AI News: The AI Stories Everyone's Freaking Out About

There's been some drama and some shakeups in the AI world this week, and I want to break that down as well as all the new models and various things that we can do now that we couldn't do before. So let's get right into it, starting with the fact that we got a few new video models this week, one out of ByteDance and also one out of Black Forest Labs. So let's start with the ByteDance video model called SeedDance 2.5. This new model can generate up to 30 seconds of video. You can also input up to 30 images, 10 video clips and 10 audio clips as reference materials, all in a single pass. It also offers timestamp level control for targeted editing of audio and video content. So you can really, really dial in exactly what you want the video to do and what you want it to be in your prompts. And for me, I'm just really blown away by how quickly we got this good with video generation. Now there is audio on these videos. I'm just muting it so I can talk over it, but video generation has come a long way. A few years ago, I was trying to make Monkey on roller skates and it just looked like a little blur flying across the screen. Now we get videos that look like this and yeah, it's just absolutely wild. Now the obvious downside to all these new video models getting better and better is it's getting harder and harder to determine what's actually real and what's not anymore. And it's become a big sort of like AI slop problem that we're running into. Now this one to me still feels pretty AI looking and most of what they show in their demos, you can really tell are AI, but, uh, yeah, this, This stuff is getting better and better by the day. This is like the longest subway car ever. Now, if you do want to play with this new Seed Dance 2.5 model yourself, you could do that over at dromena.capcut.com. Now, I'm pretty sure it's going to start working in Runway and Leonardo and all the other platforms where you can generate video soon as well. But as of the recording of this video, this is the only place I know to use it. Keep in mind, it is not inexpensive to use. So I paid for an $18 a month plan, and that essentially got need like 3 video generations. So you're not gonna get a whole bunch of demos out of me because I'm not just gonna go spend hundreds of dollars to demo tons and tons of stuff. You could go look at the, you know, various previews they put up on their website. But the first prompt I gave it was a colorful underwater scene with alien unnatural sea life. A man in a scuba suit swims around to observe and take photos of these bizarre creatures. And here's what that looks like. You can see that it actually put an AI watermark up at the top here. It's pretty impressive looking. There is some weird coherence issues, like when we get towards the end here, he's clearly looking at a jellyfish from this angle, but then when it goes to this angle, it's no longer a jellyfish, it's like some sort of crystals or something. Yeah, I don't, I don't know why it changed like that. But you know, again, I'm just impressed by how far we've come. For my next prompt, I wanted to give it an image and see what it does with that. So I gave it just an image of myself and then said, "The room changes into a nightclub and 048 starts dancing." That's basically me just tagging this image. The image was called 048.jpg. And here's what that generated. Perfect. So again, you can play with it over at dramina.capcut.com. I mentioned we got two new video generators this week. The other one was a Flux 3 video. They teased this one a couple weeks ago and I talked about it on another video, but now it's actually accessible for anybody to use. And you can pretty much use this one everywhere. It's in Leonardo, it's in Runway, it's in all the platforms where you can generate videos. This one to me looks a little more AI, a little less realistic. You could see an example that they give here on their website. Clearly not fooling anybody into thinking that's a real video. Now this one can generate clips up to 20 seconds long, and they claim this one is a model designed to model reality as accurately as possible. So I think it might have a better understanding of physics or something like that. Although to me it still looks like very generative art. So I logged into my Runway account and tested the exact same prompts that I tested with the SeedDance 2.5 model— a colorful underwater scene, etc., etc. And again, not very realistic looking, but here's what it generated. So interesting that it actually generated multiple scenes. It wasn't all just one single static shot. You've got the scene of the diver swimming towards the jellyfish here, switches angles to like a lower angle, then it switches to it looking at this like alien-looking underwater monster, and then finally towards the end it changes angles to the diver swimming in this direction. So it's constantly switching angles throughout the video off of just one prompt. I also gave it the same image and the same nightclub dance prompt here. And well, here's what that generated. Probably closer to how I'd actually dance in real life, honestly. Now, one thing that's interesting about this one to keep an eye out for, if I scroll all the way down to the bottom of their blog post where it says what comes next, They are planning on making an open-weight