saw that our No, I don't mean like only they are not giving it only to I didn't mean government. Is it like enterprise enterprise? That's where they are. Yeah, there that's where they are making their money. But obviously we are going to get benefits of it as well, but that's not what what's making them money. What's really funny to me is seeing the entire picture right now. Elon Musk personally himself, he's in court against Sam Altman, the the issue they have with open AI. Now historically Elon Musk didn't like Sam Alman like we know they are right now in court but they also didn't Elon Musk didn't also like Anthropic as you showed he used to make fun of them but I guess Elon Musk hates Anthropic less than he hates uh open AI so he's I would say partially it's good for him that anthropic is taking business away from open AI because it seems Like I I think I read somewhere that Anthropic um now makes more money. Their revenue their business side revenue has increased um more than Open AI. I saw something. So they are getting ahead which is crazy because Open AI just the word the the word chat GPT it became a household thing even my parents have heard it. Uh but now anthropic has catched up which is crazy. But the big picture, what's really funny is that all of these people, all of these companies, not just these three, but also Microsoft, Amazon, uh, Google, they all have their own models that they are training and they are all also have or like most of them they have their um, GPUs for for the training and also the compute that you need. So what's really funny is that they are kind of like com competing but at the same time they are giving each other so much money and they are investing in each other's uh infrastructure. They're literally paying each other. It's quite circular in a way. They are paying each other to just make sure whoever is going to win that they will have a stake in it. M that that's kind of how it is like if if uh Anthropic is going to win or like going to dominate the market Elon Musk will benefit from it because it's will be on their infrastructure and if OpenAI is going to win Microsoft is going to benefit from it and you know if Microsoft is going to win it's all like in a circle which is uh really funny and not to mention Nvidia being a big part of that. So, speaking of a bubble, I'm not saying like it's a bubble and it's going to burst tomorrow. It it might, but it's quite circular circular and it seems like most of these companies still don't know how to make money, but they are just investing. It's a circular investment going on. Well, there's further irony in the fact that um not only that Elon Musk was, you know, [ __ ] on anthropic all this time. Like they're literally renting GPUs from a company that has its own AI. Yeah. So, it's quite again there must be this Colossus Supercomputer 2 that I heard of um as well that that plays a part in all this. But where are they renting their where are they renting their computer from or getting their comput? is have um 220,000 GPUs they've got. Have they got 440,000 GPUs and only half of them are were available to rent? Don't know. The other question here is around um Curser and that deal and and who how are they going to provide for Curser. So, it's sort of like there's a lot of passing around that's going on. It's hard to really keep track of everything and understand it, but the only people that really know are these are these Billy Big Boots. Um Yeah. One one last thing on on that uh topic. Um SpaceX Twitter is all now SpaceX. Uh they want to IPO and I I was watching um Lemonade uh stand the the the podcast. So they they want to IPO for $1.5 trillion. It's it's an insane number. Yeah.$ 1.5. Yeah. They're going for 1.5 and Open AI and Anthropic they are also going for about a trillion just short of a trillion each each all within this year. And apparently there is just not enough liquidity and there is not enough space for three big companies like this to to IPO at the same time. So maybe this is all relevant. Elon Musk is also trying to sue Open AAI to slow them down. And again, yeah. Yeah. So it it's quite complex. Um these these people they I don't know. I don't understand the finances here enough, but it seems like taking it from a pocket to putting it in another. it it sounds fishy but also it's it's for it's 4D chess you know um but to round this whole segment off is the programmatic usage is that whilst cl uh whilst Anthropic gave us extra usage or increased usage since the deal with Space X they've also taken a tremendous amount of usage away by basically paywalling any sort of programmatic usage of your clawed max subscription. Yeah. Now, uh I'm going to rip off Theo right now because he did a great explanation of this just to give everyone a an idea of what's going on. I'm going to try and position my microphone somewhere. I don't know. Um let me let me do a little rundown. And we are here. So if we draw if we just draw draw a line here and we get some text you've got say when you go to claude uh.ai uh is the kind of ideal use case um where you can use your Claude subscription and then all the way over here you've got API usage. Okay, these are the extreme ends of what Claude Anthropic offer, you know, how you can pay them money and and seemingly across the X-axis here is also the less preferential way to to to use your subscription. So then further down the line, you've got Claude's uh code here, which is, you know, slightly less preferential because it's so flexible um but it's still within the realm of feasibility. But then you've got something like clawed uh-p which uses clawed code but allows you to wrap things around it. It's outside of the clawed code uh UI, let's call it the UI, the the TUI, the the the console where you know that you probably all know and love. This enables things like T3 code conductor um your own tools and all the rest of it. it starts to this is where it starts to sort of break down and then you've got the um clawed um uh SDK which is a similar sort of thing allows you to build tools but you write the code for it. So whilst this pipes I hope this makes sense for maybe some of our less technical people but claude-b pipes into claude code and uses everything around it. This uses it almost uses the code that claude code uses. It's sort of like if you were to draw a tree like you've got claw um you've got claw code here. Um you've got the let's say clawed uh core let's call it rather than claw code and then the SDK will also share that that code as well and but it's separate from clawed code. Hope that makes sense. But that that's like a little rough diagram there. Um what have we got after this? And then you start to get we have to move this along here. You start to get uh tools that are using claw these these the claw p or the claw SDK. These are things like open claw. These are things like um open code open code right and I think open code is probably a slightly more because it's still in the wheelhouse of code it becomes um you know uh more favorable on the left side as it is the right side. So there's the scale of of what we're talking about here. And again this is totally stolen from Theo. So maybe go watch his video is that they drew a line a few weeks ago right here where they were like if you're using open code or using open claw then we're going to charge you API usage. It's going to um rightly or wrongly steal from a you know as soon as it detects that open code or you're using open claw it's going to take from like a bucket like a purse of cash um that you've set up and sometimes it will warn you sometimes it won't and there was a report that I think they got refunded but um Hermes as an example which is an open um openclaw competitor the uh that there was a there was a Hermes written in a git comment in their in their code base and uh Claude Code picked up on that and decided to decided to charge them 200 bucks and they didn't know anything about it. Um I think that got refunded but you see this kind of that was the kind of I think the point in this whole equation which triggered in anthropics mind that they can't accurately detect whether you're using open claw or whatever. So, a new line needed to be drawn and they've now drawn that line over here. This is where they draw the line. Basically, anywhere outside of any sort of clawed UI is where you need to start paying for it to sweeten the deal. They they're going to be introducing as of today and you'll probably get an email or have already got an email today saying that you get the equivalent of your subscription in usage. Sorry, in AP it it's basically in API tokens, right? Hang on. Hang on, hang on, hang on. you get you get um usage um you get a purse or or a pocket of of tok um cash to use in these circumstances, right? And funny enough, I actually started to use P-P in a couple of cases because it's quite nice just to send it off, do something, and then not have to worry about it because that's exactly what it does. It just it doesn't even enter clawed code. It just does its thing