AI News: OpenAI Pauses, AI Cancer Vaccine, and Qwen3.8

OpenAI just paused development on new models. AI just helped with the potential cure for skin cancer. We got a brand new open source model that you can run locally on your computer that's actually pretty good. We're going to talk about all that and more in today's AI news breakdown. But first, I want to start with the thing that I've had the most fun with this week, and that's this new TripoP 2.0 model. This is a model that actually generates 3D objects that you can export and 3D print. You can pull 'em into Blender, you can pull 'em into Unreal Engine, and they're actually really, really good, at least at rendering the 3D portion. There is a little bit of wonkiness, which I'll talk about in a second, but you can see that this is generating borderline game-ready assets like right outta the box. So of course I had to play around with it and I wanted to give it an image of myself and see if it would generate a 3D version of me. So I gave it this image here 'cause this was kind of the Closest image that I had to like a T-pose. And you can see this is what it generated right off the bat. It removed any sort of facial hair or, you know, any hair at all, but it got the pose pretty spot on. Now, after generating the 3D mesh, there's an option to generate a skin that goes over the mesh, and that's where things got a little weird. So here's the textured version, and to be honest, it nailed the shirt, pants, and shoes, and my skin tone too. Like, if you look at the picture here, those are perfect. But what What the heck did it do to my face? I don't know what's going on with the lips. If you look at this, this was the lips it generated on the original 3D mesh here, but then when it added the texture, it gave me like extra clown lips or something. I don't know what's going on. And then you can also see the nose got a little bit weird. There's some skin in my sunglasses. So yeah, the face got a little bit wonky, but everything from the face down looks like perfect. Like, yeah, if you just see this version without My head in it, it looks like identical to our, uh, original photo here. But as soon as that face comes into the picture, yeah, that's not quite me. But I wanted to try with something a little bit more fictional and fantasy to see what that would do. And we'll actually use an image generator of your choice. I use Nano Banana and I give it the prompt, a humanoid wolf with horns and wings wearing torn jeans and a black t-shirt flexing and standing on a tree branch. You can see my first I didn't put a t-shirt and jeans, and, uh, I got a little extra fur in areas that looked a little too weird for me. So with my second prompt, I had it add some clothes. I thought this second one came out the best here, so I generated a 3D image from this one. And here's the original mesh that it gave me, and it looks pretty good. And then when I had it add texture, then it got really cool. Like, it really, really nailed it with this one. So it seems to be quite a bit better for like fantasy creatures where AI generated the image versus giving it real images. to work with, because this is really good, like some really decent detail in there. It's really making me want to create some 3D objects and then pull them into Codex and have it generate a game using these 3D objects. That'll probably be a project for a future video, but I was really, really impressed with this. I thought this looked really, really good and got me inspired to want to do more game-type assets. And as of this recording, if you want to play around with Tripo, they actually allow you to generate 2 for free. So I was actually just generating these on the actual free plan. These Those are my 2 free generations. Another cool thing about this is you can generate an unlit version. You can generate a cartoon style, which just kind of puts like an extra black outline around it, a sketch style. And I really like this hologram style. Like if I could get this to look a little bit more like me, I can actually see using holograms like this inside of my videos. You know, it would be really cool to like pull out my phone and then have a hologram like this, like animate out of my phone and the hologram talks to you for a minute or something like that. I think that'd be really kind of fun to figure out how to do a video. I just need it to look a little bit more like me before I feel comfortable doing that. One thing that I find really interesting, if I go to my wolf with wings here and I click on unlit, it actually removes the wings completely. So the unlit version just has the bones that were in the wings, which is kind of cool too. It gives me like more effects that I could use if it wasn't a game. Maybe if you're battling this character in the game and you hit it a couple of times, you destroy the wings and now he can't fly anymore, but the wings look like this. Cartoon style again has that more like outlined look. Here's sketch style on the wolf. And then of course the hologram style where you could see through everything. Once again, also removes the wings. If you are building a game around something like this, you can export the hologram style. Maybe when you kill the character, it turns into like a ghost like this and floats up or something like that. I've been having a lot of fun with this tool. So just like last week, I wanted to kick this video off with the thing that I enjoyed playing with the most. Next up, I want to share some really exciting positive news out of the world There's so much like doom and gloom around AI narratives right now that it's really fun to finally share some news that I think