AI News: The AI Launch That Crashed The Market

There was a ton of interesting news in the world of AI this week. I'm not going to waste your time. Let's jump straight into it and let's start with a brand new app that was launched by OpenAI this week called the Codex app, which is not to be confused with the brand new model they also released this week called GPT 5.3 Codex. That's an actual large language model they released this week. The Codex app is an app that they released this week for you to actually code inside of. And if you don't pay for ChatGPT, no worries. For a limited time, we're including Codex with chat GPT free and go. So you can actually use it on the free plan to play around with it. They call this one a command center for agents. So you can actually work with multiple agents in parallel, meaning you can get one agent working on a coding project and then open up another one and get that one working on a coding project and then open up another one and get that one going and they will all sort of run in parallel with each other. They also took a play out of the anthropic playbook and give you the ability to use skills inside of Codex. So skills are like bundles of instructions, resources and scripts that the Codex can then connect to and run these various workflows on your behalf. They have a whole bunch of examples of different things they've coded up and I will link to this article in the description if you want to take a peek at some of these. But you can see they built it looks like the same game multiple times. One using only 60,000 tokens, one using 800,000 tokens, and one using 7 million tokens. And you can see how the game sort of improved with each time they ran through it. Now I have downloaded the Codex app myself and this is what it looks like. It's essentially an ide. Like if you ever use VS code or Cursor or Windsurf or that type of thing. However, this one to me feels like it's designed to be simpler. Like there's just less going on, less options, less things to think about. You just tell it what you want and it goes off and does the thing. It writes the code. You don't really have to have all sorts of different things going on on the screen that you might normally have in one of those other development environments. It's also one of the few ways that as of this recording, you can use their brand new model 5.3 Codex. They give you the ability up here to Open whatever you're building in different IDEs. So if you have Cursor, Windsurf, Anti gravity, Warp, you name it. Whatever you code you can jump over and sort of go more in depth with in one of those platforms. They've also got easy built in get actions and you can connect it directly to GitHub and sort of save your progress in GitHub and keep all of your code stored in the GitHub servers. You've got the ability to toggle on and off a terminal down here and you've got the diff panel where you can see what COD changes it's making when it does make code changes. You've also got your threads panel over here you've got a skills section where you can install new skills which really cool by the way that they just sort of bake this in right here. You've got various automations in here. So you can start with a template. So let's test it out. Let's go ahead and leave it on the GPT 5.3 Codex, their newest model. And I'm going to create a new project up here. I'll call this one Space Game. We'll open this and we can see we've got a thread for Space Game here. I'm gonna give it a super simple prompt. Build me a retro space shooter with five levels using three js and let's just let it go and it's gonna go off and start building that. But one of the big benefits of using codecs is parallel agents. I can have multiple projects going at the same time. So let's come up here and let's create a new project. And I am going to call this one Portfolio. Portfolio Website. We'll create that. Open this folder. And now we have a new project called Portfolio Website. I'll give it some instructions. Build me a single page personal portfolio website using HTML, CSS and JavaScript. Include a hero section with my name Matt Wolf and about section. A grid of six placeholder project cards and a contact form. Make it dark mode with modern minimal aesthetic. Let's go ahead and let it work on that one. And look our space game still going down here. And now our portfolio website's being built. But let's build something else too. Let's add a new project. I'm going to switch folders here. I'll call this one Pomodoro Timer. We'll open this one. Now we have another project. I'll give it a prompt here. Build a browser based Pomodoro Timer app. You could go ahead and pause if you want to read the whole prompt. Let's go ahead and get this one started. But this time let's actually invoke a skill as well. You got web design guidelines here and front end design. Let's go ahead and use this front end design skill with it. I'll just say use the front end design skill. That's all you have to do if you want to use one of those skills. Just tell it you want to use one of those skills. And now, as you can see, I've got it building a space game for me. I've got it building a portfolio website. I've got it building a Pomodoro timer for me. And they're all working in parallel. It looks