and it doesn't and it just exits straight away sort of thing, right? So, you get a bit of money to spend in these areas. The moment you use this, it takes out of this extra purse. As you just said, the problem is it charges you API usage. So, if you've got a $200 a month plan, I've got the $100 or if you got $20 a month, you know how expensive Claude is, right? That doesn't last very long under API usage. It's still nice, don't get me wrong. It's better than what we have. Yeah. but or it's more clear what we have and we'll get into that but it's not not necessarily going to last you that long. Um, I mean, this wouldn't be the end of the world if they hadn't already had this, you know, had this line that this was all, if this never existed, it wouldn't be such a big deal. But now they've had to draw a line and and um I've forgotten his name, Matt Pulan, something like this. He said it really well in that it's such a relief that we have this clarity now. And this, you know, again, Theo explained this and it was such a sigh of relief just to you can clearly see this transition. Now, I should probably remove this um line. Yeah. Um it's nice to have this clarity because it really like I bet when the anthropic guys come up with it, it was like, "Yeah, this makes sense." And it does make sense, but it's still a kick in the nuts, you know? still click in the nuts that that it's almost anti- you know a lot of people are building their own harnesses around claw code and and all the rest of it and at the end of the day you paid you're paying anthropic money every single month you should be able to use your tokens however you want however they're so subsidized that they're probably you know they're they're losing so many tokens um open AAI taken the complete opposite stance they are you get your open AAI subscription you can use whatever you want with it. You've bought those tokens and whatever. I don't know how long that's going to last because we've reported in the past that OpenAI are taking the initific not the initification model but the the startup model of just giving you everything just like Uber did, just like Netflix did, just like all of these other companies. They just give you everything for a short price until you're hooked and then they take it away. So that's and and that's all Sam Orman knows. We've we've reported on this in the past. So that's long and short of what's going on. That's how it kind of, you know, the landscape of things discuss. Yeah. Well, no, you explained everything perfectly and it is Yeah, it's quite sad because I was uh enjoying T3 code. Yeah. So now T3 code has been updated to basically run clawed code inside of it UI which doesn't it's why would you do that? It defeats the purpose, right? Yeah, it it looks like an iframe. It defeats the purpose. Uh you can still use it just for clarity, but it's like you hit the limit 40 times faster or something like that and and then Yeah. So it's pretty much unusable. Yeah. Um, I think this is this speaks to the reality of how expensive these AI models are, how expensive they are to run, to train, and to run. And we've been getting this subsidization from Claude, but they just can't do so much because they cannot offer more compute for free. Um, and they have to give it to their enterprise clients and that's what they are doing. And it it it sucks. It sucks. But they've been giving away way too much basically for free, right? Like you can you can draw like what three 5k worth of tokens within a month. So you pay 200 bucks in one article, but it's Yeah. Everyone's different. So you pay 200 bucks, but you use as much as 1,500. No, 15,000, right? Not 1,500. 15,000. Yeah. Yeah. That that that's crazy. So they subsidize that by, for example, me not buying the 200 bucks plan, but not using it all that much, but you going nuts and using all of it. But they can't do this forever. And the point is AI is more expensive than uh these companies wish it to be and more than what we wish it to be. And we are facing this reality right now and everybody is kind of waiting on maybe another model coming up uh you know being much more token efficient. Maybe the these companies are working on that. They know that they can get it much more efficient because we likely need way more tokens, way more automations. Like open claw was just the start of it. Like everything will become open claw if if we would have enough tokens. We just simply don't have nearly enough tokens as much as like we we would like to have. So that is going to happen. And this kind of like uh speaks nice to uh the the next topic like the next video which we have about the 12 million context um model. Mhm. Yeah. Um but yeah. Yeah. Just to wrap up on this this though, um like the interesting thing is is that Claude called like there is you can do all of this with everything that Claude has offered you, Anthropic has offered you, right? You got your routines, you've got um you've got Claude code for code, you've got schedules, you've got loads of things, managed agents, right? Loads of things they offer you. I think the biggest I mean, you know, Theo was really angry, but unfortunately he's he's kind of biased because he the wind got taken away from his sale of his product that he was pumping a lot of time and energy into, right? whether he's got this secret master plan of actually wanting to sell it in the end even though it is open source or whether he's got, you know, whether there's something along those lines or whether the fact that he literally has just pumped a load of time and money into building something that's now again [ __ ] But they they there are better ways to do things. The the clawed app, the desktop app, clawed code, they they are a little bit buggy. They aren't the best. And people were building clean alternatives, you know, that that people prefer to use, right? And they could do that with their Clawude Mac subscriptions, which was a win-win. It really fueled and um inspired the developer community, right? And with these new changes, they've basically shunted the developer community. the even though they provide core code, the real power users were using core-p or the or the agent SDK um and those people are the ones most hurt by this. Um yeah, but it is what it is. And uh I do I do see why they've done it. I I they have every right to say why, you know, we'll give you we'll give you subsidization as long as you use our tools. There's probably an analogy out there in some some way, shape, or form. It's like, okay, why would you go buy a car, but the moment you put in an engine, a different engine in it, why should that come car company help you out if you break the car, you know, or crash the car or something like that, you know, it makes sense that they are taking control of the products that they offer. Um, and if they'd never done it in the first place, it was if it was never there in the first place, I think this would be a much easier pill to swallow. But because it was given to us then kind of taken away and then given to us and then taken away again. Yeah. No, but the clarity is there which Matt PCO said he used a term called away from keyboard. Anytime you're away from the keyboard um is where they're going to punish or punish. I shouldn't use that term. Is where they're going to charge you. And that makes sense because Open Claw made it so you could have agents running pretty much 247, you know. Yeah. So anyway, that's the end of that one. In good news from uh Claude is the agent view which is it's kind of related to what we just spoke about the uh the whole subsidization thing in that uh cord agents allow you to effectively monitor all the sub aents all the agents that are running on your machine dep doesn't matter what project you can dip in it will tell you whether something's uh needs your attention or anything like that. Um, it's great, but what what better way to than just just to I can't even say it. [ __ ] hell. We should just show you, right? We should just show you what's been going on. So, oh [ __ ] I've just removed it. Oh, no. Hang on. I just Here it is. Here it is. Um, so it kind of looks like this. Um, can you zoom in? Uh, please. Let me see if we can Yeah. Uh, I don't want to The audio is on top. So it looks like this. You can see it. Yeah. Um bloody blacklisted, didn't we? Now, uh at the top here, you've got needs input. Um, and you can dip in and just have a look what's going on and dip out and then go to your other agents, see what they're up to and just go through them. Tells you what's completed. Really nice way of um, managing all your agents, especially if you've got different agents doing different things, running on different work trees, running on different projects. It's a welcome feature, but again, this removes developer tools like T-Max, you know. So, I think it's a byproduct of then locking down what you can and can't do with Claude because T-Max will um allowed you to effectively have an agent view. I'm not sure how it tech technologically it does it, but um yeah, it might be that they're