people can get excited about. We've been told forever now that AI is going to help cure cancers and diseases and things like that. And well, this news came out this week that an mRNA cancer vaccine succeeded in phase 3 melanoma trials. The phase 3 trial included 1,137 patients who had stage 2B to 4 melanoma that had been surgically removed prior to the treatment in the trial. Apparently the vaccine extended the amount of time a patient has without the cancer returning. Now, this specific article here on Ars Technica does not mention AI at all in the article, and I think part of the reason for that is it's just like out of fashion to talk about AI now. Most of the narratives that you hear online that use AI are usually narratives around how AI is ruining everything. However, this vaccine does use AI. You can see here that it is called mRNA-4157. Well, I found this article here From December 2023, back when it was cool to talk about AI, called Advancing the Fight Against Cancer Through mRNA and AI. And if we look at the actual drug they're talking about here, mRNA-4157, so the same mRNA-4157 they're talking about in this new article. Back in 2023, they were on Phase 2B. This one is reporting about Phase 3 trials. Now, to be clear, this isn't referring to generative AI. This is not like a large language model that went and found a cure for this type of cancer, but it is using AI. It's using machine learning and the neural networks and everything to help fight against this type of cancer. Again, this was at a time where it was cool to say things were solved with AI. It says on the design end, a series of fully integrated AI algorithms take next-generation sequencing data from tumor and blood samples, reviews their genetic mutations, and predicts up to 34 of those neoantigens that are most likely to elicit an immune response. This algorithm has the potential to learn over time through pairing clinical and immunogenicity data and will hopefully become better at selecting the most clinically active neoantigens. So it's still using like a prediction machine, not too unlike a large language model of today. But I've been seeing a lot of arguments and discourse around this specific news online where people are claiming that this didn't use AI simply because that original article never mentioned AI. Well, in fact, on Moderna's own website, they literally break down how it's used AI to get to where it is. Anyway, this is the kind of news that I'm hoping we see more and more of out of the world of AI. And I do think generative AI and large language models are also going to be getting us closer and closer to these like health and science breakthroughs. We're already starting to see it a little bit, but when it comes to AI, I think we can all use some like positive use cases and positive news around how AI is being used. The next update comes from HyperAgent, and they built their entire platform specifically around this concept of of agent teams. Instead of asking one AI to do everything, you create specialized agents with different jobs that can work at the same time and with each other. And it's not just multi-agent, it's multi-human too, because apps like OpenClaw, Hermes, or even Claude Code can be difficult to work collaboratively with if they're living in one person's local setup. But with HyperAgent, you build an agent once and your entire team can use it, either directly inside the HyperAgent app or by messaging with it like you would a coworker in Slack. I personally used HyperAgent to build my own B-roll generator for videos. You just give it a section of a script and it turns that into photorealistic B-roll, motion graphics, and other visuals I can actually use in my videos. And because it lives inside HyperAgent, other people on my team could use the exact same workflow without rebuilding it from scratch. You can actually try my B-roll generator for yourself along with a bunch of other cool skills in HyperAgent's marketplace, or build your own team of AI agents with HyperAgent. If you sign up with the link in the description, you'll get $100 in in bonus credits on a paid plan. So go check it out, and thanks to HyperAgent for supporting my channel and sponsoring this portion of today's video. We did get another new large language model this week that everybody is talking about, and this one comes out of Alibaba. So this week we got the new Qwin 3.8 27B model. Now, if you have been following along, you might think, oh, I've already heard about Qwin 3.8. Well, last week I did mention Qwin 3.8 2.4 T, which is a 2.4 trillion parameter model. And yes, that was an open weight model, but it's so big that nobody's running that one locally. This new version, Quin 3.8 27B, is a 27 billion parameter variant that you actually can run on consumer-grade computers, which is why there has been so much talk about this one this week, is because A, it's actually a really, really good model that works really well for people, but B, it's small enough to actually run locally. To sort of hammer this point home, Hater here wrote this post. Let this sink in. Quinn 3.827B is an open weight model you can run completely offline on a 24 to 32 gigabyte laptop, and it's already better than Opus 4.8, GPT-5.6 Terra, and Luna. And he shared the screenshot to the Artificial Analysis Agent Index here, and we can see that Quinn 3.827B comes in above 5.6 Terra and only slightly behind Kimmy K3, which is another open model, but a model that is just also very, very big and likely not going to be running on your local Now, to be fair, 24 to 32 gigabytes of VRAM, most people still aren't gonna have that. If you have a newer, newer computer, maybe, or if you're using one of the Macs with one of the M chips that has a large unified memory like the Mac Studios, you're probably gonna be able to run it on one of those as well. But if you're on like an Nvidia graphics card, your GPU is gonna need to end in a 90, right? 