AI News: Huge Updates From Anthropic, OpenAI and Google

It's been another huge week in the world of AI, as kind of with every week. We got big updates out of OpenAI, out of Google, out of Anthropic, out of Perplexity, and I'm here to break down what happened and what I think you should know about what happened in the world of AI this week. I'm not going to waste your time. Let's get right into it. Now, the news you're probably going to hear the most about this week from the most other YouTube channels is Claude Opus 4.7, the brand new model out of Anthropic. I do want to talk about this model, but it's not actually where I want to start today's video. We'll circle back to this one in a few minutes. Instead, I want to start by talking about the new Codex app from OpenAI. There seems to be like a theme of this week, and that theme is that pretty much every AI company rolled out a brand new user interface update. OpenAI, Anthropic, and Google all put out brand new versions of their app, and we got something pretty cool from Perplexity as well. Now, the updates to Codex are probably the most interesting to me because it feels like a step in that direction to OpenAI's super app. There's been a bit of talk recently that OpenAI is going to sort of consolidate everything into one single super app like ChatGPT, their image generator, their code generator. All of it was just going to be one single platform. You know, before Sora died, we thought Sora was going to be part of that as well. And it really feels like that super app might end up being Codex after all. Codex can now operate your computer alongside you. It can now generate images, it can remember your preferences, it could learn from previous actions, and it can take on ongoing and repeating tasks. What is this starting to sound like? Well, more and more of like an agent, kind of like OpenClaw or kind of like Claude Cowork. They also say they're extending it beyond coding, which was also kind of evident in the fact that they just mentioned that you can now generate images straight inside of the Codex app. And they also rolled out a feature that we kind of saw in Codex and Perplexity, but this one seems even more powerful. So it's got background computer use, so it can see, click, and type with its own cursor. And it says multiple agents can work on your Mac in parallel without interfering with your own work in other apps. So you can be working in one app while this is doing something in a different app. If you remember when we played around with Claude Cowork in the past, Claude Cowork would essentially take over your computer, and while it was doing its thing, you couldn't do anything else. This can actually go and work with apps while you actually do other stuff. The app also includes an in-app browser now. And if you're vibe coding, this can actually do something pretty cool, which allows you to comment on like the webpage you're looking at and give that context to Codex. So they just highlighted the screen and commented to go and make that smaller and then added nice brick theme logo. So just by commenting on the site itself, itself, it's going and making those changes. It's now using GPT Image 1.5, so you can actually generate images directly inside of Codex. There's a bunch of new plugins available inside of it, and in order to use it, all you have to do is have the Codex desktop app installed and be signed into ChatGPT. All right, so I do have Codex installed, so let's go ahead and open it up. It tells me there's an update available, so let's go ahead and make sure we're running the latest version here. And we're now on the newest version. We'll create a new project here, so we're starting from scratch. Generate a mockup image of website for a site that sells surfboards and tacos. Let's send it in and see if it actually generates an image directly inside of Codex for us. All right, so it's using the ImageGen skill here, and we've got a nice design website for surfboards and tacos because, you know, those go together, right? So let's go ahead and say build this site for me based on the mockup image. And now the fact that it actually generated this image directly inside of Codex is new. You would have normally had to go to ChatGPT, generate it there, bring the image in. It's automatically using specific skills that it knows will help to design this. So it's using the front-end design skill, and now it's starting to build it out. All right, so it took a little bit over 6 minutes, but it says built it as a polished static storefront. So let's go ahead and open it here. We can see it opens it straight inside of Codex. So inside of Codex, I mean, it doesn't look exactly like the mockup because it created SVGs instead of actually pulling in images. But if I come up here and click on comment mode. Now you can see I can hover around and it's going to select different elements on the screen. So let's go ahead and select this area right here and say generate an image that has surfboards and taco truck like the original mockup and use it as the background for this section. Make sure it doesn't interfere with the text. Let's go ahead and it looks like if I hit Command+Enter, it will actually submit that as a prompt and we can see it's actually starting to run. So let me shrink this up over here and it took just shy of 2 minutes, but we see it generated this image here. And if we expand our website here, it swapped out the background with that image. We can see