trying to lock down any sort of outside interactions with Claude code or but you know that whole thing that we spoke about. So, um, it's that it's pretty cool. Peak without replying. So, you can dip dip in and and and reply to them if they're asking you a question. What do you want to do next? And this and that and the other. Um, if you're running a if you're running a a task or whatever, you can just send it to the agent view by typing /bg whilst it's doing a bunch of stuff. Uh, it's quite nice. It's quite nice. So, I was having a play just before the the uh show and I was able to again spin up a spin up an agent doing some stuff in one project slashbg go to another project slashb you know it's good that you can work across multiple projects. It's not bound to the specific folder that you're in. It's literally its own little interface. So, it's so it's pretty darn cool. So, bit of good news from uh from Anthropic there. You're are you actually what what I always ask you this because you're you're jumping around. What what are you still paying for Claude? Are you off Claude? Are you using GPT? No, I'm using the 100 bucks one now. So, max $100 and I'm using warp 95% of the time this week. Uh yeah, and I'm using the vertical tab and I have just multiple instances and each doing its own thing. But you're on the free version of Warp, right? No, you're not the one that we got. What are you doing with Warp and your AI? I don't use the Warp AI. Okay. Yeah. Yeah. Yeah. So, you're just splitting it. But I like the the UI. So, I I'll show you on my on my video. Um uh the old you go it looks like this. Yeah. Yeah. So each of these if you're running something you can just do slashbg and then in another tab in another warp tab just type um claude agents and then you just be able to see all of the agents that are running. In fact you might even be able to do it right now. Just type /bg in any of those preferably one. I don't Yeah, I'm not um updated maybe. Okay. Yeah. No, no, no. Type it. Just type it. I think you can just type it. Where? Like you mean not inside of Claude? Inside of Claude. You can just type it. Yeah, I'm typing. Yeah. Yeah. Press enter. No, it's Yeah. Okay, fine. Um, it does say that you just type it to enable it because it's a beta preview, but you still also might need to update. So, um, yeah, don't. But the the point here is like, um, you just I'm sorry. Hang on. Whoa, whoa, whoa, whoa, whoa, whoa, whoa, whoa. Boys and girls, let's freaking hold the phone. Did you just prompt your agent to cancel? Yeah, because I was uh the escape key. Yeah, but I wanted to know that, you know, have the context of cancer. I do use it. Yeah. Yeah. Um, mate, your brain is fried. Your brain is fried. But, oh man. Um, but doesn't does it know if I cancel? Which mean does it know if you cancel? because it started to create an experiment and create an experiment is um is a skill that I have made. It's semi-skll. So it starts something that I wanted to cancel. I didn't want that thread to go there. I mean realistically maybe to tell it it's not going to I don't I don't think cancel I know what you're saying. I don't think cancel is going to do the thing that you think it's going to do. I think it's going to just work out. whole that whole workflow just got truncated. I think that's all that means. Yeah. And if I press escape, I don't know what it happens. Maybe I should That just stops it. That just stops it and it's re you do that. Yeah. But when I send the next prompt, that's my worry. I don't want my context to be, you know, I think I think I I think uh Claude will work out that that's not it's not gonna like that tiny little that That's the thing that was that your experiment that was running for several seconds. I don't think it's going to affect your the the context. It will be in there, but I don't think it's going to affect it. I don't think it's going to get So, which one is worth worse? Uh saying like uh by the way, I don't I don't do this anymore. I I just you know I just myself I think cancel is worse because there could be many ways you run the server. It could be a docker. It could be a start the front end. It could be start I can understand why you might run say run the server but arguably you should understand the project. Cancel is definitely the worst. Okay, for now, for now, until now, we will see. We will see like how how fried uh my brain will get. I mean, for each window, I have like this many agents. Yeah, but I do clean up. I do clean up. Yeah. And you might find yourself doing that, setting them all off and then just jumping into agents agents view. Do you know what I mean? um and just managing them like that because it's really really quick to dart around into the agents. Um let's move on because I think we've spoken all we can about that. So this this is a demo I want to do. We spoke about Perplexity um we spoke about Perplexity personal computer last week and it was a beta. I was I was uh and they didn't even bloody send me an email actually. God damn it. It's a tool that I think is the Mac uh Siri Apple Intelligence that we never got sent set off my bloody uh iPad again. So, let me just show you cuz I I I started this because I wanted to do a bit of research before the show and then the intro looked really cool. So, let's just let's just Did you hear that? That's my uh thing going off. So, let's let's fire this bad boy up. Um, entire screen this one. We're just going to do that and go perplexity. Hopefully, you can hear it. I don't know whether you can. Allow me to use the mic. Can you see that? I see it, but I'm not hearing. Can you hear it? No. Okay. This is delightful music. Okay. That Well, the design is really good. I mean, designers at Proplexity, they've been always good, right? Can you hear this? No. Can you hear that? No. No. No. Let me try one more thing. I don't hear you. I can't hear you. And I don't hear the screen either. Now I can hear it. Yeah. So, here we go. Press both keys. Yeah. Ask anything. So, here it looks like it's contextual. It's like linear but like knows what you're doing. And we covered clicky last week as well, right? Hold function key. Yeah. Okay. Well, let me uh sign in off screen here. That was nice though. These intros are getting very like elaborate now and and they kind of taking over your whole screen. So I should be I like them. Yeah. I the first the first one that I saw that was really good was from Arc. That's right. That's right. And this whole dragging thing as well. That that's really cute. Yeah. All it's really doing is just switching it on there. But if we do that, granted, enable screen recording. Here it is. Boom. Boom. Boom. Quit and reopen. So, it should be able to Can it control your Mac? I mean, well, this is what we'll try. What do you What should we do? Should I say um don't know what should we do? Okay. Oh, that's really satisfying by the way. Yeah, but I don't know if I like the command command thing because it's kind of like two keys that are far. So, for for myself, I have I like the Fn key press twice. I have the control plus space bar. I have option plus spacebar. I have command spacebar. And I use these daily. Mount refill. Oh, to do one command. Command. Yeah. What's 100 credits? What's $10? No, thousand credits. $10. 1,000 credits. Um, well, you can probably change your uh change this now, actually. Yeah, you probably change the uh the the form the shortcut, you know. Yeah. Um if let's go to this. Okay. Okay. So, it's like it's like uh so I would probably that's that's my double command there. So, okay. Or so I'm holding. Now I'm holding and and speaking. You can see this thing at the bottom here. So that's cool. Stop that. Holding what? Um uh function function. Ah okay. So I can speak to it because here here's the thing as well because we spoke about clicky and I have not got this to work. Like you're supposed to be able to say hey clicky agent and it will just do something for you. But it's not worked at all. And I'm paying for this as well. But they've moved it up to the the top here. You see that? That's quite nice. Yeah. And like it feels my mouse my trackpad clicks when you do that. It's really satisfying. But I've never never got clicky agent to work. I um really disappointed in that actually. And I know it's a new product, so just bear that in mind. But uh yeah, I don't know where it is now. Yeah, I want to I want to quit it. Um, but it's it's the same sort of thing. I I think I should be able to um maybe I can go um find the pricing of Claude. I don't know whether I should maybe it's not picking up my microphone or something. They should show that. Um, I like how with whisper flow you see that if it's not working, you see it. So, you saw there that I had to top up my credits. So, this is using a separate crediting system that's outside of my Perplexity um, uh, subscription. Yeah. And you know what? I have a bone to pick with Perplexia. I've given them this feedback and they um I use comet browser here and I'd often go into um assistant and get it to do things on the website