5090, 4090, 3090, one of those are gonna be the only ones really strong enough to run a model like this. Apparently you can run it on a 16GB, but it's going to be insanely slow. Now looking at artificial analysis here and scrolling down to the intelligence index, not the agentic index we're looking at earlier, but just the overall intelligence index, it sort of falls middle of the pack here, scoring a 52. Now the intelligence index is a whole bunch of different benchmarks combined to sort of determine the smartest model across a bunch of different benchmarks. So it's one of my favorite Benchmarks to look at because you don't have to look at a whole bunch of individual benchmarks. It's like it took all of those, weighted them, and decided an intelligence score based on all of the benchmarks. And I mean, a 52 is really not that bad, right? It falls just below GPT-5.6 Luna Max, DeepSeek V4, Gemini 3.7 Flash. And I mean, it's not even that big of a jump to get up to these state-of-the-art Frontier Lab models. And if we look at everything in front of Quinn 3.8 here, none of these are going to run on a local consumer GPU, not a single one of these models. is small enough to run on even like a 5090. So pretty, pretty impressive model for being open weight and ability to run locally. If you do want to run this model locally and you have a GPU that can handle it, the easiest way that I've found is to download LM Studio here. It's a totally free app, and this allows you to run local models. With LM Studio installed, you could come over here to the model search, search for Quin 3.8, and this is the official version straight from Quin on Hugging Face. And they have different versions. You can get 4-bit, 5-bit, 6-bit, it in 8-bit quantization. Now that just means like the amount of compression essentially on it. The more quantized it is, you're going to lose a little bit of intelligence essentially. So you want to run the best one that your computer will let you run, the, the highest bit. The nice thing about LM Studio here is that if you click on any one of these models or variations, it'll tell you if it'll run on your computer. Full GPU offload possible. So on mine, I can run the full 16-bit version unquantized if I want, and it will still work. But if you can't on that version, I would grab the highest quantization number that you can, that your GPU can offload. There's also an uncensored variant that you can find inside of LM Studio, and I actually installed this version as well. You can see I'm using the 8-bit uncensored version, and when I say it's uncensored, it is very uncensored. I asked it to give me directions on how to do some pretty wild stuff that no other model is going to tell you how to do, and it told me with no problem. So yeah, Uncensored means it's very, very uncensored, so beware. Now, once you have that model downloaded, you can click on My Models. You can see I've got the uncensored variant and a 4-bit variant installed here. So if I select uncensored, click on my little cog icon, and then click Use in New Chat, you can see it then pulls in that model. And now I can use this model completely uncensored in chat, and it will think, it will Use code. It will pretty much do anything you expect new models to do. Now, this isn't like an agentic harness. LM Studio isn't going to do similar stuff to like Codex or Claude co-work, where you give it a task and then it sort of creates a checklist of things to work through, and then one at a time works through those things, giving the model new prompts to work through each of those things. That's what those agentic harnesses are doing when you're using large language models. LM Studio is just going to act like a single chat. You give it a message, it's going to respond, it's going to be If you do want to use this model as an agent, similar to what you'd get out of something like Codex, my preferred agentic harness at the moment is actually the Hermes harness here. And what's cool about Hermes is that you can actually get it to find models that are running in LM Studio. So because I have this model running in LM Studio, I can actually click on developer here, select the model I want to use. So right now I've got this Quinn 3.8 model, and then I can turn on the server here. And while this server's running, If I jump back over to Hermes, go to my settings, go to model at the top here, you can see that it actually finds LM Studio running and the models that are available inside of LM Studio. Right now I'm running the uncensored model, so I'd keep that selected, apply that. And I like this Hermes harness because to me it feels and looks similar to what you get out of Codex or Claude Cowork. So it feels very familiar. And now they just added a new feature where if you like something like GrokBot, they actually have a bots mode that works more like that where you can create little bots just like in GrokBot. And of course, I had to test this new model with my prompt of create a clone of Mega Bonk. And well, it tried. It took a long time. I think it ran for like 2 hours using my local GPU. But you can see that it kept on giving new prompts. It was writing code. It went through like a whole task list of things and worked through this project just like you would expect out of Codex or Claude Cowork or something like that. And as far as the Mega Bonk clone that I was working