AI News Recap: 2.5 Pro out now, Figma MCP, Debunking Apple's claim, Sam Altman is go - Cmd+AI #7

Welcome to Command AI, the show where we press command F on the internet's AI news and somehow still can't find our sanity. I'm Samuel Gregory. This is Gabaza and we're here to decode the weekly chaos in AI web design and development. This week we have the illusion of the illusion of thinking, debunking Apple's claims. Translate the Oxford English dictionary in a mere two seconds. Deepel claims is Gemini winning 2.5 Pro Flash go on general release and light mode joins the chat. Figma MCP server is the latest in front-end developer workflows and Sam Alman's on a rampage against meta. Okay, maybe that one the last one's a bit of a stretch, but we'll get into all that on this week's episode. Does that work? Should I just wear these for the intro? Yeah, it it's awesome. Apple has been debunked. It's something we spoke about last week. Did you read this uh this claim? Uh I read I I mean we talked about Apple's claim but I don't know about the new claim about you know so yeah what you remember what Apple said right that there's models aren't actually thinking they're not actually reasoning where you've got Alex Alex Lawson here publishing the illusion of the illusion of thinking right he's saying that the there are a few things that Apple kind of messed up on which is namely they kind of ran these models to their absolute limit. So their context window was like at its peak. River crossing tests include mathematically impossible instances with insufficient uh so there's a boat capacity for six actors or something. It was one of the games that um Apple put the models through that were impossible but yet when they recognized it was impossible or unsolvable they were then penalized for it. and the evaluation uh scripts failed to distinguish between genuine reason and failures and practical output constraints judging by models solely on completed moves. I don't actually understand that one. I'm going to be completely honest, but they were already cut at the knees sort of thing when they were put through these tests. I don't know that's a phrase to be honest, but we're going to go with it. So yeah, is the illusion of the illusion of thinking and it's sort of I don't know. I feel like this is interesting. Maybe Apple comes up uh backing their claim that to see it's like a like a a [ __ ] online battle between No, you're wrong. No, you're No, you said this. No, I said this. I want to find saying that because the context window was, you know, it's its capacity. That's one of the main issues we have with LLMs, right? Yeah. It appears that we have such a bigger context. Well, we're getting bigger. But even still, even if you have um even if you have a big context window, it's still good practice to once you finish your task, once you've done with what you want to do, free up that context. It's like prompt engineering 101 sort of thing. I don't know whether Apple are going to come back with anything. uh particularly, you know, they've said their piece, they've said their piece, just end it there. Do you know what I mean? It could end up being just a bit of a feud that's unnecessary. We'll see. Yeah. Internet translated in 18 days by this is your side of the pond. This is a German German company called Deep L. Yeah. Yeah, technic technically this is a Nvidia story really but deep are a translation company. They've I mean it's just like Google translate but I think they offer enterprise solutions and and um you know it's more than just this chat interface that we have but they were provided a [ __ ] ton of um Nvidia GPUs. So um 128 basically GPUs going off and they it translated the entire um internet in 18 days that was previously set 194 days. So nearly a year by the previous supercomput well I mean not threequarters of the year two-thirds of the year uh down to just 18 days. Um, and this is a 90% reduction in translation time. What I thought was more interesting is that it can it translated the Oxford English dictionary um in 2 seconds down from 39 seconds which is just absolutely insane. But this is all thanks to kind of like Nvidia GPUs. I am an investor in in Nvidia. So I'm I'm happy to see this this news go out. Support for this episode comes from Flowbase. If you are building in Web Flow, Framer, or Figma, Flowbase can help you build much faster. With over 4,000 components to choose from, you have a huge variety of wireframes and super clean, nicely designed sections that you can put together by just simply copy pasting them to your project. Over 15,000 icons as well, over a,000 illustrations. They also have a super helpful Webflow app called Boosters. With boosters, you can add things like sliders and mares and countups to your project with a few clicks and without any coding. So yeah, big thanks to Flowbase for supporting the show. If you want to try them out, go head to flowbase.co. And just for you guys, I want to show discount command flowbase all capitals for 20% off any plan. This is a limited offer, so get in there early. All right, that's it. Let's continue with the show. 2.5 going on general availability which basically means it is production ready baby. Um and I know Deepseek have overtaken Opus. Wa mental. It was like who I mean 2.5 is clearing the bar pretty heftily and Deep Sea are scraping it. So this is this is definitely one of is something I want to be uh looking into. But you know given that we are talking about Gemini here, the preview 0605 model is just the latest. It also means that flash is also on general availability which is awesome. Get it in your apps and all that kind of stuff. Um, and they also released a new 2.5 flash light on public preview, which is like an even faster basically just cost-sensitive highcale application sort of thing. So, really a really nice suite of of products here. And I know what you're going to say. You just want one. You just want one. Well, let me tell you this. Yesterday I tried Gemini app on my phone like the Google Gemini app on my iPhone and it was super cool like the vision. It was awesome. So I was like showing some stuff Yeah. in my kitchen that I were not sure uh like what they are or like what they are called in German. I was actually using it. I wasn't just like trying to I I was like, "Okay, what what is this and what