variant. Now, whether or not this will be something you'll be able to run on like a local consumer GPU, not sure yet, but it looks like we will get one that you can run locally and fine-tune and do whatever you want with. So that'll be interesting as well. Now, I'm super impressed with how far they've come, but I'm also not a huge fan of sort of how prolific AI slop has become on the internet. I think there are some people out there that are using these AI video models like in really, really cool ways, generating like really cool trailers where they're putting together a whole bunch of scenes and, you know, prompting it hundreds of times until they get the exact shots that they want and telling these stories and bigger narratives and color grading and really putting together something that looks impressive. But I also feel like most people are using these things to like make quick sloppy TikTok videos as well. So that part of it, I'm not a huge fan of. It's something that as video models have gotten better and better. I've been fairly vocal about, like, I think it's really, really impressive that we can generate this stuff. But I was also always concerned with how people were going to use these, you know, the deepfakes, using them to generate things that never really happened and trying to convince people they are real. The reverse, where people are doing things in real life that aren't bad, but then potentially saying, no, that wasn't me, it's AI. Like, somebody could commit a crime now and it's an actually decent defense to be able to say that wasn't real, it was AI-generated. That kind of stuff is very worrisome to me. And I don't think simply watermarking is always going to be a great solution because people are always going to figure ways around that. But anyway, again, I'm really impressed with the technology. I love that these things exist. It's my goal to make people very informed and aware that this stuff exists. So if you're out there on the internet and you're seeing things that you're like, wow, that's really, really impressive, but I can't tell if it's real or not, I almost feel like you need to start with the assumption these days that it's probably not real until proven otherwise. Like, it's kind of this weird shifting of the way we need to look at videos online now. But anyway, yeah, we got two new video models this week, and both are scarily impressive. Next up, this new release is a device that's meant to help you never forget anything. It's called Second Brain Note from Genspark, and it's basically a super thin AI recording card that's about the size of a credit card and snaps onto the back of your phone like a MagSafe wallet. It's under 3mm thick, has card slots built in, a 35-hour battery, and with one button you can start capturing a meeting, a call, or even just a random idea you don't want to lose. Once the recording syncs to Second Brain, it turns that conversation into organized notes, pulls out the important details, and connects it with the rest of your work context across email, calendar, Slack, Notion, Google Workspace, HubSpot, and more. So instead of digging through meeting notes later, I can ask What did we decide about short-form strategy? And then have Genspark Super Agent turn my recent project conversations into a full slide deck from the strategies we discussed. And the longer you use it, the smarter it gets. Second Brain Note is launching as a limited first release, and you can get 10% off through my link in the description. Free users get 300 minutes a month, while Genspark Plus and Pro members get up to 24 hours of AI transcription and summaries per day. You can check it out at the link in the description, and thanks to Genspark for supporting my channel. Channel and sponsoring this portion of today's video. We did get some new large language models this week as well. Probably the one that's been the most talked about is Qwin 3.8, and I actually talked about this one a couple weeks ago back when they first announced it on July 19th, saying that essentially it's coming soon. Well, it finally came this week, and just as a reminder, it's a 2.4 trillion parameter model. And in their post from a couple weeks ago, they said it's an open weight model that's second only to Fable-5. Well, now that it's released, we can see how close to Fable-5 it actually is. They posted their benchmarks here and yeah, it's a really, really solid model for open weights. On their page they show it building like a corporate compliance knowledge graph. We can see what that looks like here. I'm a UI designer, build a UI prototype for me. You could see it built some decent looking UIs and I mean, yeah, it's developing stuff at the level that you might expect out of one of the newer Opus models or maybe the GPT-5.6 models. And again, it is open weight. Now when they said second to only Fable, they were clearly kind of cherry-picking which benchmarks they were referring to. When it comes to software engineering, the benchmark that most people are paying attention to now is this DeepSWE. Models that are reaching high numbers on this one are usually the models that people are most impressed with when they actually have it start coding for them. And we can see on this DeepSWE, QYNH 3.8 Max scored a respectable 56.6. However, it still came in under Fable at 70, GPT 5.6 Soul at 73, even