like my portfolio website is done. So let's open this guy up. And here's the website it created for me. It's got my hero section with my name and about sections, six projects and a contact form, and it'll automatically scroll for me. That was just a single prompt. I mean, it's a very basic website, but. But it's a decent start. All right, jumping back to Codex here, I can see our pomodoro still going. But we have a retro space shooter that's been built already. Let's click into this one. And to get this to play, I'm just going to say run this for me. It should be able to do the terminal commands for me. All right, I'm starting the server. Server launched. And here's the space game that it made for us. All right, so fire is space bar wasd, keys to move around. All right, enter to launch. Oh, wow, that's really hard. It looks pretty cool, though. Try this again. This is like, impossible. And it also kind of freezes up. So I need to do a little bit of bug review because it just freezes when you die. But, you know, one or two more prompts and we're there. And then finally, let's check out our Pomodoro timer. Let's open this index file here that it created. And here's our Pomodoro timer. Let's click start. We get a little countdown that's working, a little animated, you know, line that goes around with the circle there. We could pause it, we could reset it. So this seems to be working. I did tell it to make a sound when the timer's up. I'm not going to wait 25 minutes to find out. I'm just going to assume that it does work. But that's Codex. You can literally spin up a whole bunch of agents building a whole bunch of projects at once. So if you're somebody that has a lot of ideas and you want them all running, you just fire them each up. Now I actually have a video coming out on the Next Wave on Tuesday where we tried Codex and I didn't quite get it yet. You'll see my frustration with it, but after learning about it a little bit, I've sort of figured out a little bit better and kind of understand it a little more now. So when you see that on Tuesday and you see me struggling, well, you can come back to this video and see that I figured it out by now. But this week's also been a busy week for Anthropic. They also released another model this week in Claude Opus 4.6, which I made a video about earlier this week breaking that down. But they also released the ability to add plugins inside of Cowork, and when they added this feature, it freaked a lot of people out. So much so that it actually caused stock prices to go down in the tech industry because of this new plugin feature. New plugins tailored for specific industries like sales, finance, data marketing and legal launched on Friday that sent software stocks into a spiral, raising fears that the bread and butter software as a service business model that's driven the tech and digital service industries and is now at risk of disruption. If that's not an understatement, I don't know what is. The SaaS world has been at risk of disruption for the past three years. We're just seeing these coding models get good enough where everybody's like, oh crap, this happened faster than we thought. So what are these plugins exactly? Well, they're kind of similar to skills that we just saw in Codex, but a little bit more depth to them. They can actually have multiple skills. They can connect to APIs, they can do a lot more. And these plugins are specifically designed to work inside of Claude Cowork. So we did cover Claude Cowork in a previous video, but let's go ahead and fire it up real quick here. And when we open up Cowork, we can see that assuming you've updated it, you've got a plugins button here. I can click into plugins and we can see the types of plugins that are available. There's plugins for bio, research, customer support, data, enterprise, search, finance, legal, marketing, product management, productivity, and sales that are specifically built by Anthropic. But it looks like you can actually add custom plugins and pull stuff from GitHub. Not something I've experimented with a ton. But this is why the market's freaked out. They're like, this could replace a lot of SaaS products if people just install a plugin that does what our SaaS used to do. And if you don't know what SaaS means, it's a software as a service. It's things like CRMs, stats, dashboards, you know, various things that charge you a monthly fee to use them. And well, with things like Codex and like Claude Cowork with plugins you could just kind of build a lot of your own tools. Now, are a lot of big enterprise companies going to go and do this? Probably not for a while. They'll probably still rely on a lot of the enterprise solutions that are out there. But you know, it's only a matter of time before every company's hiring an in house vibe coder who's sort of vibe coding up their software to replace the marketing stack that they're paying six figures a month for. You know, I can see why companies are concerned now. I'm going to be honest, keeping up with AI news right now is borderline impossible. And I know a lot of you guys rely on these videos to help you keep up without losing your minds. But what helps me keep up is actually just a lot of AI tools. When I'm researching, I spend like 99% of my time in my