it put the truck over to the right so that it didn't interfere with our text. Exactly what we were looking to do. Let's create a brand new project again here, but this time I want it to create an app that runs on my desktop. Let's try create a desktop app version of Connect Four that allows me to play against the computer. It needs to work on this Mac and run locally. All right, so it did work for about 13 minutes, but we now have a local Mac OS desktop Connect Four game. Let's tell it to run the app for me. All right, I'll launch the local app. I'll go ahead and say, yes, you're allowed to do that. And look at this, here's the Connect Four app that it made. So I can drag along the top. Let's see, let's drop it here. And then the computer put it there. All right, let's drop one right here. Okay, and it knew to block me. Put one there. I'll drop one there. I mean, it's actually doing a pretty decent job. That was kind of a dumb move. Nope, it actually beat me. I'm bad at that. There's even a difficulty slider. All right, now I wanna try something else real quick. Open the app and play as me to test the experience. Okay, so let's see if it will actually play. Okay, so I'm actually not doing this. The AI is playing as red and the, well, I guess the built-in game AI is playing as yellow. So it's literally testing the user experience for me and the computer just won. Me as played by ChatGPT just lost. So it can actually take control of the app that it just built, use the app, test the experience. Itself and then report back to me. That's pretty sweet. When Codex first launched, I was fairly unimpressed by it, but the more I've used it, the more I've played around with it, the more it's really kind of growing on me. I mean, I like the 5.4 model already for coding, but the simplicity and the usefulness of the actual platform, the IDE itself, is getting better and better in my opinion. But like I said earlier, OpenAI wasn't the only company to make major updates to their apps this week. Anthropic also rolled out updates to Claude Code, the biggest new feature of this is to be able to run sessions in parallel. So you can actually kick off work across multiple repos and move between them as results arrive. So let's open up the Claude app real quick. It's automatically downloading the update, so we'll relaunch to apply it. And the first thing we'll notice is the interface is already a little bit redesigned here. So you've got Claude Code here, Cowork here, and just Claude Chat here. So slightly different. In the previous version, those were up in the middle. Now they're a little bit more hidden. But we could go ahead and open a recent project here. I'll open my AMA overlay, which If you saw my recent video, you know what that's about. I'll just give it a simple prompt, review this code and tell me what can be improved. And it's starting to review the code, but let's go ahead and kick off a new session. This time let's jump into Future Tools. Once again, review this and tell me what can be improved. And now we can see we've got two different sessions going in parallel, one working on the Future Tools website, one working on my AMA overlay app. We can even drag these around and pin common threads that we're gonna jump into. So let's pin both of these up here. And now I can sort of jump back and forth between these two threads that are working. It's a nice little quality of life feature if you're using Claude Code a lot directly inside of the app to just be able to jump back and forth between multiple projects or even be working on the same project but having it work on different features within the same project. Some of the other new features that came with the new version of Claude here, you have an integrated terminal, an in-app file editor so you can edit the files directly manually inside the app, a faster diff viewer, and an expanded preview where you can open HTML files or PDFs in-app, in addition to running local app servers in the preview pane. The idea here is you don't really have to use CLI or command line interface anymore. You should be able to do pretty much everything you'd want to do with coding directly inside the Claude Code app, except for, you know, jump between models that aren't Claude models. If you're building apps that use AI, whether that's a chatbot, an AI agent, or really any tool powered by a large language model, there's this annoying gap between getting something working and getting it working at scale. Your prototype runs great, but the second real users start hitting it, everything slows down or falls apart. That's the problem Crusoe Managed Inference solves. Basically, instead of setting up and managing your own servers and GPU clusters, which is a whole job in itself, you bring your AI model to Crusoe and their team helps you deploy it on infrastructure that's purpose-built for speed. We're talking 5 times more throughput than standard cloud providers, which just means it can handle way more users and way more requests at the same time without choking. And if you're working with really long prompts or pulling in a ton of context from documents using RAG, Crusoe has this technology called Memory Alloy that keeps things fast even as the context grows, where most platforms would start slowing down. The part that I think matters most, especially if you're a smaller team, is you don't have to become an infrastructure expert. Crusoe's team actually works alongside you to optimize your setup