can I do with it?" Because I just had a lot of like stuff that my mom leaves in my house and I didn't know what to do with them. I was like, "What are these?" It was always trying to be very helpful to the point that I felt like it's annoying. I was like, "No, just let me ask you and then stop with this like trying to push the conversation." It was pretty cool. But at some point uh that I was like cleaning the kitchen the conversation changed. So uh from talking about the kitchen we were talking about uh Alpha Century this the you know the nearest uh star system and we're uh and then I was talking about like the exact Yeah, sure. Uh how many how many stars are there? Let me ask my camera. It's good. We um we have a we have a question here. Basically, is it available in cursor yet? 2.5. This is 2.5 Pro. Yes, it is available in cursor. They basically I think yesterday they made the 0506 or you know 0605 whatever it is. The last release is the same release that's available on uh you know is they're just pointing all to the basically the the the same release. So it's good to go in cursor. So Gemini 2.5 Pro is $125 in 10 out. Uh flash um is 30 cents to in and then 250 out. And then flash at preview where's where's flash? 10 in and then four out, 40 cents outwards. So, um, pretty pretty reasonable. Still pretty reasonable. I think for the performance to price ratio, it's crazy. I'm very excited about this next one. And I'm surprised you didn't bring it up to be honest. I'm surprised I was the one that had to go and find it. Uh, the which one? The Figma dev MCP. Figma Figma MCP server. Like, this is incredible. I mean, do you know what MCP servers are? Do you know do do you know what the Yeah. Do you know what they are? Yeah. Yeah. It's basically allowing LLMs to use um another service kind of like it's like APIs for LLMs. Yeah. Yeah. And it determines for LLM MC we are throwing a lot of let's let's dumb it down a little bit. So you can use natural language like a prompt and the the edit the your cursor winds surf whatever will figure out that you want to do something in Figma. So um where is it here? Get code from Figma and implement this file. But basically uh I go nuts in my video. I go nuts. I say can you get the design and put it in this file and it will just again just the LLM figures out that you're talking about a Figma and it runs a tool call to get the the selected frame that you have and then pulls it in and makes it as code. And obviously you can go as narrow as possible. You can you can get a button that has different variants or you can go to a whole page and just bring in a whole page and it will build the code and you know it's 90% there probably 95% there like it's never going to be perfect well it might be perfect soon but like it's not perfect all the time but it's pretty gosh darn great and I can just say go get me the variables and it will pull in all and this is where it comes in very powerful is that you pull in all your it pulls in all the variables and the beauty of natural langage language and LLMs and stuff like that is that you pull in the LL pull in the variables and I tell it to make them into Tailwind for variables. I'm adding the extra layer of complexity and saying well give me tailwind make them Tailwind or make them SAS make them less you know all this sort of stuff. So this is crazy and more people need to know about this. Is your team using this? No, but I think this is because a lot of people are scared of AI and especially like designers like scared in the sense that but this is not to do with nothing to do with the technical designers don't use cursor right but this is not this is not for designers it's for developers yeah that's what I wanted to ask like is there a way to also reverse it use for designers yeah reverse it to to to do something in yeah not yet any they would need to implement that they would actually need to implement the ability to do so. The good thing is is I don't need to leave cursor. I can just make all these tool calls and you can you can have we're getting into MCP land right now but like I can make searches. I can search the internet. I can do everything. Wind actually this is news. Windsurf just released a browser. So you can literally browse the internet inside of Windsurf. And as a de again, this is just purely developer stuff. As a developer, it's great that I don't need to leave my IDE. It's great that I don't need to and I can just get my information all from in here because chat bots are just are just winning. I saw that news uh with Windsorf, but I don't get it. Like why do you need a browser there? It's efficiency. It's just it's just marginal gains of efficiency. And actually, this is something I want to figure out. I I want to pose the question and like I wish I could film my desk set up right now, but I find it so annoying. Uh mouse and keyboard, right? I just want to stay in the one place. Maybe that's where um uh who what's Lenovo made that little red nub. Remember that little red nub in the middle? Like I I think that might need to make a comeback because like I just I hate moving over to my mouse and back and forth and back and forth. It drives me nuts. Scientists get on this because I'm tired of lifting up my arm. Now Samman's been on a bit few podcasts and he's he's been he's been laying off a few little little bits and bobs here or there. You know how vague I sound. Basically, he's he's anticipating GBT 5 this summer on this I think it was this podcast, which is he's saying that this will be the be all and end all to all of the stupid versions that they have 03, 04, 4.1, 4.5, 01, all this kind of stuff. He's saying it's the it's going to be the end of all that and GPT5 will be the one that will just fix it will work out basically whether it needs to go heavy or go hard. So he's on this podcast uh uncapped with Jack Alman which is his brother and then he went on this one which is OpenAI's own podcast. So they've launched their own podcast which could be quite interesting straight from the horse's mouth as it were. Um he also alluded to better self-drive technology. So all the cameras and stuff like that, the liar that Teslas have, he's claiming that they've or alluding to not claiming alluding to that they've figured out a way to do it with just normal cameras. like no liar, no anything like