that it's on but then it just stopped working and basically it's that same thing. They don't they want they wanted to hide um any sort of use of your computer or your your browser behind the payw wall. So, they started paywalling everything. This is supposed to be Hang on. There we go. This This over here is supposed to be um just for you to chat with the website you're currently on, not actually in not not get it to actually interact with the website that you're on, which is a real and I shared this with them. I said, "This is a real down thing down grade because you'd be chatting away and you'd be like, "Oh, how do I do this? How do I do that?" Blah blah blah blah. and okay well well claude does this. It says do you want me to do that for you? You say yes. It takes over your browser and starts to do things. But now you've got to go you've got to open up a new tab. You've got to go into computer mode and then kind of start everything again. You know it takes you out of that flow or it takes me out. That's my that's the way I like to work. Takes me out of that flow. I've also just noticed that uh computer use isn't really working. I mean, I'm under the I'm under bit of pressure here to try and get it to do something, but um what's that? That's your credit system. So, I'm not really digging deep into it. I don't necessarily want to do that, but I do think this is quite nice. And I they did have a let's see, I can't go to settings. They did have a floating uh bar which they they spoke about that that that was the thing that that I thought was really nice. It was always this floating bar and you could always like interact with it. I suppose that's quite nice the double double command thing. Um and that's even listening as well. So I guess you could say um could we try the whole clicky experiment? uh create a HTML uh presentation on this Claude article. Yeah, I reckon it's got something. Yeah, I reckon it's one of the reasons why Clickie isn't working, too. I reckon it's some sort of external. It might be my system that I've I've personally got set up, but we'll leave it there. But still a nice tool. I do still want to dig into it, but I do I do have a bone to pick with Perplexity. And I also um have cancelled my subscription to Perplexity because they've completely nerfed my favorite browser and all of its AI goodness. So, yeah, don't know what to make of it, but uh I could do everything I need to inside of Claude. And going back to what we were saying earlier about Claude actually having a lot of features involved. It just might not be the best implementation of it all. Um I can do a lot with Claude. I can get it to control my browser and everything under my subscription. So yeah, Plexi supercomputer pretty cool. Take a look into it. I will be doing so. But um sorry I couldn't get anything to work. Finally I can stop talking. Subract sub subq, do you want to do you want to give this a whirl? By the way, well, I can try. Have you as you've already said, none of us are qualified for this, but we'll just look we'll just read the article and we'll just respond to it and you know um just go. So yeah, so this is way above my pay grade sub quadratic uh agent. Not agent AI. Wait, actually let me subi. Yeah, it is sub quadratic. I think that's right. Yeah. So let's let's I'll start from scratch. All right. So this is way above my pay grade pay grade uh subqai a new startup a new model out is out and they are claiming they have 12 million token uh and like context window basically that they can have and they are better than pretty much any other mod out there. So um they have a really good video. I don't know if we can play the video or do you want to play the video? Play the video. It's just the audio. But you're you you've never got audio to work on your machine anyway. So, just play it. So, the video is really good. Yeah. 12 million tokens. And this is like when the context window is at 1 million tokens. Yeah, they are better than others. And and and it can process the tokens faster. So it's faster, smarter, and has a bigger token window, which is crazy. Do you smell that? No. Smells a little bit like [ __ ] So, have you registered for early access? Yes. I haven't gotten it and I did this like over a week ago. So, here's something that I barely understand. So apparently with the standard we have the GB GPT models that we have each word is like calculating its distance like in the vector space with every other word in the so like every word is connected to the next one like and all of them are connected to one another. So when there is a word, the next word is predicted based on the previous word. So that makes it like really difficult to scale the context window because you can imagine the last word in a 1 million um context window is related to the first one. So it's a really uh huge calculation to be made. But in this way they are only doing the calculations between the words that matter. This is kind of like also how we humans do it. Do it by the way. Like when you read there there are like methods to speed reading. You can read a page on a book and retain a good like understand a good chunk of it without reading the whole thing. There are methods to it. So you just read the important words and you kind of like read the what's making up the context and it seems like that's what this AI is doing where only some of the words are connected. Mhm. Uh so anyway, so it seems like they can scale now some I I saw some tweet some tweets people saying this is a [ __ ] why they were saying if what this guy is claiming here if it's true the 12 million token uh context um window is too small. They were saying if it's like truly what it is, what uh this guy uh is claiming it to be, it it should basically scale infinitely. Uh and why why stop at 12 million? Why not go for a 100 million? Why not go for a billion? So that's the type of tweets I was seeing. But like scroll down a little bit. Yes. In all fairness, keep going. Yes. And in all fairness, this guy, I believe, um the CEO or co-founder, he replied and said like it's not just the architecture that the new architecture that that we have. We have like other limitations. So we can't literally just take whatever model is out there and change the architecture and scale to unlimited. the there are like other problems that um need to be solved as well. Anyway, theoretically, if it's true, they should be able to achieve much bigger scales than 12 million and much faster speeds. And here we see some interesting tests. Obviously, these are handpicked and from what I understand, it's uh just done by the company. So uh we will take it with a huge grain of salt but uh Sweepbench verified like Opus is at 80% and they are at 81 but Opus 4.7 is at 87%. These are at 81. I don't know why this is shown as green. I think just Yeah. Yeah. Exactly. It should be more right. Yeah. Yeah. So 94% uh at this one. 95. Uh so it's better than the other models at at some some of these uh tests and tasks and I have asked for early access but I have not gotten uh access yet. So remember it's it might be less but uh less than Opus 4 uh 5.7 or whatever it is but it's one the cost you know the cost is um is something to to take note of. Um I've I've signed up as well. Uh there will be two ways you can interact with it. There's the code by the looks of things as well. So you can have it'll be like a clawed code um plug-in sort of thing um or directly through the API which would be for building tools and things like that. So yeah. Yeah. Small company by the way. It's a small company with not a lot of funding and from you know it's not from Silicon Valley. So a lot of people are like hm is this real claim and I can't work out whether it leverages existing models or whether it's their own model because a small like have they trained their own model not completely from what I understand they are changing the archite architecture or they are distilling yeah so yeah yeah so they've they've Well, I mean, you can't Yeah, obviously because they're comparing against closed source models, they're obviously you can't distill those because they don't have access to the the weights, right? Um, however, like they could they mentioned a claw code plugin. It could be that they're wrapping is is something they can achieve like that. I'm not it's yet to be seen, I guess. But, um, it's cool though. Uh here they say internally uh evaluated. So for what's that worth? And by the way technical report coming soon. They it this is from a week ago still coming soon. So I I would like to see that there are like some third party results. Um, but it seems like where this can shine is like at higher context windows. And yeah, this can be potentially very interesting to to do some like heavy working with. Um, we didn't report on it so much, but it's something I've been exploring on my channel, which is Turbo Quant, which is a way of quantizing the cache, right? And uh, the one by Google, right? The one by Google. Yeah, we did cover it, didn't we? We didn't I I don't think I understood it