Opus 4.8 at 59. Now it's a big improvement over their previous generation, Quinn 3.7 Max, from 21.6 to 56.6. But I mean, it's still not really at the same level as the Fables and the 5.6 Souls of the world. So if you're looking for a state-of-the-art model that's going to be great at coding, you're probably not going to see similar results to Fable and 5.6 Soul from this model. However, if you're looking for a model that's just really good at answering questions and thinking through problems, it seems to be pretty good at that on GPQA. Google-proof question and answers, it got a 92.6, which is on par with Fable and just slightly below GPT-5.6 SOL. Now, as always, I will link to the documentation in the description below this video if you want to check out the full benchmarks here. As far as open weight models go, it does appear to be a pretty good model. Saying that, if you think you're going to install this locally on your computer and run it outside of the cloud, well, it is a 2.4 trillion parameter model. Like, it's not going to run on any consumer upgrade GPU anytime soon. You can play around with this model right now if you'd like over on Quinn's website. Head over to chat.quinn.ai and log in, and you can play with 3.8 Max right here on their site. Now again, for me personally, if I'm doing like a complex coding project, I'm probably gonna go use one of the GPT-5.6 models or maybe Fable-5. If I'm using an open weight model, it's because I want to run something locally in most cases, in which case I'm going to need to use a much smaller model like Google Gemma 4 or one of the GPT-OSS models, which can be run locally for simpler tasks. So this model kind of falls in this area for me where it's something I can't run locally because it's just too big and too powerful for that. And if I have to run it in a cloud anyway, I'm just gonna go use the best of the best models, which is gonna be out of Anthropic or OpenAI right now. But I know you're probably curious how Qwin 3.8 looks on BucyBench, and well, you know, I got you covered. Here's Gary Bucy as far as Quinn 3.8 Max is concerned. A couple things that are interesting to keep in mind: it cost me $0.21 to generate this, it used 35,990 tokens, and took over 10 minutes to generate this SVG. But it's not horrible. It doesn't have some of the same problems that some of the other newer models have. And while we're talking about newer models, we also got a new model out of Meta this week, as well as a new coding harness from them. So this week, Mark Zuckerberg took to X to announce that they're releasing Muse Code in beta today. It's their terminal coding agent, their CLI, similar to like the old Claude Code CLI, the Codex CLI, Groq CLI. Basically, you run it in the terminal and it'll write code for you. And along with this harness, apparently they released a new model called MuseSpark 1.2, but it's a code-focused model, so it's not just like a complete general-purpose model from my understanding. Now, when we look at the benchmarks here on this one, DeepSuite again is the one that I feel like if you're doing any sort of software development or having it you know, vibe code stuff for you. This is the one where you're probably gonna notice the difference the most. It falls in at 59.3%, just below 5.6 Terra and Opus 5. But notice 5.6 SOL and Fable models aren't even mentioned on here, so not quite at the same level as those models. Here's a game example that he posted along with his tweet of a game that was developed using this model, and I mean, it looks pretty decent. And here's a website design that it made as well, which looks, you know, decently designed as well. It looks very similar to what I've gotten out of the Grok and the GPT models as well. They all seem to have like a similar aesthetic when you ask it to go make like an interestingly designed interactive website. Now, because MuseSpark 1.2 is a model for coding, well, it should be able to do SVGs. So of course I ran it through BuseyBench. You probably already saw the teaser of this earlier, but here's what MuseSpark 1.2 thinks that Gary Busey looks like when generated with code. Taking a closer look here, this one took only 35 seconds to generate, cost only 4.5 cents, and used 10,597 tokens to generate. So while Meta may not have the most state-of-the-art coding model, I do feel like it's a pretty solid, inexpensive alternative to the actual state-of-the-art models out of companies like OpenAI and Anthropic. And since we're talking about Meta, I have to bring this up really quick because this has been absolutely fascinating to watch. It seems like every company is racing to see which company is the best at committing cyber crimes. So this is according to the information here. A Meta AI model hacked another company during cybersecurity testing. And if you're wondering why I think this is so funny, maybe you haven't been paying attention recently to all of the other claims that have been coming out of some of these companies. I talked about this a couple weeks ago where OpenAI hacked Hugging Face. Supposedly their model was inside of a sandbox environment, but it figured out how to break out of the sandbox environment, get access to the open internet, hack Hugging Face, steal the answer key to like a benchmark it was being tested on, and cheat on the test. And it was this like major cybersecurity