so you stay focused on building your product and they handle the hard infrastructure stuff. Check out the link in the description to learn more about Crusoe Managed Inference, and thank you so much to Crusoe for sponsoring this portion of today's video. Google also made their app available for more users this week as well. The Google app for desktop is now available for Windows users around the world. Now this is basically Google Search and Google AI mode directly inside of an app, so it's not like they took AI Studio and put that into an I wouldn't be surprised if that wasn't far off. Or I guess that's kind of, I guess, what Antigravity's supposed to be, but Antigravity is a fork of Visual Studio Code. But whatever, I'm getting in the weeds there. I have a feeling over time Antigravity is going to be what Codex is to OpenAI and what the Claude app is to Anthropic. But right now Antigravity is still kind of very Visual Studio Code-ish. And if you're like me and your daily driver is a Mac mostly, well, you have the Gemini app on Mac now. So pretty much anything that you you can do inside of Gemini for the browser, you could do inside of the Gemini app on your computer. Obviously, you can chat with it just like you would the normal Gemini in your browser, but you can also generate images with Nano Banana, generate videos with Veo, and again, do most of the stuff you'd expect to do with Gemini. If you head on over to gemini.google/mac, you can download the Mac app here. Let's go ahead and fire it up on my computer. And if I open up the sidebar, I've got all of the that I would've had just if I was using the browser. If I click the plus down here in the bottom left, you can see I've got Create Image, Create Video, Create Music, Canvas, Deep Research, Guided Learning. Again, anything you can do in the browser version of Gemini, you now have available to you in a separate desktop app. And just like you'd expect, down on the bottom right, you can switch between Fast Thinking and Pro. Google also rolled out some other updates this week that I think are worth talking about, including an update to Chrome that I'm really, really a fan of. In fact, my favorite feature of Perplexity Comet is also now a feature in Chrome. You can now turn your best AI prompts into one-click tools in Chrome. So basically they've added slash commands to Chrome. When you write a prompt that you'll use again, you can save it as a skill directly from your chat history. The next time you need it, select your saved skill in Gemini in Chrome by typing forward slash or clicking the plus button. Your skill will run on the page you're viewing along with any other tabs you select. Comet has this assistant up here, and I have a bunch of slash commands down here like news review and skim news and YouTube video and comment review that help me do things when I'm on a specific webpage. Comment review looks through the comments on my YouTube videos and helps me find sort of general themes and things people are saying in my YouTube comments. News review reads news articles for me and pulls out important pieces so I make sure I don't miss them. Again, this is Comet, but you could do that in Chrome now. So let's go ahead and fire up Chrome here. So I'll head Future Tools. Let's go to a news article here. We'll go ahead and read this news article and then ask about this page. Now let's type, break this news into bullet points for me. All right, so it gives me all the news here into bullet points, and I'm not seeing the ability here to turn this prompt into a slash command. But to be fair, it does say starting today we're rolling out skills to Gemini in Chrome on desktop. So maybe it just hasn't been fully rolled out yet, but when it does roll out, it should look something like this where there's a button that says save this prompt as a skill. Google also rolled out a new text-to-speech model this week called Gemini 3.1 Flash TTS. This new model's available in Vertex AI, it's available in Google Vids, and we can go and test it in Google AI Studio in a moment here. We can see in their own demo here that one of the big standout features is that you can tell it how to speak by using things like excited, amazed, whispers, panic, sighs, laugh, and it will follow those extra instructions on how the voice should come out.

Welcome to the next generation of Gemini text-to-speech. Watch how these tags transform the tone. We can drop into a dark, suspenseful whisper or speed things up to absolute panic mode because we're running out of time. The model breathes.

It even laughs.

The best part is being able to control the timing with a dramatic pause.

So that's pretty impressive. Again, if we head on over to aistudio.google.com here, come up to Build in the top left here. Over here where you select your model, we'll select this. Let's go to Audio, and we can now see Gemini 3.1 Flash text-to-speech preview here. They have some quick start templates that you can use here, and we can see it even has like a two-speaker back and forth. So if you want to generate something that sounds like a NotebookLM podcast generation, you can actually do that now and control the conversation. Let's see how this sounds.

Welcome back to Future Tools hosted by Matt Wolfe. Today we're diving into the intersection of AI and actual usefulness.

Exactly. I've got so many thoughts on what happened this week. It really is shifting daily.

I mean, did you see Matt's latest video from Wednesday?

