AI News: Fable Banned, New Open-Source Leader, Midjourney Shocker

It was a wild week in the world of AI and I don't wanna waste your time, so let's just get right into it. Starting with the story that pretty much rocked the AI world this week. For the first time ever, the US government forced a publicly released commercial AI model off of the market. Anthropic had to shut down Mythos-5 and Fable-5 for every user on Earth. And in last week's news video, I spent a good chunk of the video talking about how much I loved this model, but unfortunately we only got it for less than a week. On Friday evening, June 12th, it was gone. So here's what happened. The US government basically said, "You need to suspend all access to Fable 5 and Mythos 5 to any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." Now that's a nearly impossible task. They really had no way of shutting it down for everybody but us. American citizens. So the net effect, they must abruptly disable Fable 5 and Mythos 5 for all customers to ensure compliance. Now Anthropic claimed it was over a very minor vulnerability and went on to talk about all of the ways they were sort of securing Anthropic so that it shouldn't be an issue. Now in my opinion, Anthropic kind of brought this upon themselves. Back in April when they first announced Mythos, they pretty much restricted it for everybody because it was just too powerful to release openly. They literally claimed that it could wreak havoc in the wrong hands. Wrong hands. The White House also claims they'd heard Omidai liken the dangers of Anthropic's technology to a nuclear bomb. So they literally spent months talking about how dangerous and how scary this model is. And then last week, just days before any of this happened, Dario published this essay literally arguing that the government should be able to block or reverse a model's release, like FAA style. Frontier AI models, like airplanes, should be required to go through technical testing and auditing and the release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety. In the same essay, he said, we now globally and collectively need to activate a slow and rickety policy apparatus to deal with risks and opportunities that are going to compound surprisingly quickly from here. And well, the government did exactly what he was claiming the government should do in his essay. And now the loudest advocate for AI regulation is frustrated that it's getting regulated. According to David Sachs here on X, a highly credible trusted partner of both Anthropic and the US government who was testing Fable came forward with a jailbreak of the guardrails. The admin asked Dario to fix the jailbreak or de-deploy the model, and Dario refused. Of course, in the blog post where Dario announced they were shutting down Fable, he did claim that the jailbreak isn't that serious. Now, according to David Sachs here, the admin issued the export control, but they did it reluctantly. He goes on to say, it's frankly bewildering that Anthropic hasn't wanted to comply with safety requests that it previously said were its highest priority. And also, as a random side note that's like super fascinating about this whole thing, is that the whistleblower, you know, the trusted partner that went to the US government, turns out that it was Amazon CEO Andy Jassy. Jassy was among the tech leaders who raised concerns to senior Trump administration officials this week about security risks and Anthropic's most advanced models. What makes this so interesting is that Amazon is one of Anthropic's biggest investors and vendors. And that's what makes it so bizarre, is it doesn't seem to be in Amazon's benefit for this to actually get shut down because Amazon has so much invested in Anthropic. Now my take on this is kind of that it's less about security and more about like the government and Anthropic butting heads. Like you get the real impression that the US government just does not like Anthropic. According to this article on Political, the crux of the issue was the lack of seriousness that Anthropic was applying to it. Had Anthropic taken it seriously and rather than dismissing it as isolated, moved to fix or PAUSE] access, this would've never happened. Combine that with the drama of the US government deeming Anthropic a supply chain risk, Pete Hedgeth sort of talking crap about Dario on Twitter, all of this kind of stuff makes it really evident to me that the US government just doesn't like Anthropic. Anthropic stands so heavily on their principles, and when Anthropic actually dismissed this as like a no big deal vulnerability, the US government went, fine, shut it all down. You're not taking us seriously. But again, the facts still stand. Understands that Dario did spend months saying that this is really, really scary. It could wreak havoc in the wrong hands and that this stuff absolutely needs to be regulated. Like, he spent a lot of time saying that kind of stuff. Now cybersecurity defenders are trying to come to the rescue, and a bunch of them all signed this petition here to bring back these models, basically saying, you're taking the best safeguard we have against cybersecurity issues away from us. Going on to say, to pull the best capabilities away from defenders without a good reason when our adversaries are rapidly Dancing is dangerous. They also claim that these Mythos class models are not uniquely good at these tasks. They do say they are quite good at finding flaws and weaponizing exploits, but they're not uniquely good at