on, well, it made a screen that looks good. The problem is, as soon as I hit play, Yeah, nothing happens. I get nothing on the screen. I— yeah, it's just the game's broken. And it took so long to do this locally that I just went, yeah, I'm not gonna keep on working on this anymore. You can see it tried. Apparently I just leveled up. I don't know how because I can't see anything, but it didn't quite pass my Mega Bonk test. I did, however, throw it into BuseyBench. This was the image I got, and I don't know why this happened, but it used Used 70,000 tokens. It took 74 minutes to run and generate this, and it cost me 25 cents. Now, I didn't run this locally. The reason it cost me 25 cents was because I was using Open Router, so I was running this one specifically in the cloud. It would have been free, obviously, if I ran it on my own local computer. But 74 minutes to generate—like, that's wild. That's just—I don't know why. Anyway, the major idea here is that these local models. Are getting better and better, even though they're getting smaller and smaller in terms of like parameter count and actual file size and the amount of compute required to run them. So we're not far off from having models that are, you know, Opus 5, GPT-5.6, SOL, Fable level that we can run on high-end consumer GPUs. And that's really exciting. And speaking of frontier-level intelligence, we did get an update out of OpenAI this week on where they are in their development cycle. This week put out this article called Pacing Model Development in an Era of Cybercritical Capabilities. Now, they've been sort of teasing GPT-6 or Project Astra or, you know, whatever their next sort of iteration of GPT is, and it keeps on getting bumped out because of, well, the tale as old as time. It's just too scary and too powerful to put out right now. By the way, they also said that about GPT-2 back in 2019. Now, obviously these models have come a long way and they actually have gotten out of containment and hacked websites and things like that. So it's a A little bit different of an era, but still same narrative. They did say here there was a 2-week pause in reinforcement learning training on our latest models intended for deployment while we further hardened and red teamed the research. And their largest planned frontier reinforcement learning run remains on hold while we conduct smaller scale training and evaluation to assess model behavior. Yeah, right now the models are just kind of on hold. Now it sounds like the models are trained because reinforcement learning typically happens in post-training. So the full model is trained, Then they typically will do reinforcement learning to sort of tune it to the preferences and add guardrails and things like that. So it sounds like this Astra model, GPT-6 model, is pretty much like ready to go other than this reinforcement learning phase that really kind of dials it in essentially. When we'll actually get it, who knows? They're apparently holding things back and slowing things down because they really want to make sure they do this right. Right, and they don't want it breaking its sandbox and hacking more websites. Now, what we actually have been getting out of OpenAI lately is more feature updates and upgrades to their platforms like Codex. For example, this week they released a computer history feature. This to me sounds very similar to that Microsoft feature that everybody hated that they were rolling out on all the computers that basically watched everything you did on your computer and can recall back to it. Well, ChatGPT is doing something similar. Computer history turns your activity across apps and websites into memory. Memories and a timeline that ChatGPT and Codex can reference. You can ask natural questions about recent work, pick up where you left off, understand patterns and how you work, and turn repeated workflows into skills or automations. So you theoretically can go work all day, at the end of the day say, hey, review everything I did, is there anything you can automate for me? And let ChatGPT look through your history and figure out what it can automate for you. Now you do have to choose to turn it on, and it only starts after you've turned it on, so any past history won't be there. You can also control which apps and websites contribute. You You can pause collection from the macOS menu bar, and you can inspect or delete your history at any time. Why would anybody wanna do that? Here's a screenshot of what it looks like. Here's like a timeline. At 9:00 AM, you checked your task list, reviewed recent changes, planned what to tackle next. 9:10, reviewed your project feedback. You were in Chrome, ChatGPT, and Slack, it looks like. 9:20, prepared a launch update. Could see what apps are open, and it's suggesting skills. Turn the steps you use to prepare a launch update into a reusable skill. So it basically wants to watch your computer for both most likely training data but also so that it could figure out what you're doing on a repetitive basis and help it make it less repetitive for you. So if you're in the ChatGPT app, we go to our settings here down on the left sidebar. You've got computer history and you have to turn it on to allow it to start watching your computer history. And you can allow all apps or customize which apps you want it to be able to watch. Once you've done that, it will start watching your computer history from that point on. Another new feature they rolled out this week is is that it now integrates with Apple Messages. You can install an Apple Messages plugin inside of the ChatGPT app on your Mac and have it search messages, catch up on conversations, draft and