Internet browser. So transitioning to an AI browser just made a lot of sense for me. And the one that I use every day is Comet, because it not only integrates AI into web searching, but it can perform tasks for me right in my browser. So while I'm doing my normal research routine, you know, reading articles and looking up press releases and watching launch videos, the AI is already working in the background and baked into every page I open. For example, I'll open up a bunch of announcements from different companies and instead of reading through every single line, I'll ask Comet to pull out what's actually new and what's marketing fluff and what's actually genuinely worth talking about. And then I'll jump over to social media in the comment sections and have Comet summarize what people are excited about versus what they're skeptical of, which is huge for deciding what angles to cover in my videos. If I'm watching a product demo on YouTube, I'll ask Comet to jump to the most important moments, grab quotes, and tell me what the new feature or product actually does in plain English. And when I'm done, I'LL have Comet organize all of my tabs into a single research workspace and export a clean outline straight into a document without needing to copy and paste and fix wonky formatting. This is what makes Comet really stand out, because it's not just Chrome plus ChatGPT. It's like my own assistant that can do the grunt work for me and take action with the tools I'm already logged into. What makes this all work so well is that Comet can actually see what I'm seeing and understand the page and context. Comet is free to use, so if you're interested, check out the link in the description. And thank you so much to Perplexity Comet for sponsoring this portion of today's video. All right, moving along. We actually got a new video model out of XAI this week with Grok imagine 1.0 so you can do 10 second videos now at 720p resolution with dramatically better audio. Now in my last week news video I showed you that they released this in the API. Well now you can just do it straight over@grok.com Imagine. We could skim through some of these videos here and you can see the kinds of quality that this model is putting out. It's actually pretty decent. I mean none of these videos are fooling me. Even these ones that are supposed to look realistic still have this like plasticky look to them. But I mean it's, it's a decent model. I actually don't think GROK gets enough credit. Again, I said this last week. I think a lot of people don't like XAI models, mostly because they don't like Elon and not because the models aren't good. The GROK large language model is actually a pretty decent model and their video model and image model is solid as well. Is it as solid as this next one? Probably not quite because this week Cling AI introduced a brand new model called Cling 3.0. If we look at some of their videos, I'm not gonna play the audio cause a lot of times they put music in here that I can't, you know, show and stuff. But if you look at this, the realism of these videos is so much better. I'm going full screen here so you can see a better look at what these look like. And yeah, this model is probably going to be considered the best model for videos right now. It's really, really good as far as realism goes. This is as good as any model I've seen so far and you can also create 15 second clips with it. It's got upgraded native audio, so I think the lip syncing is even a little bit better. So it doesn't look like it's fully rolled out on their website yet, but from what I understand, it is available via the API. So I think there are some other sites where you can use it. So I jumped over to Kreia AI here. I flipped it to video mode and when I click on the settings here and click a model, I actually see cling 3.0. So it looks like you can actually use cling 3 in krea and generate all the way up to 15 seconds. I gave it an hour and it still hasn't generated yet, so I don't think it's working. Moving on, here's something from Krea that I did test and I do know is working and that is the Korea Real Time on your phone. This is actually pretty cool. So here's the Krea app and you can see on the screen there's a button in the bottom left that says Real Time. And if I click on that, it turns on my camera. And look at that. This is actually in real time. Me playing with this on the camera. It's actually pretty cool. It's pretty fun. Now let me switch to wireframe. And now you can see my wireframe here. Let me switch my camera around so you're seeing what I'm looking at. And now you can see like there's my camera, there's my lights, there's another camera, more lights. Pretty cool. You can. You can see my bookshelf kind of in the background with a bunch of equipment and stuff on it. There's another microphone, speakers, mixing boards. Pretty sweet. All right, now let's flip it back to me and let's set myself on fire. My room's on fire. Ah, it's pretty fun. Pretty cool. I like it. Let's give our own prompt. Fish man under the ocean. Ooh, look at that. Now, apparently I'm playing a guitar under the ocean. That's interesting. But I can move it around and you can see it's changing. It's kind of interpreting it. Weird. Their built in filters seem to be better than