this, basically meaning the other models can do this stuff too, and they're still out there. Now, there is a very good chance we'll be getting these models back pretty soon. I don't know how accurate this article is here, but according to Korea Joongang Daily, Anthropic's confident of re-enabling Mythos Fable 5 access in the coming days. Now, I think one of the bigger picture issues here is more about the precedent, right? Anthropic is worth almost $1 trillion. The last valuation had it at $965 billion. If you're an investor in one of these AI companies, it makes it extremely scary that the government can basically say, shut this off, and you just have to do it overnight. Now, personally, I have very mixed feelings about this whole drama between Anthropic and the US government, because again, on one hand, Anthropic was essentially begging the government to do this to them. On the other hand, I really liked the Fable model. It was really good at coding up apps quickly with just like one shot. If you watched my last week's news video, we made some really cool stuff with it. At the end of the day, we don't know exactly what went on behind the scenes with the government and Anthropic. So almost everything we're hearing is just sort of speculation and, you know, sources told us, but it definitely does make, you know, investing in some of these foundation labs a little bit scarier of an idea. But again, there's a good chance we'll have it back pretty quickly. Maybe even by the time this video is live, Fable is starting to roll back out again. Another cool AI product I wanna show you is from Box. If you're not familiar with them, Box is an intelligent content management platform that helps companies use AI to unlock information and insights hidden inside your content. And this is becoming a much bigger challenge than most people realize. Like I run both my YouTube business and the Future Tools site and newsletter, and even at that scale, I'm dealing with files spread across different folders and formats and tools and systems. Now imagine that problem across an entire business. Box's latest State of AI in the Enterprise report found that 96% of organizations recognize that AI agents need access to company-specific content, but only 36% have actually connected those agents to trusted content across their business. So the bottleneck isn't really the AI models anymore, it's getting the right information to them. That's where Box comes in. Box's AI features like Box Extract, Box Agent, and Box Automate help transform all that scattered content into structured, usable knowledge that can be searched, analyzed, summarized, compared, and put to work by both people and AI agents. And because companies don't want to bet everything on a single AI provider, Box is model agnostic, supporting models from providers like OpenAI, Anthropic, and Google. Their research found that 68% of organizations are concerned about vendor lock-in. So this flexibility is huge for businesses. And I personally like the option of jumping between models when I'm building with AI too. I can see this being especially valuable in industries like financial services, insurance, healthcare, government, and media where massive amounts of content need to be accessed securely and intelligently. Now, if you wanna learn more about Box AI, click the link in the description below. And thanks so much to Box for supporting my channel and sponsoring this portion of today's video. We also got another brand new model this week that actually happens to be an open source model and people are saying it's really good. The company ZAI just dropped GLM 5.2. It's their new open weight flagship model that's built specifically for long horizon coding and agentic tasks. The model has a 1 million token context window, putting it pretty much on par with all of the Frontier Lab models, and it's got an MIT open source license. So theoretically, you could download it and fine-tune it and do whatever you want with it. Now, according to Hugging Face here, it's a 753 billion parameter model. So I mean, most computers aren't gonna actually run this, but you can do whatever you want with it on the cloud. What makes this really interesting is that again, it's an open model, but on some of these benchmarks here, it's like up there with Opus 4.8. It actually beats out GPT-5.5. On SuiBench Pro, it actually performs better than GPT-5.5, but not as good as Claude Opus 4.8. I don't really pay a ton of attention to this. I think the more important model is DeepSui, which, you know, it's not beating GPT-5.5 or Opus on that one specifically, but still really, really, really good numbers. Yeah. For an open weight model. If we look at the code arena here, GLM 5.2 is actually coming in number 2, just behind Claude Fable 5. Now this is a blind test where you give it a prompt, it gives you 2 responses back, and users vote on the best response. And that's how this score is decided. And based on this sort of user testing blind taste test here, the only model that's performing better on the web dev arena is Claude Fable. It's beating Opus 4.8, it's beating GPT-5.5 by a lot. When it comes to the agent arena here, it's right up here in the top 10 along with the Anthropic and GPT models beating out all of the Google models. So yeah, it's a good model. And the fact that it's an open source, open weight model shows that these open models are catching up very, very quickly. And the other thing to keep in mind here is if we come back to our code arena is look at