Oh, absolutely. It blew my mind, but it also raised some pretty huge questions about the implications of tacos on the Star Star Wars universe, right? So let's get right into it because we have a lot to unpack today. Bibbidi-bobbidi-boo-yah!

I mean, it was pretty good. It messed up. It kind of didn't get the voices perfect and it skipped over one of the laughters. But again, if you want to make that like NotebookLM-style podcast, it is ultra controllable now. Google also rolled out an update to Nano Banana. So recently with Gemini, you've been able to have personal intelligence where it actually knows stuff from your calendar and your Google Drive and your email, your Google Photos. It actually learns about you, well, now it can actually use that information inside of Nanobanana. So you can have it generate images about you or about people that you have photos of and your dog and your family and things like that. Because it has this additional context, it's going to be a lot better at generating images personalized to your needs. All right, let's give it a try here. My Gemini account is actually not like my main account that I do most of my emails and calendar scheduling and photos and stuff like that through. So it doesn't have a lot of about me, but let's go ahead and see what happens. I'll do create image so it knows to use Nano Banana and then say create a scene that depicts what you think my life is like. It knows I use a computer. I guess that's about it. I don't even have a cat, but pretty much everything else about this image is wrong. But I can't really blame Google on that because it is not linked to my main account. But give it a shot if your Google account is connected to all of your services like your Gmail and your calendar and your Drive and your photos and all that kind of stuff, it'll probably do a lot better job. Perplexity rolled out a new feature this week called Personal Computer. Now Perplexity's had their computer platform where it goes and does like agentic stuff for you. You tell it to go and build an app for you and it goes off and builds it and comes back and says, hey, we've done that for you. And it's kind of like a very user-friendly version of like what OpenClai is. This new one, Personal Computer, brings that to your machine. It can work across your local files, native applications, connectors, and the web to complete complex and even continuous workflows. Flows. The original Perplexity Computer is run on Perplexity's cloud systems. This one runs on your own device. It brings Perplexity Computer into a more useful place— your own machine— and integrates your local files, tools, and apps into an orchestration system of Perplexity Computer. One of the best ways to experience it is on a Mac Mini. It stays available 24/7 for work that needs a persistent machine or secure local access to your files and native apps. You can ask it to do your to-do list. Computer will read your notes, to-do list, reason how to accomplish each task, and work across all of your local files, iMessage, email, connected apps, and the open web to get it done. Actions are auditable and reversible. A system that acts on your behalf needs to be useful and legible. So it's still using the Perplexity model, it's still using their API and their servers to run the actual AI models. It's not like these are local models that the large language model is processing on your machine. The inference is still happening on their servers, but it can work with all of your local files. So similar to like what OpenClaw Cloud Cowork can do where it can access the local files on your computer. This can do the same thing, but it's using Perplexity's models. Now this is something I'm actually going to do a test with and make another video about. We'll do a deep dive on this specific project because I do want to put it to the test. And well, I've got a Mac Mini right here to test it with. In fact, Perplexity sent me this Mac Mini to try and test this. They didn't sponsor this video. They didn't sponsor the video I'm going to make about this, but they did send me a Mac Mini and say, hey, here's a computer to go and try it with. I mean, what world am I living in where companies are just sending me Mac Minis and saying, hey, go try our product with this. But yeah, I'm gonna do that. I'll make another video and give you my honest true thoughts on what I think about Perplexity computer running on a Mac Mini because that's what they sent this for. And my might as well try it. So look out for that video. So we also got an exciting update out of Canva for some features that are coming soon that I'm kind of excited about. Now, full disclosure, I do have a small investment in Canva, but I never let any of my investments affect what I say about companies or tools. I always talk about the good, the bad, the ugly. I try to be as unbiased as possible when I talk about stuff. Saying that, these new features do say coming soon. They're not out yet, so I haven't even tried them myself yet. But essentially, they're going to make it so you can pretty much prompt anything you want into existence inside of Canva. These are the Canva AI 2.0 features here. So we can see in their example, they just prompt, can you design a social media campaign to advertise our summer running shoe sale using the brief in Notion? And it went and created a Canva image, and they're just editing this image through text here. It'll connect to things like Slack, Notion, Gmail, and other tools. You can actually schedule tasks inside of Canva to run automatically