when I do them myself. Now I'm a statue. Ooh, that makes me look really old. But it's fun. I actually really enjoy this. Hey, hey, look at that guy. Let's give it a thumbs up. Yeah. Peace out. Okay, that's enough of that. It's pretty fun. We got a new update out of Ideogram here. You can now do prompt based editing on IDEOGRAM and apparently you can try it for free. So let's do that. Why don't I just start by taking an image that somebody else made and let's just edit that one and let's say give her a Padres base ball cap. Nice. It took about, I don't know, 25 seconds or so. Now I believe we're editing this second image so I can say, make the background a baseball game. Not only did it make it a baseball game, it actually made it Petco Park. That's pretty cool. Give her purple sunglasses. Nice. It's actually pretty smooth workflow. I think if they could just speed up a little bit, I'd like it a little bit more waiting. You know, that 20 to 30 seconds, I get kind of impatient. But I'm really impressed. I mean, it even got the logo right. It put the right ballpark in the background with very little context. If you ever need to just make some small tweaks to an image, ideogram's a great free way to do it. Now, we did get some new large language models this week, including the new Claude Opus 4.6 model. Now, this model is really, really proficient at coding and agentic tasks. Here's the benchmarks, if you're interested. And this one's available right now inside of the Claude website. The Claude app, CLAUDE cowork, Claude code. They pretty much made it widely available right away. You can see over on the Claude website over here we now have the option for Opus 4.6 and you can even turn on extended thinking if you want it to spend a little bit more time thinking through. We also got another model out of OpenAI this week. In fact, they came out 15 minutes apart from each other. They were both going to release at 10am Anthropic decided that they wanted to front run OpenAI a little bit and released it 15 minutes early so they can get in front of OpenAI. But this one is called GPT 5.3 Codex, and it's also a model designed for coding as well as agentic tasks. They give us a lot less benchmark data on this one. But we can see the data they did give us is really geared towards agentic tasks and towards coding. And I'm not going too deep into either of those models in this video because I did make a breakdown of both of those models in this video here called New Claude and ChatGPT models just dropped. It's war. So check out that video if you want to go a little bit deeper on those two new models. Again, they're very focused on coding and agentic tasks. Things like the open claw tool that people have been playing with a lot lately. What I do want to show you though, is I came across this from Alistair McLay and he built this, like, flight simulator game where you can fly around in New York, Hong Kong, Rio, your hometown, literally anywhere. And it's actually really impressive. I think it's useful using Google Maps maybe. But check this out. They've got these cities here that you can pick, and when I try to actually search a different city, that seems to be broken. But let's go ahead and just do San Francisco and let's pick a strike fighter and then let's take off. And you can see here's the game. I'm using my ASWD keys here. I can fly down, I can fly up. Spacebar actually shoots guns, so I could like shoot at other planes if I want, see if I can take one down. Ooh, I actually got some points for that. But yeah, it's pretty cool. Like, this is blowing my mind. Somebody vibe coded this and they said they did it in an hour. Let's just go ahead and crash us into the ground and see what happens. Oh, I'm gonna collide with another plane. Nope, I went through the other plane. Okay. I just blew myself up. Pretty impressive though, for something that they said they coded in about one hour using Opus 4.6 and GPT 5.3. So they use both the new models that came out today, and within little over an hour they have a game that looks like that. Kind of crazy where we're getting with this stuff. If you want to try this game, I will link up to this tweet in the description below so you can check it out. We do also have some AI drama this week. Anthropic showed off the ads that they're going to be running during the super bowl. And I'm not going to play the ads for you. I actually also discussed these ads in this new Claude GPT models video that I mentioned earlier. But basically the deal is they made these ads. And in the ads, the person is talking to an AI and halfway through talking to the AI, they start selling them stuff. And then at the very end, the ad says, ads are coming to AI but not to Claude. And they have several ads with pretty much the same sentiment. People talking to another person. Halfway through the conversation, the person tries to pitch them something, and then it says, hey, ads are coming to AI but not to Claude. And there's very mixed emotions online about this. All the, like Super OpenAI fans are going, oh, this is super deceptive. We know that's not how the ads are going to be. The ads are going to be outside of the chat, not actually part of the chat response, which is true. So they're saying Claude is like, super