the price per million tokens here. Claude Fable is $10 in, $50 out. GLM 5.2 is $1.40 in, $4.40 out. Opus 4.7 and 4.8 are $5 and $25. So you're gonna get like close to Claude Opus 4.8 level coding ability, but at like a quarter of the cost. So according to all the benchmarks and according to the leaderboards here, this is the new state-of-the-art open model. So of course I had to put it to the test. I came over to z.ai here and then selected the GLM 5.2 model and I wanted to test it with a similar prompt that I gave Fable. I wanted to see how good it can recreate that Megabonk game. That's kind of like my current testing benchmark to compare to. So I literally just gave it the prompt, create a Megabonk clone. Now I obviously gave it that prompt in English and all of its response here I believe is in Chinese, but I could go ahead and select all this and translate section to English, translate full page. And there we go. I've created a complete Megabonk style clone game for you. Core gameplay, visual design, feedback system. And let's see what this looks like. This screen looks good here. Let's open it in a full screen here. And off to a good start. I mean, aesthetically it looks good. Uh-oh, nothing happens. I can't start bonking apparently. All right, so that was just one prompt. Let's go ahead and give it another prompt over here and just say nothing happens when I click start bonking. I can only see the home screen. Now while we're waiting for that, one of the other things that this GLM 5.2 model is good at is creating slides. So I went and gave it the prompt, create a beautiful and accurate slide presentation for me that walks through what an agent loop is, how they work, and several useful examples of them. This time it actually gave me the response in English. I don't know why it did it in Chinese on the last one, but it gave me this slide deck presentation. It's 10 slides and it opens it in another webpage, a field guide to agent loops, how autonomous AI systems think, act, observe, and repeat until the work is done. A model that keeps going. An agent loop is this cyclic control flow that turns a one-shot language model into something that acts in the world, sees what happens, and tries again. So I mean, it looks pretty clean. I like the little graph here. Nothing like ultra insanely impressive, but definitely clean. 4 moves repeating, perceive, reason, act, observe, and then it loops through less than 20 lines of structure. The REACT loop, the original in 2022. So this says pattern 1 of 4. So now I think it's giving us those use cases here. The tool use loop, the self-correcting loop, many loops, one goal. You've got the coordinator agent with a researcher, analyst, writer, and Loops are not free. Reach for a loop when the task is genuinely multi-step, intermediate results change the plan, the agent has real tool calls, you want an auditable trail. Skip the loop when one forward pass will do, latency is the product, cost compounds per turn, you can't bound the iteration count. And then there's the final slide. Again, it's a clean presentation. Am I blown away? Not really, but it looks good. All right, jumping back to our Megabonk, we can see that it's writing code. It's actually responding to me in English this time. That's good. Okay, so it explains the problem to me and what it did. Did to fix it. It appears to be done. Let's go ahead and try it. Still not working. All right, so after the third prompt, it did work. So here's what we got. I could finally start bonking. This looks nothing like Mega Bonk. So it had no idea, I think, what Mega Bonk is. It just went off and made its own bouncy game. Also, spacebar is supposed to make me jump, but spacebar is not doing anything at all. Okay, it's working sometimes. Space or click to jump. Spacebar. Spacebar. There, oh, it jumped that time. Now it jumped. It's like, it's janky. So does it hold up to Fable level of coding? Definitely not. I do like the color scheme though. If I'm looking for one sort of positive out of it, it's got a good color scheme. All right, let's ask our like gotcha question here. Instead of max, let's just go to high. My car needs to be washed. The car wash is 100 meters from me. Should I walk? Or should I drive? You should definitely drive. If you walk to the car wash, your car will be left behind and won't get washed. Okay. Okay. So it could answer that. At this point though, I'm starting to wonder if like this is fine-tuned into the training data. I don't know. Anyway, it's a promising new model and I don't think I've paid anything. Anyway, you can play with it over at chat.zai. And as far as I can tell, it is currently free to use 'cause I haven't paid anything and I can't even find a way to pay if I wanted to. Meta rolled out some new AI AI features on Facebook. I haven't personally tried any of these yet 'cause, well, I don't really use Facebook as much as I used to. I do have a Facebook account. I just don't think I've logged into it in like 6 months, 9 months. It, it, it's been a long time. Actually, let me know in the comments if you use Facebook and if it's worth using. I just haven't in a long time and had no plans on going back. But let's check out some of these AI features because I don't know if this makes Facebook better or worse. Better. So Facebook has a new AI mode and it's a new way to get answers to your questions right on Facebook. So AI mode uses Meta AI to give you answers grounded in what people are saying publicly across our apps like Groups and Reels. So