theoretically write a guidance document, generate summary presentation, create a summary report. It learns from your designs. Keep creating, and Canva AI 2.0 remembers and adapts to your unique style. So let's say you're a YouTuber, for instance, and you make YouTube thumbnails. Over time, it theoretically will learn your YouTube thumbnail style and get better and better at generating that. It'll also run web searches and help bring stuff right into that design, similar to like what Nanobanana does right now. And it says generate any element, so anything you can imagine— graphics, audio, 3D, right inside the editor. They're also releasing an offline version of Canva and some AI-powered activity generation and thousands of ready-to-teach resources to Canva, mapped to curriculum so you can search by country. That's interesting. That seems like a little outside of the scope of what I thought Canva was for, but I don't know, seems cool. We also got some new large language models this week, as I kind of hinted at in the beginning, the most talked about being Claude Opus 4. 4.7. Taking a quick look at the benchmarks here, it's interesting because they did share Mythos Preview, the model that they basically said is too powerful for you to access, but they wanted to show you how this model they are giving us access to compares to both that one and the one we previously had access to in Opus 4.6. So where the biggest leaps seem to be is in agentic coding. Opus 4.6, 53.4%. Mythos Preview, 77.8%, a pretty huge leap. And this version they gave us this week, Opus 4.7, it kind of split right down the middle at 64.3%. That was SuiBench Pro, SuiBench Verified. This is Software Engineering Bench, by the way, kind of the same thing. Mythos scored 93.9%, but we don't have access to that. Opus 4.6 scored 80.8%. That was the previous generation model. Opus 4.7, they kind of split it down the middle and said, we'll give you like the halfway version. So the people that are really going to notice the difference on this model are going to be coders. It is quite a big leap still above Opus 4.6 and what we were getting out of GPT-5.4. But if I'm being totally honest, the majority of people probably aren't going to notice like a big leap in using Claude or any of your other tools, again, unless you're a coder. Now they do say it is substantially better at following instructions. It's got improved multimodal support, so better at understanding images and things like that, and it's better with the memory here. If you're just like an everyday Claude user, you might notice the difference in instruction The way I sort of heard it described was the older models, you had to sort of prompt engineer 'em a little bit better. You had to be very precise with your prompts. It may have taken you a few prompts to get to the desired destination with your chat. Well, as these new models come out, they sort of get better and better at understanding what you're asking for, and you need a little bit less like prompt engineering. You need a little bit less detail in your prompts to get what you're looking for, and maybe you'll get exactly what you're looking for the first time you give it a prompt as opposed to like the third time you give it a prompt, you know, like the third time through the conversation. There are much deeper dive videos on this new Opus 4.7 model, and a lot of them are going to be geared towards coders. So if you are a coder, you now have a new best top-of-the-line model that you can be using. When I write code using Cursor or Claude Code or one of those tools, I'm going to be using Opus 4.7 now because it is clearly the best model at coding according to the benchmarks. The model Minimax M2.7 is now officially open source. They claim it's got state-of-the-art performance on SuiBench Pro at 56.22%, which was, you know, pretty much state-of-the-art until Opus came out. So 56.22%, using Claude's own charts here, that was better than Opus 4.6 and better than Gemini. Not quite as good as GPT-5.4, but now obviously Opus 4.7— this is SuiBench Pro here— sort of blows it out of the water now. But Again, the important thing being this is open source. So this is a model that can be fine-tuned. And according to this community note here, Minimax M2.7 weights are publicly available, but under the license are actually prohibiting commercial use. So you can't build something on top of this and then sell what you built that uses this model. According to this community note, we also got a new model out of Alibaba, as we seem to get almost every week now as well. Quin 3.6, 35 A3B, and it's open source as well. Here's the benchmarks on that one. This dark purple one is the SuiBench Pro, the one we were just looking at for those other models. And well, not quite as good as Minimax and not nearly as good as the new Opus 4.7 model. But again, this is an open source model, so you can fine-tune it, you can run it on your own cloud servers. And at 35 billion total parameters here, probably could run this one locally too if you have a pretty decent GPU. So pretty cool to see models getting decent benchmark scores that can be run locally and that are open source. All right, I've covered a lot. This was a really, really busy week. I just have a handful more things that I think you might be interested in, but we won't go too deep into any of these. So let's jump into a rapid fire. Starting with some additional news out of OpenAI this week, they introduced GPT Rosalind for life science research. This is a new reasoning model designed for