even send replies for you, all within the ChatGPT app on your Mac. So to set this one up, you're going to open your ChatGPT app, come over to Plugins this time, search for Messages, and you can install the Messages app here. Now if I click Try Now, I can tag Messages and tell it to tell me about my most recent Conversations, which I'll likely be blurring. But let's see what it says. And yeah, although I have to blur it, it did tell me what my last five text messages were. But apparently can also like send messages and go back through old threads and things like that. All right, I know I covered a lot, but there was quite a bit more that happened. But I don't want to waste your time with all the details. I'm just going to give you the quick overview of it all. So let's jump into a rapid fire. Pew pew pew pew. Meta just launched a brand new Mac app, so you can pretty much do all of the various Meta AI stuff that you would do in one of the Meta platforms, but now in a dedicated app. It has an obnoxiously large dock icon, but when we open it up here, we can see that it has a very similar look and feel to something like Codex or Claude Cowork. A little less feature-rich at the moment, but I get the impression that they're going for kind of the same thing. We've got our scheduled tasks here, we've got artifacts, We've got media. You can see some of the stuff we've generated over time here. We've got our chat history. And of course, it's going to use the meta models to either respond to your questions or prompts or to generate images for you. Perplexity this week launched their Brain product, which is kind of like a second brain or knowledge base wiki thing, but it's built into their Perplexity computer. We can see it's got a sort of Obsidian-like knowledge graph here. It can organize and store the memories. So I'm pretty sure what they did was they took Andrej Karpathy's Wiki concept that he broke down on X a couple months ago and just built that concept straight into Perplexity Computer. And in fact, if you want to learn more about that and how I built my own of the similar thing, but I just built it myself instead of using Perplexity's version, check out this video that I put together called Build an AI Second Brain Knowledge Base. That'll get you there as well. Here's a feature that Google rolled out at the end of last week, but because I record these on Thursdays and publish them on Fridays, I didn't see this news until Friday. And well, they gave people the ability to remove watermarks on AI-generated images, video, and music. Now, they still have the invisible SynthID watermark, so people can still double-check to see if they were generated with AI, but they're now allowing you to physically turn off the watermark. And a lot of people aren't a fan of this idea because people believe it might just lead to more fake images and videos being spread around where people are pretending for them to be real. Now, again, you can grab them, throw them into Gemini, and ask if they're real or not, but still, how many people are actually doing that. Claude got a new feature this week. You can now send emails in Gmail and manage files inside of Google Drive. So you can ask Claude to reply to a thread and it drafts and sends the response. You control when it needs your approval. So you can connect Gmail, Google Drive. The interesting thing is I saw this and thought, I could have sworn Claude already did that. I mean, I've been able to do that inside of Codex. I don't know why I thought I was able to do it in Claude, but I guess now you can. OpenAI this week rolled out a feature for teens. So basically OpenAI is trying to determine whether you're an adult or a teenager by how you chat with it. Now, if you tell it you're a teenager, it just knows from that point on and will feed you the teen version of the model. However, if you don't ever say your age and you just chat with it and it thinks you chat like a teenager, well, it might downgrade you to the teen version that way as well. So if you're an adult that talks like a teenager, well, you could end up on the teen version. I actually did I did a livestream with my buddy Joe earlier this week where we went off on this giant rant about should teens even be using ChatGPT or not. Check out that livestream if you hadn't. It was a lot of fun. It is 3 hours, I know, but you can, you know, 2x it. There's a lot of good insights and tests and things that we did in that video that I think a lot of people will enjoy. But we did talk about the whole like ChatGPT for teens, and I think my take on it will probably surprise people that I don't really think teens should be using ChatGPT that much. I mean, there are some use cases where I think it's helpful, but I think they should resourceful and actually do more digging than just getting every answer from ChatGPT. But again, it's a rabbit hole and a rant that I went down on my recent livestream a little bit deeper. Along similar lines, college students can actually get free access to Google Gemini. So this week Google announced we're offering 1 year of Google AI plan free of charge for eligible college students around the world, plus new enhanced study tools so students can make the most of the school year. So if you are a college student and you want Google to help you on your homework, well, You could do it for free. Alibaba released a new music model this week called Happy Shrimp 1.0. I don't know what's a better name, Nano Banana or Happy Shrimp, but this looks to be a Suno competitor where you can generate music. Here's a quick sample of a future garage track. That's not what I was expecting at all from a garage track. Now, we