this sounds to me very similar to what like Grok is doing with X, right? If you go and use Grok, it's grounded in all of the tweets or posts or whatever that are on X, which makes Grok really, really good for like real-time information and real-time sentiment out of users because it is looking at what everybody's tweeting about. So we can see in this example they typed "summer escapes near me" and then it says the Bay Area has tons of quick getaways, coastal towns, lakes, etc., etc. Think Half Moon Bay. It pulled up some images I'm guessing that people posted on Facebook and then it gave some additional prompts that they can use like wine country day trips from San Francisco, or then it says popular day trips. And apparently this is all being informed by what people are posting on Facebook, I guess. They also introduced new editing capabilities that leverage AI. So inside your camera world, There's like collage cutout templates. There's new transition effects. And apparently this stuff is opt-in only and can be turned off. There's also new photo presets. So you can actually change your clothing and hair. You can have like a picture where you're wearing the clothes you're wearing and then you press a button and it'll swap out your shirt to like be your favorite football team shirt. I mean, it's all kind of gimmicky. It's probably not gonna make me start using Facebook. All right, there's one more quick rabbit hole I want to go down before we jump into a rapid fire. And that's the— What's the big announcement that came out of Midjourney? And if you didn't hear about this announcement yet and you know what Midjourney is, you know, the AI image generation platform, you're never gonna guess what they just released. Midjourney just started a new branch called Midjourney Medical. And the first thing that they showed off was essentially this like replacement for an MRI machine. It's this like dunk tank that like lowers you down into the water and then uses ultrasound to bounce sound waves around inside the water and then it uses those sound waves to get like an image of your body and your internals to detect issues and stuff. Basically, it's like designed to do what an MRI does, but at like 1/60th of the speed and 1/10th of the cost, I think they said, or maybe that's the other way around. Basically, it's like an MRI-type machine that, uh, does it really quickly and really inexpensively. So here's the video that they put out that sort of explains it. You've got all these little transducers here that are both speakers And microphones. And we can see as it zooms out here, there's like thousands and thousands of 'em, just shy of about 9,000 of these. And then these things are put in this like ring so that the audio waves that are coming out and being collected just bounce around inside the ring. And then of course your body goes down in these rings and then the sound waves are sort of bounced around everywhere to get that imaging of your body. You can see this is like the sound waves sort of bouncing around and creating that image. And they can see up to 25 different biological biological structures. So like, you know, spinal cord and your rectus abdominis muscles and your genome. And at the end of it, it creates like this whole scan of your body similar to an MRI that can all be sort of sliced out and looked at in like layers. Now you might be thinking, that's cool. That seems like something that they'll put into hospitals and make it a lot less expensive to do scanning on people. That's not their plan. Their plan is to create what they call the Midjourney Spa. The first spa is gonna be opening in San Francisco in 2027 and it's They're gonna have hot tubs, saunas, cold plunges, cozy rooms, and of course these body scanners. So you're gonna have like this whole spa-like experience where you can go get scanned and then go chill in the spa or the sauna or whatever. Like they're trying to make this a cool, relaxing, cozy hangout where people wanna go. And the idea being you go get scanned more frequently so that you can detect when problems happen because like you get scanned once and maybe it detects some stuff that could be completely benign but if you get scanned multiple times, then you can see if things are like growing or if things are becoming more and more of an issue. One thing that makes this super interesting though is that Midjourney has no investors. They're completely bootstrapping this. They've made, you know, tons and tons of money from Midjourney and they're taking that money and they're using it to go and do this, like something actually useful in the health space, something that's actually going to cut the costs of something that was oftentimes prohibitively expensive for a lot of people. It sounds like they have this goal of like democratizing health and democratizing advertising the ability to get these kinds of scans, which I think is a, a cool and noble effort. Like you wanna see brainpower going towards stuff like this. Now, if you're wondering why Midjourney would be the company to go and do something like this, well, I see a couple reasons. One, they're like an imaging company. So if you, if you don't know much about David Holtz, the CEO and founder of Midjourney, before Midjourney, he had a sensor company that made it so like you could do motion tracking with your hands. Like there was little sensors that can see what your hands were doing. And so you can like drag things around on your screen and almost do like Minority Report sort of stuff