incident. You give this OpenAI model a goal and it will do whatever it has to, including committing cybersecurity crimes to reach its outcome. OpenAI released this article breaking down everything that happened, and I talked about how it's interesting because it's very scary that these models will go to any length length to achieve their goal. But also, it did feel a little bit like OpenAI was using this as an opportunity to market how good and how smart and how impressive their models were. Well, not to be outdone, a few days later Anthropic was like, our models commit crimes too, you know, right? So on July 30th, Anthropic released an article investigating 3 real-world incidents in our cybersecurity evaluations. In a review of our cybersecurity evaluation transcripts, we found 3 incidents in which Claude model reached the internet from within or while interacting with third-party evaluation environment and then gained unauthorized access to real systems of 3 different organizations. So it accidentally got access to the internet and hacked into other websites and essentially committed cybersecurity crimes as well. And now, of course, as Meta's trying to get back into the large language model story here and release new models, they're like, yeah, yeah, yeah, us too, us too. Our model also broke containment. This was the previous generation, Meta's Muse Spark 1.1, not this newer coding model, but it appears to be a very, very similar story. Story here. It was in a sandbox environment, but the AI was able to access the public internet because of an error. The model subsequently exploited a security vulnerability in another third-party service in a manner similar to previously reported instances with other companies. This has led to one of my favorite memes I've seen in a while with Sundar from Google, you know, saying, come on, commit crimes. But you have to admit, it's kind of interesting to watch all of these companies sort of race to put out press releases about the cybercrimes that their AI models are committing as a way to show how powerful they are and I guess how valuable they are. And then also possibly for like the whole regulatory capture reasons. They obviously want the government to get involved, but they want themselves to be in the mix when the government gets involved, assumingly. Anywho, I mentioned earlier that there was some shakeups in the AI world as well, a little bit of drama and a little bit of company restructuring, let's call it. So let's talk about those real quick. First, starting with the drama coming from someone I never expected to see the drama from, which was Hank Green. Now Hank Green's one of my favorite YouTubers. I love his videos. I pretty much watch every Hank Green video that comes out. I'm a fan. Well, he's in some hot water right now because he got caught using AI, I guess. So here's the quick TL;DR. Hank does a show called Ask Hank Anything where people literally ask Hank a bunch of random questions. They submit the questions beforehand and then he researches them, and then on the show he answers the questions. And then if any questions pop up during the show that he doesn't know the answer to, he'll research it and then come back and do like a Future Hank thing where he then splices those questions in. Well, in one of those segments where Future Hank was talking, he said the words, "I appreciate the pushback." That's it. That was what started this whole thing. He literally said, "Oh, I appreciate the pushback." and then went on to give his response. And people said, "I appreciate the pushback" is something AI says. Are you using AI to write your scripts? That's essentially how it started. Hank went to X and basically said, "I produce under a ton of pressure. I did use ChatGPT for research on the script, but I do want to note that the script didn't say 'I appreciate the pushback.' That was an ad-libbed line." He then deleted this X post and then went to Reddit and wrote a much longer essay about his thoughts on his whole process with AI here. Now, I'm not going to read the whole thing. Of course, I'll link it up below if you want to. But basically what he said was, I don't use AI to write my scripts, but I do use AI to find research papers. I still read the research papers myself, but AI helps me find and uncover research papers that I might have otherwise missed. So it saves me some time in doing the digging. However, in this same post, he does go on to say, I need to come to terms with the fact that the level of dopamine I've been getting from interacting with LLMs, with doing more and more and more and more, is not healthy for me or good for the world. It is careless and has disconnected me from where people are on this. And I mean, this whole thing has just created a ton of controversy around the internet about Hank Green actually uses AI and how horrible it is that he uses AI and that he's probably lying and does use AI to help with his scripts. And people essentially just trying to cancel Hank Green for this. Okay, but here's my take on the whole thing. I actually to understand where people are coming from on this, believe it or not. So Hank Green is a science educator. He is a seemingly very, very intelligent person, and people come to his channel because they're expecting Hank's take, Hank's opinions, Hank's commentary. And if all of a sudden