research across biology, drug discovery, and translational medicine. It's optimized for scientific workflows, combining improved tool use with deeper understanding across chemistry, protein engineering, and genomics. We can see in their chart here, this model is much better at things like chemistry, phylogenetics, experimental design and analysis, and tool usage than any of their other models that they have available. And they even compared it to some other non-OpenAI models like Gemini 3 3.1 and Grok, but it's not compared on here to any of the Claude models. But they're taking a play from the Anthropic playbook, and this is for trusted access only, so they're only giving this to scientists and researchers who apply and they grant access to it. So this isn't something that just any of us can get our hands on. However, it does show that we're getting closer and closer to these AI models helping with more and more scientific breakthroughs and drug discoveries and figuring out ways to cure diseases and things like that. And I think that's when AI is going to become truly valuable to the world. Like that's what most people want these AI models to do. It's cool that we could get like little productivity increases out of 'em in our business and stuff, but the real game-changing use cases are going to come in biology and chemistry and like, you know, the material sciences and things. So love to see models like this being designed for those use cases because that is like life-changing use cases. Microsoft Microsoft released a new version of their image model called MAI Image 2 Efficient. We played around with their new image model a couple weeks ago, and it was a pretty solid, pretty good image model. This new model appears to be just as good, but it's a little bit faster and a little bit cheaper. MAI Image 2 Efficient leads on render speed, taking, it looks like on average, about 13.7 seconds to generate an image, where the Nano Banana Flash, the one that I've always thought was really fast, takes about 19 seconds. Now, it does specifically say it handles short-form text like headlines labels cleanly, but you want to use their existing MAI image 2 if you're looking for longer or more complex in-image text. So the original model's a little bit more detailed while this one's a little bit faster and a little bit cheaper, but the trade-off might be some of the detail and the longer text if that's what you need. Midjourney also released a new version of their image model, version 8.1. They claim their new iconic aesthetics are back with native 2K HD rendering. It's 3 times faster, 3 times cheaper versus V8, which is kind of crazy to me because V8 was already really, really fast. I didn't really think it was the most impressive image model. I actually thought Microsoft's image model was way more impressive, but this one was really, really fast. So 3 times faster is like almost instant, I'm assuming. Here's some of the examples they shared, and this one specifically right here actually does kind of remind me of the old Midjourney aesthetics that I really, really liked. These are likely cherry-picked images, but you know, if you're still using Midjourney, they are still rolling out new updates. We don't really hear about them as much anymore, but they're still out there cranking along and putting out new updates. If you do a lot with video like me and you use DaVinci Resolve to edit your videos like me, well, they just rolled out some cool new AI features as well in their new DaVinci Resolve 21. Probably the coolest AI feature for me that I will use the most is this one here called AI IntelliSearch. Instantly search for people and content with AI IntelliSearch. It'll analyze the media and use the IntelliSearch tool to find specific objects or keywords in dialogue Customers can even search for individual faces. Results appear as whole clips in the media pool for faster clip organization and editing. So you can dump in a whole bunch of like B-roll from an event, a whole bunch of interview clips, and then you can search those interview clips for when people say specific things and it'll help you find just that clip. Or you can say, help me find the B-roll of this keynote session where Jensen Huang was speaking, and it might just find from the video a keynote where Jensen was speaking. So just a much easier way to organize your video footage and search for it in DaVinci so that you can then and pull it on your timeline. I know most people watching this probably don't care about that, but if you do any video editing, feels like a pretty game-changing feature to me. There's also this interesting change an actor's age with AI face age transformer tool. Analyze a face, enter the subject's age, and adjust the age offset slider to add or remove age-related features such as wrinkles and facial fullness. These aren't features I've tested yet because I actually haven't downloaded the new version of DaVinci yet. But if you ever see me in future videos where I either look 30 years older or like 15 years younger, it probably wasn't my new skin care routine. In probably the weirdest AI news of the week, the company Allbirds, which sold shoes, is now an AI company. And when they announced they're switching to an AI company, their stock jumped by 600%. The people who are talking about an AI bubble have to be having a field day with this one. Like, when a company that makes shoes says we're pivoting to selling AI GPUs, like, we're— yeah, that, that's just weird. And for context, Allbirds had a $4 billion IPO in 2021. They