with your hands, and it manipulated what was going on on your screen. He has a background in sensors, right? And then he got into Midjourney, which is image generation. This almost feels like the evolution of those two things, just sort of like smashing those kinds of projects together, like sensors and image generation. And I think the bigger play here is probably more of like a data collection for AI play. Like, if he gets hundreds of thousands of people in these scanners, Well, now he's getting tons and tons and tons of data on people's like bodies and can probably better help predict things based on more and more data that they're able to collect. So I think there's like a bigger AI play here where they're going to likely train models on all of the scans that they collect. I don't know though, I'm just speculating. But there are some people that aren't convinced. Hank Green here posted this on X called, "Let's talk about whole body ultrasounds." He personally doesn't like the fact that they're comparing it to MRIs and CT scans. Claiming that all of these scans are better at different things. So basically claiming this is a replacement for an MRI is not quite correct because an MRI is going to do certain things a lot better than what this is going to do. So he says here, MRIs are bad at imaging parts of your body that move. That's why we don't use MRIs to screen for colon cancer or lung cancer. A chest CT is much more likely to catch an early lung cancer, while a colonoscopy can identify colon cancers. Different scans do different things. Now, ultrasound is really good at soft tissues, so fluid-filled things, and things close enough to the surface that sound can get in and useful information can get back out. So he's saying it should be great for looking at thyroid or a breast lump or your testicles or your lymph nodes or gallbladders, kidneys, fetus, blood flow, fluid around organs, lots of abdominal stuff. Ultrasounds will be good at that kind of stuff. But there's also stuff that ultrasounds will not be so good at. Air is a problem because sound waves do not pass smoothly from tissue into air. They mostly bounce. That's why the scanner scanner needs to happen underwater because if you were just in air, this stuff's not going to bounce around properly. But there's also air inside your body, like in your lungs. So the ultrasound will be able to tell you things that are useful about the surface of your lungs, but it's not going to look through your lungs the way a CT scan can. Basically, his argument is don't compare this to a CT scan or an MRI because those will do things that this just can't. Now, I think he's hoping this thing is a success and will do well, and hopefully AI will be be able to fix some of the things that are the shortcomings of a product like this. But again, the point that he's making is that don't say it can do everything an MRI can do when it just can't. And don't say it can do everything a CT scan can. That is just not true. It won't be able to do all of the exact same things. So I guess Hank's argument here is more the way they're marketing it and not the fact that it's not something that's really actually kind of cool. And for me, I just wanted to talk about it because even though at the moment it has nothing to do with AI, I mean, there's future AI implications of it, but right now Now he flat out said there's not even any AI in the system. I just think it's absolutely fascinating that this is the next thing Midjourney was working on. Anyway, I've spent a lot of time covering just a few things and I have quite a bit more I wanna tell you about, so I'll just kind of rattle through 'em quickly in a rapid fire. Pew pew pew pew pew pew. Starting with a cool new feature that OpenAI rolled out inside of their Codex platform called Record and Replay. So this appears to be something for if you have like a repetitive task, you can call upon this record in replay. It will record a video of you doing something on your computer, learn how to do that thing, and then in the future you can just tell it to do that thing again. So the example they give here is, watch me upload this YouTube video so you can handle these uploads for me in the future. We can see down here that it starts recording his screen. He goes through the process of uploading a YouTube video, grabbing some information from a spreadsheet, you know, grabbing the files on his computer, setting it as private, and then he types done. And then it watches the video, understands what he did, Creates a new skill out of it, and now he can go back and he gives it a PNG file, I'm guessing for the thumbnail, a video file, and a subtitle file, drags and drops them into his Codex chat box and says, upload this YouTube video using the YouTube upload skill that we created. It then goes and does the exact same steps. It selects the file, it grabs it from his computer, it uploads it, grabs the thumbnail, grabs the subtitle file, saves it as private. So it followed his exact step-by-step instructions that he did in the original video. That's pretty handy. Claude Design rolled out a new feature. You can now edit directly on the canvas inside of Claude Design, so you can have it generate a design and then just edit it straight from within there. You can now bring in designs from GitHub repos, design files, or raw uploads, and Claude builds with your components, checks its output against your design system, and makes corrections before you even see it. They've also made it easier to move back and