deceptive. But they're also helping this ad get more reach by retweeting it, talking about how deceptive it is, bringing more attention to Claude. And then you have all the anthropic people saying, yeah, this ad is perfect. Like, we hate the chatgy bringing ads into their model. And we love that Claude is standing up against that and basically planting a flag in the sand and saying, hey, we will never do ads, because now they kind of can't, right? Like, let's say two years from now, Claude decides they want to put ads in their platform because they need the money. Well, everybody's going to point back to these ads and be like, you hypocrites. Look what you said two years ago. And Sam Altman did respond to the ad, essentially Streisand affecting it, right? Because his response has 9.3 million views. The ad itself only has 4.1. So his response to this has double the amount of views as the ad that's circulating on Twitter. So, I mean, I don't know if this was a smart move to even respond, but he said, first, the good part of the anthropic ads, they are funny. And I laughed. But then he goes on to say, I wonder why anthropic would go for something so clearly dishonest. Our most important principle for ads say that we won't do exactly this. We would obviously never run ads in the way anthropic depicts them. We are not stupid, and we know our users would reject that. They also get a little jab in there. More Texans use chat GPT for free than total people use Claude in the U.S. so, yeah, there's a very sort of like, public beef going on over these ads. The Super Bowls this weekend. So, I mean, if you're watching this after the weekend, you've probably already seen these ads. If you're watching it before the weekend, I'll link it all up in the description. All right, I have a few more things I want to share with you that I found interesting this week. So let's jump into a rapid fire. Starting with something that was pretty big news towards the end of last week and through the weekend was the new social media network for AI agents. Like social network, not even for humans. It's called Mult Book. It's basically Reddit, but only agents could go on and communicate with each other. People sort of freaked out about it because agents were on there saying things like, am I actually conscious or am I just pattern matching what consciousness feels like and saying all of these things that were creeping people out, making them think agents had sentience. But in reality it was users telling their agents to go and say that kind of stuff on Molt Book. It was kind of this weird thing where some people were freaking out because they thought, oh my gosh, we're finally hitting the singularity, AGI is here, whatever. But the reality of it was humans were telling AI bots to go and post stuff to make people think that they were sentient. So it was really, really weird. Also, it turned out it was a huge security risk. People were figuring out how to get other people's like private API keys and passwords and things like that by sort of like hacking the back end of the site. And it was just this big kind of mess. I think they've actually got a lot of that stuff patched up. But yeah, this was kind of a whirlwind for a few days here. And if you do want to dive deeper and learn more about this story, this is another one where I did make a dedicated video about it. You can see it here. Autonomous AI agents have gone too far. So check that video out if you want to learn more about the story that I'm referring to right now, because I did sort of deep dive and explain it and also talked about a whole bunch of additional crazy stuff that came out after molt book. OpenAI also had another launch this week, a little bit less interesting to general consumers, but they launched something called OpenAI Frontier. This is a platform that helps enterprises build, deploy and manage AI agents that can do real work. Frontier gives agents the same skills people need to succeed at work. Shared context onboarding, hands on learning with feedback and clear permissions and boundaries. So we can see here. Frontier lets enterprises connect data warehouses, CRMs, ticketing systems, and internal apps into a shared context layer so multiple agents can see the same institutional knowledge instead of being isolated bots. So again, super interesting and helpful if you're running a large business where you need something like this, but not something I've really played with or that I think most consumers are going to need access to at the moment. Perplexity had some news this week as well. They rolled out an advanced version of their Perplexity Deep Research, and now they're the most powerful deep research model out there. According to this benchmark here, scoring quite a bit above Gemini Deep Research, which in my opinion was the a deep research agent that was the best that I've used. So I need to test this one out some more. But they do have this new Deep Research agent available. The update is available to all Max users right away. So you know they're expensive, like $200 a month plan. And they also introduced their Model Council. This is one where when you ask it questions, it can actually ask multiple models and they all sort of come to a conclusion together. So when you select