one day you have to question whether what you're hearing actually came from Hank's mind or came as a result of like an AI response, now you're not sure if these are Hank's real thoughts or the thoughts of an LLM that the LLM them fed into Hank's brain and then he recited. Even if it wasn't AI writing the script for him, if he's reading LLMs a lot, the responses of the LLMs are still getting into his head. So even if it's not the script, it still could be him feeding back to you what an LLM gave him, even if unknowingly and not, you know, from a script. So I totally get why people are sort of upset about this. The person they thought they were getting commentary and original thoughts from, if that is a thing may have been thoughts fed to him by an LLM. Saying all that, Hank's response was, look, I use this for research. It saves me a lot of time on research. The people that are getting mad that he's using this for research, it makes no sense to me because he's like, in his words, using these AI models almost as like a glorified, better search engine. So I don't think Hank did anything wrong. In fact, I don't think there's anything wrong with using AI models. But even if I found out that like Hank was using AI to help with some of his scripts, I actually don't think it would bother me. Me that much because I can actually see the use case of me like brain dumping a whole bunch of thoughts that are my thoughts from my brain into a LLM and then saying, help me organize these thoughts and then say 'em in a better way than this ramble that I just gave you. And it's still the things that are coming out of my brain, just AI helping me sort of organize and arrange them. And if I learned that Hank was doing that, I would think nothing less of Hank for that, assuming these were his thoughts and not him just regurgitating what an LLM gave him. But I think the bigger story story here is that this is bringing into the public consciousness creators are using AI to assist them in what they're doing online. And the watchers of those creators are having to come to grips with the fact that some creators are using this stuff to assist them. And some people are finding, I, I don't really like that. I don't appreciate when other creators use LLMs. I think the thing that bugs me the most is probably the amount of judgment that other people put on creators. And this is also the reason this story has probably impacted me personally so much is because I am a creator who also deals in AI. So like, I'm seeing this overlap, but the amount of judgment people have on creators is tough because creators are like real people too. They have lives, they have families, they often have careers and other commitments outside of what you're just seeing on video. And they're also looking for ways to sort of optimize and improve their workflows. And every creator has a sort of line that they're willing to cross as far as how much they're willing to automate without sort of giving up the amount of them that still goes into the process. And it's something that I personally struggle with all the time of like, how much should I let AI do? And where is that point where it's no longer my creative process, my voice, my brain, my input on the product anymore? So for me, I found that line as I let AI help me with my research. I let AI help me with outlines and essentially what I talked about, brain dumping stuff and then letting it reorganize the brain dump so that I say it in a logical order instead of just random, you word vomiting, but I don't let AI edit my videos for me because it's not good enough yet. I don't let AI write scripts for me. I don't let AI decide what I'm going to talk about. I come up with what I'm gonna talk about and let it sort of help me outline it. I don't let AI make my thumbnails for me. I don't let AI write my titles for me. I don't let AI decide which news stories to talk about in these videos. There's a lot that I'm still not willing to give up to AI, but there are things I am willing to give to AI because it just makes my life a little bit easier. Here and takes a little bit of pressure off. And I don't think a lot of people give other creators enough grace for trying to figure out new ways to optimize and sort of streamline their processes, because everybody's looking to do that. That doesn't mean you're lazy or you're trying to take shortcuts. It means you're just trying to optimize. You're trying to work smarter, not harder. And I just, again, I just don't think a lot of people give creators like Hank enough grace on stuff like that. And like I mentioned, everybody has a line, and I feel like Hank saw where the line was and then realized his own line that he created, he probably crossed it a little bit. But I also think a lot of the internet's blowing it way out of proportion. Anyway, I'll stop rambling about that. It was just a topic that sort of hit me because I do fall in that intersection of creator and AI. There was also another shakeup this week over at Google, and it's a pretty big one. So Demis Hassabis, who was the CEO of Google DeepMind, decided to step down as CEO of DeepMind and become chief scientist for Alphabet. Now, Now the previous chief scientist was a guy named Jeff Dean, who also this week decided to leave Google and start his own company. Now if you're not familiar with Jeff Dean, he was actually