forth between Claude and some of your favorite tools. So stuff like Adobe, Base44, Canva, Gamma, Lovable, Miro, Replit, Vercel, Wix, and more more destinations soon so you can actually, you know, design in one and move between them fairly easily. And these new features are available on pretty much any of the paid plans. Perplexity rolled out a new feature called Self-Improving Memory for agents. They say this new Brain feature is a self-improving memory system. It builds a context graph of the work computer performs. At set intervals such as overnight, Brain reviews the context graph and teaches itself how to do the work better. The more work you do, the better and more efficient Brain makes your computer. To me, this sounds a lot like what that Hermes agent doing that everybody's raving about lately where it watches what you're doing, learns what you're doing, learns where it's making mistakes, and then constantly improves itself. Except now you're doing it inside of Perplexity Computer. Brain gets better as you use computer. Agents become more effective at updating context as they learn the projects, connectors, artifacts, and other sources that lead to the best outputs. They also learn from their mistakes, remembering when a user has made a correction or when a source was a dead end. This results in fewer turns, fewer model calls, and better outputs. This feedback loop makes Brain continuously self-improving. So Perplexity Computer is essentially like Perplexity's answer to OpenClaw or Hermes. It's like their agent, but it runs in a cloud instead of on your own computer. They also have what they call Personal Computer, which does run on your own computer, but I think this one just runs on their cloud computers. I didn't see anything in their write-up about Personal Computer. Now, in order to use this, you do have to be on a Max or Enterprise plan, so one of their higher-end subscriptions. I thought this looked interesting as well. I haven't tested it myself yet, but this company Palmier made a video video editor where you use Claude to actually edit the videos. So you can see here they've got their Palmier video editor open, and they've also got Claude open in another window. They say, organize my media in Palmier, and then over in their editor window, you can see all their files get sorted into just a few folders. And then they say, now edit it for me. And it says, good, now applying trims, speeds, and transforms. And it looks like it's just like editing it in the timeline for them and adding sound effects and adding text. And then it uses the various like video models that are out there to actually generate video. So you can see like this eye video is all generated. So you use Claude in combination with this video editor and it will do things like organize your media for you, chop up the video for you, actually generate media that can go into the video if you don't have the exact shot that you need. Something I definitely want to test out. Although when I look at their website, it costs $29 a month, normally $49 a month for 5,000 credits. 5,000 credits. Credits. Now, 5,000 credits gets you 3 to 7 minutes of generated video. I don't know, that seems like that could add up very, very, very quickly if you're using a lot of generated video in whatever you're editing. But still, I'll probably give it a try, maybe test it out in a future video. Since we're talking about video editing, Adobe is adding its AI assistant to Premiere, Illustrator, and InDesign. So you can do stuff inside of Premiere, like make a new sequence with the drone footage or generate images through prompts like this. The idea being that you can use these tools here, Premiere, Illustrator, InDesign, but just text prompt it with what you want instead of actually doing the editing the old-fashioned way. Here's one for my marketer friends. Google introduced what they call Ask Ad Manager. And to me, this looks like pretty much the same thing as Ask Studio that we have in YouTube, where in the backend of YouTube you can click on a video and ask Studio, hey, what is this video doing well? What should be improved? And their little AI will look at all the stats on the video and tell you what you could do to improve that video. Well, you're getting that same functionality inside of the Ad Manager now. So you could be running ads inside of Google Ads and get a little chatbot to figure out how to improve or optimize your ads or, you know, get additional data out of your ads, things like that. I used to run a lot of Google Ads in a previous life when I did a lot of digital marketing, and this would've been really, really cool to have. Pew Research did a survey recently where they asked Americans about their thoughts on AI, and there was some interesting information that came out of it. Here's the key takeaways. About half of US adults now report using AI chatbots. Chatbots. We can see back in 2024, about 33% said they used them, 66% said they didn't. 2026, it's now 49-51. But Americans, including younger adults, are deeply skeptical of AI. More adults predict that AI will have a negative rather than positive impact on them and on society. So both things are true at the same time. More and more and more people are actually actively using AI, while also more and more people are actively saying they're skeptical that AI is actually going to benefit humanity as a whole. And I actually think that makes sense if you think about it, because we're sort of lumping all AI together as just like this one