as well. I actually played around with it. There some videos on my own channel if you anyone wanted to go watch those. But um yeah, that was an interesting way because obviously it doesn't it's it's it works with any model, right? You just quantize and you store the cache and quantize the cache in a certain way based on your parameters you've you've set. Um, so and I think that's the only real like step we've had in terms of trying to make the existing models more efficient. Like I say, I don't know whether this is their own model or they are just wrapping, you know, clawed code and and working with with Opus, you know, um, and storing the cache more effectively or or doing something with the model on its way out. Don't know, but it's cool. Cool. Web flow. Yeah, let's go to Webflow. I'm preparing I have like my comparison setup. Um, we still like to cover Web Flow on this channel, you know. It still holds a pace in both our hearts, I think. Yeah. Slowly but surely, you still use it. Yeah. mostly. Well, I use Claude and Claude uses Web Flow. The Claude us Yeah. Yeah. Your agents use Web Flow. Yeah. Uh let me zoom in a bit. So, yeah, that's get getting the ads out there. All right. So, Web Flow has a new pricing plan. And wait uh did it before before actually I want to restart. Did it happen like less than a year ago? What happened less than a year ago? Like um pricing change. I don't know. I don't know when the last pricing change was. I Yeah. Don't even know. Anyway, all right. So, Web Flow has changed their pricing again. And here's everything you need to know about it. In very short, you don't really need to be worry about it too much. So, I have a full comparison for you. This is the old one. I took it from web archive. And then this is the new one. Okay. Uh let's zoom out so you see it better. Okay. So, this is the new one. You have starter basic premium. And in the old one, there was CMS and business. In the new one, you see this platform plans with team $2,500 a month. Let's let's just put that on the side. That's something else. It's really stupid that this is next to it doesn't make sense. So, let's ignore this. They're calling this platform plans. Yeah. Web Flow, I know you're you're watching. I'm really sorry, but like you're really bad at simplifying your pricing. Your pricing page needs a course. That that's how bad it is. Anyway, I've been doing this for years, so I can't explain it. Anyway, so let's just forget about this. We have the premium now. Previously, you had CMS and business. 90 plus% of my clients are on on the CMS plan. My website is on the CMS plan. And just to be clear, Web Flow have two separate plans. You have a you have a probably what they call a platform fee, right? You have a No, you have site plans and work. You pay these. Yes, that's the one. Workspace. So, these are site plans. These are for each individual site that you might have on a plan your Exactly. Thank you. So you you have a website, you connect it to a domain that's that's going to require one plan and you have another domain with another website that requires an additional plan. So previously they had CMS 90 maybe even 95% of my clients are on this a few on this and a few on the business one. Why would you go from CMS $23 a month to 39 for business? What is the difference? The main difference for at least for in my case it has been form file upload funnily enough. Wow. In the in these you cannot upload a file like let's say you have a form the the end user needs to upload a PDF upload an image to a form can't do it with these plans it that's why you had to previously go to the business plan and that's kind of like a good jump it's a big jump to go from this to this from 23 to 39 just for form file upload. load. But yeah, you would get also 40 CMS collections instead of 20. Now that has been simplified and I like that actually. Now all both of these two are just called premium and web flow in their blog post they show it pretty well. So the starter is the same. The basic has a price increase from 14 bucks to 15. The CMS and business are combined 23 and 39. Both of them are going to cost 25. So it's an increase. If you are coming from the CMS, 95% of my clients are going to pay $2 a month more compared to, you know, what they what they were paying for basically no benefits. They don't need it. But it's Yeah, it it sucks, but it's not a huge increase. You're probably gonna go into this, but what do you get for 20? Yeah. Five. Yes, I will get into it. So, for 25 bucks, you get pretty much everything that you were getting from business before. That's the good news. But you don't need it. Maybe you might need not need it, but I I would argue that it's nice. What what's a downgrade from business is the the bandwidth. So at $39 you would get 100 gigabytes bandwidth. At 23 you you were at 50. Now with the 25 from premium you just get 50. But yeah, if you are if you are at a 100. Yeah, it's it that's So what's the what on the old plan? What was the bandwidth of business? 100. 100. And what's the price of 100 bandwidth in the new plan? Did I see 45? That's exactly. So that is a jump. So if you if you were on the business plan because of bandwidth because in my case it's form upload but in your case it might be because of bandwidth or CMS items or whatever. Now the bandwidth is going to cost more. What's really good is with the CMS you get now 40 collections and 20,000 CMS items. Yeah, I know I've Yeah, I've said like I've talked about uh Framer pricing enough, but I have that just in case you haven't seen Framer at $100 a month gives you 20 collections. Web flow now at $25, like one4th of that price gives you double. So like fra framer is eight times more expensive when it comes to the number of CMS collections. Um and not to mention others like at 30 bucks which is still more than 25 you get 10 collections where with web flow you get 40. Anyway they're not talking about framer but I thought it's kind of like re relevant if you are looking into web flow pricing you might uh be also looking at framer. Framer is not cheaper. So that's it. That that's pretty much it. You I saw people being angry on on Twitter. I don't get it. Yes, it is more expensive and I I don't welcome it. But in my case, because I was I I had to go to the more expensive plan, the the previous business plan just because of form file upload multiple times. I'm kind of happy and I don't think that two bucks a month is, you know, a huge jump. So, hang on. Right. You're telling me you spent an extra what is that? $13 $16 a month for form upload. Yeah. And you couldn't build something that uploaded it and then posted it in like and just stored it in a that link in a in a database entry. Uh because it's web flow like all web flow integrated like that. That's why we we were like just using web flow form to streamline stuff just just like you're so you are basically saying like integrate like something inside of the web flow form right no just have a just or like don't use web flow form at all use a separate service so your whole entire form submits via JavaScript you send the for the the file upload load whatever to a separate service. You send the rest to Web Flow. That separate service then plugs into Web Flow CMS and says, "Here's the URL of the file that I uploaded." Yeah. Yeah. You could you could you definitely but we we're if if you see the sites that we've made um no I wouldn't I I would still pay this and not well maybe today with claude it would be easy but we've done this in the past and you want to keep the consistency and the you know the rest of the the UI of the the complex forums that you're building because the ones that we did were not simple forums. still quite complex ones. Yeah, we're off topic. However, what I will say my my perception here is that and I'm surprised they've gone up for the basic plan for this reason. I think they've brought down the price of everything that they can control the the the known knowns. They've said right we can control this so we can set a price on it. That is number of CMS items that's using the platform that's this and that's that. The unknowns is where they've increased the price which is bandwidth. There was a whole thing about two three years ago around the bandwidth being slashed basically you know and that was obviously killing them as a business right and it makes sense because they don't know whether you're going to build the next Facebook on web flow or you're going to build uh your auntie's um trip planning website and it's going to get no viewers. Yeah, it's going to get barely any views. You know, they just don't know. So that's why and they're they're they're really competing with clawed code right now and you know any other AI uh tool that a lot of people are jumping to. Um so they need to bring down their prices of their tool because otherwise people will just turn to to cord. Um so that's that's my observation. Yes, there there are a few more things that I want to mention. um 100 gigabyte bandwidth what the previous one had and 20,000 CMS items