never turned a profit. Sales dropped nearly 50%, and then the company sold itself off for $39 million. So $4 billion IPO down to $39 million. They announced that they're rebranding to New Bird AI. They're no longer selling shoes, and they're acquiring high-performance GPU assets. I just, I, I don't know, like, not investment advice, but this is probably one that I would short. I'm not actually going to, so don't hold me to it. It just doesn't sound like they're the most brilliant business people, if I had to guess, but I don't know him personally and I don't know the full story. This is just kind of a meme right now. And finally, there was this cool video out of Boston Dynamics this week where they got a robot to do their to-do list. And here's the thing that I thought was cool about this. So check this video out. The robot actually walks up to a whiteboard and you can see there's tasks on the whiteboard. Make sure all of the shoes at the front door are on the shoe rack, recycle any cans in the living room, put any clothes on the floor into the laundry basket, check the mouse traps, right? So the robot goes and reads the whiteboard. I'm just skimming ahead here, but then it goes and moves the shoes that it's supposed to move. It goes and it crushes the can and throws it in the recycling bin, goes and finds more cans, crushes them. Again, I'm skimming here. This is not real time. Goes and finds some clothes, put the clothes in the laundry basket. And then let me skim ahead here. And then the final task, take Koda for a walk, and it shows it walk out the door. But that was the end of the video. I just thought that was super cool that it went and looked at a whiteboard and saw what tasks it needed to do, read the tasks, and that was essentially like it getting prompted on the actions it needed to go and take. I'm impressed by that kind of stuff. Like, you can have a robot that lives in your house and every day before you walk out the door, write a bunch of stuff on the whiteboard, and then when you leave, it knows what it needs to go do because you put it on the whiteboard. That's like what I do with my OpenClaw through Slack, but like in the real world. That's kind of sweet. And that's what I got for you today. I actually want to shout out one video that I really think you're going to enjoy. I put this video out on Wednesday. And it's kind of like this AMA video and I had no idea how to title this thing. So the current title, which might be different by the time you're seeing this, 'cause I test a lot, is The Truth About My Channel and How Much Money I Make. And in this video, I actually walk you through how I make my video intros, like all the crazy intros that I do. I walk you through my entire editing process, like how I make these videos that you're watching. I give you a full studio tour of the studio that I'm in right now. I talk about the various computers. I use. I talk about a bunch of the automations that I have that run my business and how they all work. I go off on a few little rants like, will AI kill creators? What was AI like before ChatGPT? My thoughts on ChatGPT's business at the moment. How to actually keep up with AI. And of course, I actually do share my income numbers as well if that's something you're interested in. And the comments all seem to like really, really, really enjoy this video. And I just noticed Wes Roth commented on it and I gave him a little shout out in the video. Some AI bro love there as well. I highly recommend if you want to see behind the scenes of a business that's like mostly run by AI, you also go check out this video. Okay, I'm done rambling and ranting about that. I just wanted to shout it out because I really think more people will enjoy that video. And I'm planning on making a video like that like once a month where I look through all the questions that came in on my various YouTube videos, and then I make a roundup video where I answer all the questions and sort of close any open loops for people. All right, that's what I got for you. Again, there's so much happening every single week, it's getting harder and harder and harder to try to keep up with the news every single day. And I figured the best way that I can provide value to the most amount of people is personally drink from the fire hose every day, keep up with the AI news every single day, every little announcement, every little update, filter through that noise and find the signal, find the things that I think most people would enjoy and would find helpful, and then share just that stuff with you. That's my goal with this channel, is to make this all feel less overwhelming for you and just help you understand what you need to know and what's important. If that's something of interest to you, consider liking and subscribing to this channel. I'll do my best to keep you looped in every single Friday with just what you need to know and do my best to cut out the fluff and the noise and the hype as well. Again, I really, really appreciate you. A quick reminder, I do record these on Thursdays, publish them on Fridays. So if any big news came out on Friday, it probably didn't make this video, but it will be in next week's breakdown video. But I had a blast keeping up with all all this news, testing this stuff out, trying it out, and telling you what I think you should know. And again, hopefully it was helpful, and I really, really, really appreciate you hanging out and nerding out with me today. And hopefully I'll see you in the next one. Thanks again for tuning in. Appreciate you. Bye-bye.