thing. There's just AI, right? But AI is like a whole bunch of things, right? You had video generators like Veo and Seedance and Grok Imagine. You have image generators like Nanobanana and ChatGPT Image. You've got coding tools like Codex and Claude Code. You've got music generators like Suno and Udio. You've got text-to-speech generators like ElevenLabs. And of course you've you've got chatbots like ChatGPT and Gemini and Claude. But then you do surveys like this and say, how do you feel about AI? Well, it's nuanced. Like you're allowed to think, okay, I really don't like AI-generated music because it's starting to pop up more and more on Spotify and I don't like hearing AI-generated stuff on Spotify, but ChatGPT is kind of cool and helps me do my work a little bit faster. So I like that. And every time I get on Instagram or YouTube, I see Shorts of like AI-generated slop videos and I'm getting sick of seeing All of those things can be true. You can like this part of AI that AI is doing, and you can hate this part of AI that is not making your life better. And so you go and do these surveys and yeah, more and more people are using AI 'cause ChatGPT genuinely does have helpful use cases for most people. It could make me do my work a little more productively. I could take a picture and say, what's this weird rash? And get an idea before going to a doctor, right? Like there's things that make it useful to like everybody, but then there's also parts of AI AI, like AI-generated slop music or AI-generated videos that are popping up on YouTube Shorts and getting millions of views that a lot of people are sick of and getting overwhelmed by and not wanting to see it. And there's, you know, fake images popping up that look like real images, like, oh, this building just got bombed in this foreign country, and you believe it's real for a minute, and then everybody starts telling you, oh, that's AI, and now you don't know what to believe. And there's things that positively impact people with— from AI, and there's things that negatively impact people with AI. And so I think going around and surveying and saying, do you like AI or do you not like AI? Well, that's kind of a loaded question because it's very nuanced and these surveys don't really like nuance. Anyway, rant over. Oh, and if you're curious which chatbot gets the most use, apparently ChatGPT still dominates with 44% of Americans claiming they use ChatGPT, which is interesting because it's easy to get into a bubble and hear everybody talking about Claude and think that everybody uses Claude. But I think in the general public, ChatGPT is still like the Kleenex or the Band-Aid of AI. Like they're synonyms to most people. And finally, I like to end on robots. And well, this robot is one that'll take your crap. It's a self-driving toilet. So this company, Xiaoban, will safely navigate your home, clean up after users, and empty itself all on its own. That's right, folks, we're getting self-driving toilets. So this toilet sits on a dock, and when you need it, it can be summoned over to your bedside and positioned exactly where you need it so that you can, uh, shift over to it and then do your business. It's got like a bidet feature to, you know, clean you up. It's self-cleaning. So the toilet bowl scrubs itself and cleans itself when it's done. And then it actually drives itself over to your toilet and then has a mechanism to empty what's in the toilet, the driving toilet, into your regular toilet. So it will dump that waste into your toilet. And then once it's done dumping the waste into your toilet, it then cleans itself and scrubs itself all out, searches the room again. And then— When it's It's all done with that, it goes and docks back up to charge and refills its water tank. So yeah, I mean, this is obviously designed for people with mobility issues that need the toilet to come to them. But come on, you know, some lazy rich people are gonna get this too. But that's what I got for you today. A lot of interesting stuff happening right now in the world of AI. I do record these videos on Thursdays and publish them on Fridays. So if any new news came out on Friday that I missed in this this video, well, it'll likely make next week's video. There is so much happening in the world of AI every single day. Like we are just getting flooded with news announcements constantly. And it's my goal with this channel to personally keep up with it every single day, drink from the fire hose, all be overwhelmed so you don't have to. And then every Friday I'll make a video where I break down all the news that I think is most important for the most amount of people to know. I'll try to cut through the noise and the hype and just give you the signal so that, well, you don't have to feel overwhelmed by this, 'cause trust me, it's really, really easy to get overwhelmed. That's my goal is to help you out and prevent that. If that's something that interests you, consider liking this video and subscribing to this channel, and I'll make sure more videos like this show up in your YouTube feed. But again, that's what I got for you today. I really, really appreciate you hanging out with me, nerding out with me. I'm having so much fun keeping up with all of this stuff and, you know, testing it myself and then turning around and figuring out what's worth talking about. And hopefully you learned something and it's helpful to you. Really, really appreciate you. Hopefully I'll see you in the next one. Bye-bye. What's up everyone? Glad you could make it.