Model Council in the main Perplexity interface, your query runs across three of the models available on Perplexity at once. SO Cloud Opus 4.6 GPT 5.2 and Gemini 3.0, a synthesizer model reviews the outputs, resolves conflicts where possible, and gives you one answer that shows where the models agree and where they differ. So pretty cool to go deep on a topic and actually get all of these models sort of working together and coming to a consensus. But like the other thing that I just mentioned, Model Council is available now for Perplexity Max subscribers on the web, if you're curious. Perplexity Max is $167 a month if you pay for a year in advance. So basically two grand. I think they also have a monthly plan that's 200 bucks a month. So in order to use the new Deep Research and in order to use the Perplexity Council, you have to be on that higher end plan, which I'm not currently a member of. 11 Labs dropped a new text to speech model this week and they say that 72% of people prefer this mod over the previous release. It's got accuracy improvements like it's better at actually speaking out phone numbers, they say. In the past it would have said plus 49, 179,876,000, blah blah blah. Right now it'll actually speak it out like a phone number. It's better at speaking out currencies and chemical formulas and sports scores and things like that. And it's publicly available now and then audio going in the other direction. Speech to text model. Mistral dropped a new model this week called VoxTral Transcribe 2 and it's an open source speech model that runs on device for pennies. So if you want a local model that's installed on your computer directly where you can speak and have it transcribed, similar to like OpenAI's Whisper model or Assembly AI's models, you now have a open source, free to run local model that you can use. After seeing Google's Genie 3 recently, Roblox wanted to get in on that same action and created Roblox Cube Foundation Model. This is a 4D generation model powered by the Cube foundation model. The technology behind 4D generation and cube will eventually allow any creator to generate full scenes, including assets, environments, code, animations and more with natural language prompts. So they've got some pictures here of stuff that was generated with this model. So it can generate the assets, it can generate the worlds, it can generate pretty much everything in a Roblox world. But I don't believe this is available yet. And we got a big merger this week. Elon Musk took an Elon Musk company and merged it with another Elon Musk company when Elon Musk bought another one of Elon Musk's companies. So Elon Musk merged SpaceX, a company owned by Elon Musk, with Xai. They seem to think that if they could put data centers in space, they can use solar power 24. 7 to power these data centers. They don't take up any space on Earth, they don't use up any water on Earth. But I do know there are a lot of engineering hurdles before they actually get to be able to do that. I think there's still some like major heat dissipation issues that they need to figure out. I don't debate that they actually will figure it out. I think they will eventually. I just feel like they're just selling it as something that can happen, likely a lot sooner than it actually is able to happen, I guess. But it'll be interesting to see how that plays out. Maybe data centers in space are going to happen this year? I don't think so, but maybe. But now Elon Musk Co. Doesn't have to to negotiate with other Elon Musk Co. Because now they're both just one company. So that's good. But that's what I got for you today. A lot, a lot happened this week. It was a pretty crazy, pretty busy week. Watch the super bowl this weekend. There's going to be a ton of AI ads. Google has ads about Gemini. OpenAI has ads about ChatGPT, Anthropic has ads about Claude. And that's probably just scratching the surface of the AI ads we're going to see this week. The super bowl is going to be AI ad fest pretty much. So that should be fun. Now, before I wrap up, I should mention I record this on Thursday, Tuesdays, publish them on Fridays. So if any new news came out on Friday, I might have missed it. And if that's the case, I'll share it in next week's video or, you know, one of my other breakdowns that I do. And if you like staying looped in on all the latest AI news and you want AI tutorials and you want to be fully tapped in, maybe consider liking this video and subscribing to this channel. And I will keep you super looped in on everything that's going on so you could be the smartest person among your friends when it comes to AI, at least. Also, check out FutureTools IE and join the newsletter every week. I said just two newsletters a week with the coolest news that I come across and the coolest tools that I come across. And I'll deliver it directly to your inbox and that'll help you even further. Stay looped in. And it's totally free. And I give you the AI Income database, which is like a whole bunch of ways to use AI to make money. You get that for free when you subscribe to the newsletter, and you can do that@FutureTools IO. But again, that's what I got for you. Thanks so much for hanging out with me. Nerding out with me. Hopefully I'll see you in the next one. Bye. Bye.