and 40 CMS collections is still going to cost 40 five bucks a month but on on the previous one if you had if you wanted 40 collections and 20,000 CMS items and 100 gigabytes that would cost 89. So, it's it it really depends on your use case um to know if this one is going to be more or less expensive for you with the new pricing. But if you were on the CMS plan, you just know out of you just know it's it's two bucks a month more expensive. And if you have multiple sites, that adds up. So, that sucks. Now another thing that kind of sucks about the pricing even more especially for agencies sometimes well nowadays maybe less I would argue I can understand this but in the past we would set up the site for the client and h added I would we would add hosting for one month for the last you know the last few weeks of testing and you putting the content there. But before sending it to their workspace, we would add one month of month of hosting and on our um workspace which we would typically pay and just you know uh we would eat up the cost. Now that's going to be a lot more because you see save up to 40%. Previously it was up to 33%. So, previously the basic would cost 18 bucks a month if you would pay monthly. Now it's 25. So, ouch, that that's a big jump. The premium is 39 while the CMS it was 29. So, it's a good jump for the monthly, but honestly, yeah, if you have a website at most I like all my clients are paying yearly. Nobody is paying monthly so it doesn't matter. There there are some cases that you would pay monthly. Uh just to mention that the e-commerce plans still exist. The workspace plans still exist. The between the the differentiation between teams and freelancers still exists. And this sucks. Like I understand it, but I see so many people just don't don't get this. uh we we understand in in the in the community because we we have grown our businesses with web flow for so long but yeah uh the the pricing is pretty complex here in the blog they also compare all of them so there are also a few more things that are that have been increased and this is kind of like the backlash I've been seeing from the community on Twitter and I understand it It's that these are things that nobody cares about and now you have to pay it anyway. So for example, a starter now costs more but it has double the amount of static pages from 150 to 300. The argument is nobody needs this many. And yeah, I don't I don't know any website that like what what are you do? Why would you need 300 static pages? Like I understand CMS like the moment you you need more than what 20 the moment you need more than 20 maybe maybe 30 pages static you you have to go CMS like you you're you're going to build these pages systematically. So anyway this is something that you don't need but now you're paying for it. Same here. Uh the 20k uh item limit from 10k now it's 20. Yeah, you probably don't need this. The 40 uh CMS collections. I really like this. I actually we use a lot of CMS collections. And yeah, there are like Web Flow CL cloud. I don't think that matters that much. And now about this $2,500 a month plan. What is this for? Who is this? From what I understand, this is kind of like between the what what now it's called premium and enterprise. So inter with enterprise you get you get a huge bill. Now if you are a team and you're not ready for enterprise but you still want to have like hosting and workspace and some advanced publishing and yeah a bunch of things like this you're going to pay this but why is this next to site plans? This is what I said like web flow is just so bad at pricing. Um like look framer just so much simpler just so much simpler and I do think people get confused and they might not sign up because of that but yeah that's been it that's the web flow pricing not a big change honestly uh I'm not too pissed about it. Well, it's good to have that clarity. No changes to the workspace plans, right? Not that I know of. No. No. Cool. Cool. Right. Well, I've got to shoot really soon. I've just been looking at the time. So, let's go to unreasonable effectiveness of HTML and then if you want to Well, yeah, I I'll have to I'll have to shoot probably midway through it. Yeah. Okay. So, um how should we do it? Just you just do it. Ah, I should do it. Yeah. Okay. I can do an intro real quick if you want to sort some stuff out. Let's Let's do the intro. Yeah. Shall I do the intro or? Yeah, do the intro. Okay. Let me share my screen then real quick. Okay. Um, why can't I Why can't I share my screen? Because my screen is on the No, as in like I want to change the Oh, because I'm scar I'm sharing my screen, not my tab. Um uh so Derek from Anthropic has raised a very very crucial IP [ __ ] in it. Um no I haven't got I haven't got an intro. I haven't got intro in my head. It's okay. It's okay. Um I can go about Okay, cool. Hang on. Go for it. All right. So, TK from Anthropic, he shared a very interesting article, use HTML instead of MD files. And we'll go into it and I will show you comparisons between what's an MD file, what's an I presumably know these, but I will show you next one another. Uh the idea is that with MD files agents can read it easily and it's quite efficient in usage of tokens but with HTML it's nicer for us humans potentially even for agents it's not as efficient but it has like more context a better context and for us it's much easier to read. So why HTML? Uh there is an argument to make uhformational density. You can do just so many types of um like show show informations in so many ways. So HTML can convey much richer information compared to markdown. It can of course do simple document structure like headers and formatting, but it can also represent all sorts of other information such as uh basically like tables and design data, CSS, SVGs and code um and just so much more. Actually, I'll just show you an example that I prepared. So here we have three basically views. On the very left we have a markdown file not rendered. So this is basically the raw information and you see this hashtag and then that's basically your H1. One hashtag means H1 essentially. Then that's how the rendered one for human. So there are tools that do this for you. For example, on my Mac, if I show you the same file, I have a I have an app that renders the preview for me quite nicely. If you don't have that, you will see it just like that. When you open it on your Mac or on your computer, you see the file just like that. Now, you see this is quite simple, but it's nice, right? It especially this one. This one is not very readable. like good luck reading this. I think that I think there's a there's you've got a to you've got a missing bracket or something like that, but that should render like table. Yeah. Yeah. It it's the this it's because of the spacing like it's smaller like Yeah, but it's it's still not really easy to read. Like I would argue like this table here again this table you see because of the information density it it the you know the the length of the information it's just not easy to read but this one is easier to read. So why does this matter? This matters because you want when you are talking to your agent, you want to have like a format to communicate with which you both can read it effectively and you know communicate it effectively and when I say effectively like no information lost but also uh in tokens you don't want to use too many tokens. Now, now the idea from TAK is which by the way I've been doing this for quite some time. I'm really happy to share like I've been asking my agents to just output HTML to me instead of MD files quite often. Now this is that the same the same file on the left but rendered as HTML not exactly. So I told the agent to take this MD file and create a rich HTML uh based on this MD file and present it to me. You see it's just much nicer to look at. For example, look at this export UI panel. This is the plan for a panel. It's just a very rough diagram or like UI here. Here you have a proper UI. you can oh it well it doesn't like click but you could tell your agent to make it interactive right it it would work so that's really nice about it then when it comes to like rich information or like coding it's just much nicer to have it you can have like diagrams and so much more and actually TK has made a a a really good example that you can give to your agent to make to basically instead of using MD files or creating MD files, create HTML with kind of like this as an example. And this is really nice the types of things that you can do. For example, here like a this is a really good example of something that you can do with HTML that you can't do with MD files or not as nicely or which one was it like a clickable flow. This is you know this works. I mean there is a JavaScript here uh involved as well but you get the idea or like with SVGs I do this by the time by the way all the time I ask my agent to explain something in diagrams or in SVG and they are really good at it they like this you can imagine being just much easier to understand as a human but also