employee number 30 at Google. He's worked at Google for 27 years. He was there long before Demis was even there. So just like a quick history lesson on like Google and like their AI division, right? So Google had a division that was working on AI called Google Brain, and Jeff Dean was sort of like the man who was running Google Brain. And then somewhere around, I think, 2014, Google acquired DeepMind and brought DeepMind in-house as an Alphabet company. So now Google had two AI divisions. They had Google Brain and they had Google DeepMind. Jeff Dean was one of the guys running Google Brain. Demis Mustafa and Shane Legg were the guys running DeepMind. In 2023, Sundar and the Google team decided to just merge the two together and just have it all be Google DeepMind. So Google Brain got absorbed into Google DeepMind, and now it was just Google DeepMind. And at the time, Demis was made CEO of that combined DeepMind thing. Jeff Dean was chief AI scientist. Now fast forward again to this week. Demis is saying, I no longer am CEO. Kéré, the ex-CTO of DeepMind, is going to step up and basically be running DeepMind. Demis is going to be an advisor for DeepMind, but now he's going to be more of the science role. And Jeff Dean, he's just out. Jeff Dean moved on to start a company called Discovery Loop, which is going to focus a lot more on creating automated machine learning research for scientific discoveries. Now there's sort of two camps going on here about what's actually going on. Some people are saying this is a sign that Google is struggling and not going in the right direction, because as boy input here, would you leave a company where you're one of the head honchos, a company where you have been instrumental in building it into what it became came after spending all of your professional career there if you were confident they are on the precipice of unlocking the intelligence explosion? And I think that's a pretty good point. Would you leave if you felt like they were this close to like cracking AGI? But I actually think personally, and this is just pure speculation, that the story's a little bit different. I think both Jeff Dean and Demis are science people. They like science, they like research, they like to be in the weeds solving big problems, figuring out how to make the world a better place through their research. That's where those guys' brains live. There's also been a lot of talk about how Jeff Dean wasn't a fan of Google's contracts with the military, and I don't think Jeff or Demis are big fans of like the productization race of AI. I don't think either of these people are people that want to be what ChatGPT or Anthropic is, trying to make the best models as fast as possible and get them in consumers' hands. Like, I don't think that's where their brains live. Their brains live in solving bigger problems, curing diseases, solving climate change, solving energy issues, things like that. They want to be working on the bigger problems. But Google has found themselves in this race with Anthropic and OpenAI, who are both racing to get their products into more and more hands, get more and more active users using their tools. And now Google is in that mix going, we need to be there as well, getting more and more people using the Google AI products instead of OpenAI and Anthropic's products. And I don't think that's the game Demis or Dean want to be playing. So I think Demis is stepping into that science role so he can focus on science. I think Jeff Dean is stepping away so he can focus on building a company that's solving bigger problems. And I think Sundar and Karay at DeepMind, they're shifting their brains into how do we go and compete with OpenAI and Anthropic and make products that people really, really want and that can compete in this model race that's going on. That's what I think the story is. But again, this is all just speculation. I have no insider information other than the fact that I've had conversations with Demis and Corey, and you know, I've also been following along to the whole AI storyline for the past 5 and a half, 6 years. So, you know, just based on that, I think that's where these guys are sort of, uh, headed with all of this. And I think it's less drama than people want to put on it and more just like the science guys want to go focus on the science Let's let product people go and focus on productizing it all again. I could be wrong. All right, I've covered a lot, gone down a handful of rabbit holes. There's still a few more things I want to talk about that happened this week, but they're all quicker things. So let's jump into a rapid fire. Pew pew pew pew pew! Let's start with the Google Earth feature that I talked about at last Friday's news video. Well, that was rolled back like, I don't know, 24 hours later? So basically there was a feature where you can open Google Google Earth. It had Nanobanana built straight into Google Earth, and you can have it sort of reimagine what was going on in that shot in Google Earth, like Pompeii reimagined with Nanobanana. So it tried to bring it back to what it may have looked like back, you know, whenever that was. Well, people started using that to create a lot of deepfakes, like making images of what would it look like if there was a terrorist attack here, and what would it look like if a plane was crashing into a building in