that, which would be very interesting. They he didn't suggest that they were going to release this, but as a company, you've got to and this is really exciting to think about really, but as a company internally, you've got to innovate or you've got to conceptualize things uh from all walks of life. It seems like they're playing around with this technology and and it helps in other areas whether they release it or not. But this idea that they've got potentially better self-drive technology or they figured it out sort of thing, it's really really interesting. I'm on the LAR team. So you're on the liar team. Well, I mean I think it's not to say that we should replace liar with cameras is to say that this is uh an addition to or like this could just lead to something else. basically it could lead to well he spoke specifically about robots as well um very he thinks we'll have very capable robots walking around in the next 5 to 10 years now whether you could put liars in robots might be kind of what I'm alluding to right now but maybe they just need cameras maybe tiny little robots just can fit in cameras so but they figured out how to do it you know what I mean so absolutely like you're not going to be able to beat liar you know especially in like military and things like that but um Yeah. So, yeah, robots in the next 5 to 10 years. And we also got clarity on Zuck's, was it nine or 10 figure offer? Yeah, nine figure I believe it was. No, nine figures, which so Alman said that it was specifically to members of um OpenAI who refused it and it was a it was a hundred million for as a signing bonus. So, it wasn't necessarily a salary, and interpret that how you will, like whether I don't know. I've never had a b I've never had a signing on bonus, but is a signing on bonus less than a salary? I don't know. I can't imagine it to, but but yeah. And Alman was like, well, you know, he he's he's trying hard. He's going all in on it, but that's not the best way to build a culture is to just throw money at this stuff. So um there was a bit of slack talk from from Alman uh towards Zuckerberg there because I do I do think people really enjoy working at open AR. I think they intentionally have a very small team. I think they're only like I don't know maybe even less than 200 people or 100 people or something like that. They he tries to keep it really small. Uh but I've also heard he's got a tyrannical leadership style. But there you go. Um because did you see the Sam Alman files or the Open AI files that went out this morning? I didn't I started reading them and I was like what is this? Like it's just it feels like it's just like clutching at straws on talking [ __ ] about open AI. But anyway, there's that. It's very much like the show again, the Silicon Valley show. Ah yeah. Yeah. Payload is joining Figma. Uh payload is um a back end. I I think they do like back Oh, there CMS isn't it or something like that. Did I read this? And to build the modern web. So this is very interesting like mixing this with the you know the MCP that they have because I think the MCP is also really useful to update like huge databases you know using natural language. you can just do a lot of, you know, updates. Um, take your database and do whatever with it. So, Figma just uh bought them. Buying up. See, they've got to they've got to get some sort of head start on their Figma sites because they said CMS is coming, right? Yeah. And this is interesting that they announced Figma sites and they showed the CMS a little bit and they said it's coming but they bought this company afterwards or maybe it's just being announced now. Yeah. And their site is pretty cool. I see what you mean about the Silicon Valley thing because it's about announcing things before. I mean it's not rocket science adding a CMS. So, I think it's pretty fair to announce they're not innovating anything like Apple are doing, like innovating, but then making those promises before they even know how they're going to do them. Yeah, it's really cool. How do you think they did? Is that a video on the background there of that website? The video. Uh, I haven't checked it actually. If it's a video, it looks like it's not responsive. Uh, but it might be a shader. I don't know. I'll I'll check that out later. As you see, I'm showing this website using Dio. I'm slowly making the change. And this is the last thing I want to talk about. Quite literally, DIA saved me 66 uh euros. Here you see DIA. And this is how I used it. So I was like, if I'm going to email and request for a refund, I have to do it and then come back to it and it will probably take some emails. So I opened the tab India and I just asked I said write a refund. Um because it had the context of the email that I was charged, right? So it just wrote it and then you just press insert. They replied to me. I didn't bother reading most of it. So I just looked through it and I just prompt like a very short you know sentence to yeah continue with the cancellation continue with uh refund sent me a link about their policy like refund policy. So obviously I open this and I come back here and I I can just take this refund policy tab and now I have this as the context and I think yeah yeah right write write a followup I want my refund I didn't use the tool please and I mean they were nice they gave me the refund but all of this happened with like minimum effort in typing and also minimum effort in thinking because all of I didn't have to like re really read or like look at the details and most importantly what is like the big point for me is I don't have to go to to chat GPT app or like their tab and change the like this is not the only thing I do throughout the day. Yeah. Surprise, right? I I I do different things. So I like I change the tabs in OpenAI, but with this I don't have to change tabs in 20 different places. And this to me is like an aha moment, a wow moment. I'm like truly like I know we talked about in the past weeks. I'm falling in love with this. Like the idea is actually working. I want auto browser. So I want to be able to action things on multiple or or even singular tabs. But um but it's cool. I'm glad you I'm glad you're converted. Listen, I've got a hard stop on that bombshell. I've got to go. I've got a weekend ahead of me. Cheers for tuning in everyone. Um and yeah, keep on vibing. Peace