the agent will understand it just fine now what is the issue with that tokens and people were making jokes about uh Anthropic having the deal with SpaceX first and then publish this because this is going to consume more uh tokens. How much more there has to an argument to be made and I don't know the exact answer and there is a test to be make made if I'm chatting with my agent the the likelihood is that there is a lot of tokens that are going to be used just for the reasoning I don't know the proportions that's my point like I don't know if it's even if an HTML doc is what's happening even if an even if an htm ML doc is 10 times more token thirsty than an MD file. Even if it's 10 times more token thirsty, it might not matter at the if you look at the entire um interaction that you have with a cloud code instance because of the reasoning and what whatnot. I don't know. I would actually like to maybe test that. But yeah going back to the article which we will share there there are like other arguments to be made like visual clarity and ease of reading as I mentioned it's just much nicer with hierarchy of information uh ease of sharing go to the use cases I think is quite an interesting because uh you've kind of already discussed those ones okay well ease of sharing by the way like we can also like ease of sharing like with an a you could share an MD file but with an HTML you could just quickly put it on Versal and share it with anyone on a group chat that's that's like a big deal I would say so here when you say use cases you mean here or like in the article well this well I I'm surprised you didn't clock on this is a this is the article done as a HTML file oh it I It is the art that way. It makes Yeah, it makes sense. Yeah. So the use cases are are in this exploration and planning is a use case or code review and whatever they're all the use cases. So if you scroll on to use No, it's not. If you scroll on to use cases. Ah, but okay, but it's like where's where's use cases? Well, these are the use cases. So if you go back to the article, scroll down to use cases here, specs, planning, exploration, you go back to the HTML. Yeah, but like this is a different file. Oh, the the exploration, planning, code review, design and prototypes. They're all in the the sharing. Okay. So, it's not I see. I see. No, I haven't been that. Yeah. Okay. Okay. I I understand what you're saying. Yeah. Yeah. No, I was like at the start of the article kind of like introing like making the argument and then Yes, you're you're right. Uh then we get into the use cases and the exact use cases are these these are these are how you could use it or like the benefits basically. Mhm. Uh is there one that you particularly in um I think they're all pretty good. No, no, I think they're all the one the one I find hardest uh to to get rumhead around is the review git review stuff. But uh they're all pretty good. So, I think they're all good. Worthwhile going through. Yeah. Yeah. And uh I think the one with uh custom uh custom editing interfaces like these are like such a difficult thing to convey through text and an MD file is just one level above like a normal prompt like one level higher in fidelity above normal prompts. and it's better than prompting but you know an HTML can be much easier to to read for both and and to share. I think that that's what people mostly forget about like you go to skills right um like skills are MD files right it's really difficult to read a skill I mean difficult it's not difficult but it's it's a long text that you but I would argue you don't you don't necessarily need to interact with a skill but your point no being that uh because it's purely for Right. Yes. Yeah. But my point being is like if you want to share something and make it also human readable and HTML is just superior and I I don't know for sure but I don't think that tokens are going to take such a big hit as people are well argues Yeah. Well, argues that well, yes, more tokens, but like you're you're making up for what it produces being so much more clear that you're not having to go back and do anything with, you know, edit the a markdown file even more or this or that because it's not laid out in a way that human wants to read it. So, you know, um it's arguable, but um probably negligible to be honest. I'm not too sure. Yeah. Anyway, give it a try. Next time you're planning something with your agent, like let's say you're doing a marketing, this is what literally I did for a client of mine. We we started doing marketing research because he wants to write blog posts and create SEO pages, pages made for SEO. We did the he did it on his end on chat GPT. I did it the same at the same time on my end with cloud code. The difference at the end I said present to me everything in a in an HTML and it did and then I just I I'm logged in in Versal in the CLI. I just said like deploy to Versel just did and shared it with him. Well, if he wants to share it with me, he needs to share like his chat history or whatever like a text. Well, I shared an HTML doc with him with steps with ref references and links and like um checkboxes. It's just, you know, it's just much nicer to work with and it can be something you do with clients. Yeah. You know, back and forth. the like so I'm I'm writing a book at the moment and I actually really want that to be in markdown while I'm editing because I want to read the words and edit the words and this and that and the other right the end goal is it will be HTML for sure but like you know in there are some cases like Obsidian is a is a markdown um network of files basically and people are using that to journal and and create their second brain and things like that I that probably also belongs in markdown. Maybe even a HTML um maybe it creates both. Maybe the maybe there is a HTML file created at the same time as a markdown. But again, yeah, when when I need to hand I need to go in and hand edit stuff, that's when markdown will reign supreme. However, if I don't and I don't want I don't know about that get my agent to do everything. But why not? Why would you use markdown in even for that case and not just use like a simple like very simple HTML doc like this this you could do with HTML as well and much nicer I would argue this you could do with HTML table and it's nicer no because the book won't have things like that in it as an example there there there there is a element of simplicity like If the if the document is very simple, like it doesn't need tables and this and that, then it makes sense to have for me it makes sense to have it in markdown because I want to be able to write it. I'm not asking my agent to write things for me. I want to be able there and whatever. Whereas if I'm navigating through you know spans and all the rest of it and t and whatever then it just becomes whatever. But it's but we we are talking about many like many more use cases HTML was more appropriate but it's cool it's it's I think we're all kind of doing it but it just needed someone to really sit down and and demonstrate and articulate it with which did so yeah yeah by the way to make this I referenced that project I was like use that project as a reference but make an HTML for this MD file. But yeah, um that's it. Um with this, you're going to probably use more tokens, but have an easier time your dog. Yeah, exactly. Or communicating your ideas more effectively to your team, to your clients, to your grandparents. Yeah. Yeah. You can create a presentation with it. Yeah. Cool. Lovable. Let's wrap this bad boy up. Okay. Lovable. Um I need one second. All good. Wait, where where is the lovable? I think by the way, while you're doing that, for anyone that's listening or watching or whatever, um we are wanting to do something with some sort of community. we'd want to uh be a bit more interactive and a bit more uh let's say intimate with you guys and discussing AI and and and sort of bringing you up to speed and whatever it is that you tune into these shows for. We want to know what that is because we are just throwing things into the epha right now and talking about it. Whereas if we could be a bit more tailored to what you want, we'd build a better show and also again provide a community where we can all openly discuss things. So if you go to insiders.comai.com, you will be presented with a form. We're not collecting any uses data or anything like that. We literally just ask you what it is that you would want to talk about or what it is you want to know about when it comes to these shows andor what you want to know in this makebelieve community that may or may not happen. So insiders cmdai show.com. You've been working. Okay, I'm I'm almost ready. Give me one more second uh to get my notes. Okay. Yeah. Your uh your cable is having a bad time today. Yeah. Yeah. Okay. Introducing lovable aesthetic update. So, this is an update on lovable. Amazing video by the way. This video is so good. It it the video is amazing. It's about how lovable can now create like really good design like award-winning. They are