this image, and starting to just spread misinformation like immediately when this launched. And Google saw that and went, okay, maybe this was not a good idea, and they pulled it. So that feature did not last very long. I did get a chance to demo it in last week's video, but probably by the time you saw the video, it was already gone. While we're talking about Google, Google Maps got some new features with their Ask Maps. You could now order food, find the perfect hotel, and discover events. So Ask Maps can now handle complex multi-step tasks for you using new agentic capabilities. You can literally ask Maps to order spicy pad kee mao with seafood for me to pick up on my way home. And it'll find the restaurants along your route that serve what you're craving, and then could even consider things like your saved places or specific dietary needs. And then it can automatically add the dish you're looking for directly to your cart so you can easily review and complete your order. It could find the latest information about hotels for next week's conference in downtown Miami. Find me a decently priced, top-rated hotel with an artsy vibe, blah blah blah. And it'll give you personalized answers. You know, it can connect to Gmail and Google Calendar and things like that, and it can even resume past conversations for you. It's also got real-time transit information and you can talk to it conversationally. So now your Google Maps is going to be pretty much completely filled with AI options and you can just chat with Maps. Now we got a handful of small announcements out of OpenAI as well. GPT-5.6 Sol is now more reliable with facts and it can provide more focused answers. But probably even more interestingly, if you're a free user of ChatGPT, the default model is now GPT-5.6 0.6 Luna and you now get unlimited text chats. So free ChatGPT users can now use it unlimitedly. OpenAI also released this post called 10 Advances in Mathematics and Theoretical Computer Science, essentially showing that it's solving harder and harder problems now and are much closer to like literally cracking the code on like some of the world's biggest problems. Now these are the 10 problems that it solved here. You can feel free to pause and read them if you'd like. Now, these problems were solved by their Astra, which I think is the code name of like their next generation of model that we haven't seen yet. But the implication here is that if it can solve these problems, what's to stop it from making new drug discoveries and figuring out solutions to much bigger pressing world problems? OpenAI also released new plugins specifically designed to guide K-12 teachers, college educators, and college students. Things like Assignment Translators, which will translate the assignments into other languages, exit ticket briefs, updates for families, practice tests, interactive teaching sites, course calendars, course materials, course posters, interactive learning sites, flashcards, study plans. So all sorts of cool stuff for students and teachers to take advantage of inside of these new educational-based plugins. And then finally, let's end with gadgets. Now, this is a little bit speculation, but apparently Johnny Ive's first OpenAI gadget is a hockey puck-sized smart speaker. So what we're hearing right now is that it's going to be a donut-shaped device. It's going to cost more than $300, and you're going to talk to it like you talk to, like, voice mode inside of the ChatGPT app. So they're trying to make a Jarvis donut, I think, right? Like, that's what you would call it. It's a Jarvis donut. Anyway, that's what I got for you today. It's been a busy week. I mean, I say that every week. It's always a busy week. There's so much happening every week in the world of AI. As always, it's my goal to sort of drink from the fire hose and read it all consume it all, keep up with it, and then turn around and keep you up to date so you don't have to try to keep up with it every single day. If you want to stay tapped in on everything that's happening in the world of AI that's worth knowing about, and you want to do it with just one video a week, that's what these Friday videos are for. So if you like that kind of thing, maybe consider liking this video and subscribing to this channel. That'll make sure more videos like this one show up in your YouTube feed. Now, as a quick reminder, I do record these on Thursdays to publish them Friday morning. So if any news came out like late Thursday night or on Friday, it usually makes next week's video. But I promise to always keep you looped in on the most important things going on in the world of AI that I think are actually worth your attention. So again, if that's something that might interest you, consider liking and subscribing. I also do like tutorials and breakdowns and commentary and stuff like that on what's going on in the AI world in between these news videos. So you might find those cool as well. I'm getting really close to a million subscribers, so if you haven't pressed that subscribe button, it would really help a lot. Just You know, 2 seconds, less than 2 seconds. Okay, I'll stop begging now. Thank you so much for tuning in and hanging out and nerding out with me. I have a blast keeping up with all of this and hopefully I'll see you tuning in for the next one. All right, peace out. I love you. Bye-bye.