did test this on our live stream and we made a couple pop punk songs about the Padres. Here's what one of them sounded like. Bottom of the inning and the tension is tight. Step into the plate in the stadium light. I'll leave it there. You get the idea. It's very, very generic sounding. I don't quite think this is as good as what Suno can generate, but I also do know a lot of people are looking for Suno alternatives with all of their latest restrictions that they're having. We also did try to get it to generate songs in the style of, and we gave it certain band names, and it refused to generate in the style of certain bans, similar to what Suno also does. We've got some quick news for coders as well. Cursor is apparently creating their own version of like a GitHub. It's called Origin, where you can, you know, store your code, send pull requests, browse code, and it even syncs to GitHub. So you can just like pull everything from GitHub straight into it. They're clearly trying to compete with GitHub. You can see in their blog post, they're telling you just to bring your GitHub repos right over. So first we got Buzz, which was an app that was similar to Slack where you can have a bunch of AI agents talking to each other, but then you can also have your human coworkers in there talking to you and the agents. And then we got Grokbot, which felt kind of similar but just more with like one person talking to multiple agents. And now Salesforce is rolling this feature out into Slack. Agentic coding is now multiplayer. Introducing Slack Code. We can see here you can pull agents directly into Slack now and they can write code directly inside of your Slack and Agents and humans can now interact with each other directly inside of Slack without needing other tools. Now, this one doesn't appear to run the agents directly inside of Slack. Instead, it appears to use your existing agent harnesses like a Codex or a Claude Code, and you can just tag those apps and it will spin up those channels. It'll still use those tools to run the agents, but you can chat with them directly from Slack. So I guess it's kind of more like a native OpenCLAW built into Slack, I guess. I don't know, it's something I'll probably experiment with more in the future. Another interesting research paper out of Pew shows that a lot of the internet these days is being written by AI. Like, this is sort of dead internet theory proof. So this chart here shows the percentage of AI editing or authorship going back to January 2022. There's very little. Fast forward to today, and apparently about 10% of websites are generated with AI these days. And if you filter out old pages, basically pages that were published before ChatGPT That number jumps way up. 35% of pages on the internet are generated by AI since ChatGPT came out. That's just crazy, but I guess not surprising. More and more people are looking for shortcuts. And finally, let's end on a robot. This Unitree new robot Superman can go from a stand to a 2-meter high jump. That's about 6 feet roughly. It has a top speed of 12.66 meters per second. I think that surpasses the standing high jump and running speed records of all humans around the world. Just take a look at this video. Look, that is just insane. Like, he was over that dude's head when he jumped. And here's a video of it running. And yeah, that's faster than any human could run. If they start making police robots that chase after the criminals, criminals are screwed. Anyway, that's what I got for you this week. Another one of those weeks where there was a lot of news, like a ton of things happening, but all feels like kind of smaller things. Like the improvements are sort of these marginal little jumps instead of big leaps. In fact, it sounds like these big leaps might actually get further between Again, because all the foundation labs are sort of scared to release their newest iterations of the models because of how good they are at the cybersecurity stuff. We do have Meta Connect coming up next month, and we do have OpenAI's Dev Day coming next month. So I wouldn't be surprised if the next iterations of models both of those companies are releasing are coming towards the end of next month to sort of release around those events. But I don't know, I'm just guessing. I don't have any insider information on any of this stuff right now. I'm just as in the dark as you are. Who knows how it's gonna play out? We'll see. If you wanna make sure you stay in the loop with all of the latest AI news, whether you love AI, whether you hate AI, whether you're somewhere in the middle and you just wanna keep up with it, I drink from the fire hose all week. I read all the news articles, I talk to the people, I test the tools. I spend all week staying looped into everything that's going on with AI so I can turn around, make this video on Thursdays to publish them on Fridays to keep you completely looped in on everything that's happening in the AI world. I'll be overwhelmed so you don't have to. If that's something that's interesting to you, maybe consider Consider liking this video and subscribing to this channel. I'm like very close to a million, hoping to hit it like within the next week or so. So that little subscribe button would really, really help me out to like just pass that little milestone that I've been trying to hit. But yeah, that's what I got for you. Excited to share some of these other upcoming videos in between the news videos. I really think you're gonna dig them. And, uh, again, yeah, that's what I got for you. I'm done rambling and nerding out. I never really know how to land the ship with these videos. So, uh, I guess, uh, bye.