not calling it award-winning but when you see these you think of award-winning websites, right? And it seems really cool and my social media has been full of it. So here Henrik saying made in one prompt. This looks absolutely amazing, right? At least in the first glance. And then here we have another one also one prompt made with one prompt. Here another one. Let me go back. Is this one prompt? Yes, this one is also I've been paid to to mention one prompt. One prompt. Yeah. And I have something to say about it. These are not amazing. Not as amazing as it might seem like. And I don't want to be too negative about this. The By the way, the video again is really nice. But these designs are I want to say templated. They do look all the same to me. I'm not crazy about am I the only one who thinks this? You you have really nice images by the way. Yeah, you have really nice images. You have these serif fonts that are kind of like paired and I'm I see myself explaining good design. B what I'm saying is basically good design. And yes, this is better than you know 95% of designers out there. My issue with it, my only issue with it, it it feels a bit repetitive in a way. Uh, but yeah, again, I don't want to be too negative about it. It's just that it looks a bit templated to me. So, here's the thing. All of them have this font pairing that I mentioned, this like aesthetic looking that uh a bit editorial looking. All of them very editorial. Yeah. Um, yeah. And they all have this minimal look. The images by the way are very nice. I don't know how they are getting that done. There is a warm feeling. At least the ones that I saw they all had some sort of lo um very warm feeling and uh there is some good sort of motion happening. So the motion is also pretty much uh the same in all in a way the same. It's a mix of scaling. Um, and basically like uh as you see, let me check like show you this one. You see how like elements are moving as you scroll. So this is happening to all of them. So there is just like a tiny bit of motion with this good font and image pairing and a stagger like you see the stagger especially in this one where the right elements I'm going to give that to them the right elements are staggered so these are shown with stagger so yeah it at at first I saw it and I was like wow that that's great but the more you look into it, the more this the more similar they look like, the more they are just like the same. And by the way, uh I don't know why Henrik is sharing this one, but this one like has an issue with the images. It it could be fixed, but yeah, small one. Um and maybe this one is the least impressive among the other ones. So yeah, uh there was a bunch of buzz around it. I wanted to to share my my two cents on it that it's it's it's quite nice. It's the new baseline that kind of unfortunate by the way in a way. Well, depending on how you look at it. Maybe for bad and medium designers this is unfortunate but for businesses this is the new base which is great actually. M well I think you need to like temper your expectations and I say you I don't mean you one needs to temper our expectations like this is this is without a doubt amazing that someone can just do this in one prompt right this is going to serve so many businesses just to be able to get something up and running. You aren't the as a designer, you're not the target market. Like, why would you want to produce something in one prompt that you know you would design or that you want to that you want to craft? Why would why would I if I want to paint something, why would I just want it to be painted instantaneously? Like there's that you're taking away my love, my enjoyment of painting. Yeah. If I could just do it in one prompt. And it's the same with anything really. It isn't whether it's lovable, whether it's Figma, whatever. So clearly designers are not the target market for this. This is not going to this is never going to wow designers. This is just going to be good for people who don't care about design. This is this is great for me. Like if I want something done in a certain way or whatever and let the AI figure out use best it's basically trying to use best practices as much as possible. Although I watched Ran's video and he demonstrated that it wasn't using best practices, particularly in contrast. It was failing miserably on the contrast. Um, yeah, but good is good enough. Do you know what I mean? For for for many many people. And unfortunately, the thing about the internet is you broadcast and it's seemingly supposed to be, you know, specific given specifically to you um because it's relevant to you, but realistically it's not. This is this is meant to land in the eyes and ears of uh company owners and basic people who just want to who don't care about who just want a website up and running. Do you know what I mean? So this is this is for people who previously would buy a template. Now you don't buy a template like these new baselines, these new standards with in these um AI native or AI first platforms. These are the new promp the the new templates. So instead of finding a template now you're prompting something and you might get lucky to get really good stuff but you might need to you know have more prompts to to dial it to your liking. Working with so many businesses in the past six years and you've been doing it for much longer. you know that there are just so many details that they need and a prompt cannot do it. Um it's an amazing starting point but a website is a living being actually that is going to grow with you and change with you and your business. So yeah good starting point not going to put designers uh out of job. Yeah. Yeah for sure. And you know, again, the way in from the dev perspective, we're just seeing a screen recording of them scrolling. There's no link. Well, maybe there's a link. I don't know, but there's no link here. Um, the website might not even be working, you know, maybe they want some live data fed into some of those stuff. Maybe they want that CMSable. Maybe they want um any they want a menu that opens and closes. We don't know if it actually works. So, you know, again, there's there's many more steps away from this, but this is this is great for for you guys wanting to just Yeah, I see people selling prompts, which is that's the funniest [ __ ] I said I said I've said I said on Twitter recently, I said if you're looking for prompts and skills and you're buying skills and and you know, using libraries and things like that, you fundamentally do not understand that what AI is giving to us. Like, yeah, that's most build your prompt by asking for the prompt. Do you know what I mean? Prompt your way to the prompt. Like, crazy. Yeah. It's like, "Welcome to the shortest AI course ever." Yeah. Ask your AI. Yeah. Prompting it. Whatever you want to. I know that there is a it's a different conversation. There is a definitely like two like you have taught me stuff about using clot code. You saw me today saying cancel right so there is definitely right and and you know what's that about it's not about the AI it's about understanding the fundamentals it's understanding in this case development you told me about I said like how can I change like the port like I was like run the like the you know um and then I had like mism I had two projects like trying to run on the same port and you told me just change it and I was like what do you mean change it where? Because like in my head I I I started from web flow and I would change it you know go to the settings but then I asked my AI like where do I change it was like go just to your package JSON it's like it's it's there it says like where to run it and like you can just click and change it or ask your AI to do it. But if you don't know that thing exists, you don't know like even how to how do you do it? Like if you don't we talked about so many things today. If you don't know what an MD file, you won't do an HTML file or you know what I'm saying? Yeah. And and don't get me wrong, like I'm not saying that everyone should understand AI. I'm just saying it's a it's a red flag to yourself that if you're looking for if you're looking for the answer for things then you you don't understand AI as much as you should. Right? There's no hard there's no there's no fault in that. I'm just saying that you're missing the point if you're if you're uh asking for skills, you know. Um yeah, Aaron. Aaron, we see you. But unfortunately guys, I do need to shoot. I need to catch a train. So that is all we've got time for um in this week's AI sign and dev. Uh go to commandai show which just takes you through to the YouTube. Give us a subscribe on that. Go to insiders.commandaishow.com to register your interest in a may or may not be a a community that we want to run. Also will help us guide and we'll just put you on the screen now to let you know that we are live every Friday. So catch us next Friday where we'll discuss the week in AI design and dev. I've been Sam. I've been Kabar. Keep on vibing, baby. Peace out. Damn it. I should have I should say