GPT-5.2 Is HERE! (Plus Runway 4.5 and the Next Image AI Leaks)

As we get closer to the holidays, the news from the AI world is starting to slow down just a little bit. However, in the world of AI, a little bit is often still a fire hose of announcements. So despite this week being one of the calmer weeks in the AI world, there's still quite a few interesting things to talk about. Starting with we got a new model out of OpenAI this week. We got ChatGPT 5.2. Now, we've known about this model coming out for a little while. A lot of people thought it was going to come out on Tuesday. They ended up putting it out on Thursday. I don't know about you, but the old ChatGPT model, the 5.1 model, it was kind of starting to get kind of dumb. I noticed it was making a lot of mistakes and a lot of errors, so this new model couldn't have come soon enough. Unfortunately, I'm actually out at Rivian right now for an event. They had their AI and Autonomy day. So I haven't tested this new model yet, but I do have some of the details about it so I can make sure I loop you in. Now. It seems that this new GPT 5.2 model is a little bit more focused on math and science and getting to the top of the benchmarks in those areas. So hopefully the hallucination issues and the making up information that stuff is solved. I don't know yet. On their website, the benchmarks really only compare this to like other ChatGPT models. So we know that it's better at science questions than GPT 5.1 and we know it's better at advanced mathematics than GPT 5.1. But I don't have a comparison in front of me of how it compares to other models outside of ChatGPT. Maybe if we can find it, we can put the comparison up on the screen of the benchmarks of the non OpenAI models. So it looks like the new GPT 5.2 model has a 400,000 token context window with 128,000 max token output. So that's big. Basically the amount of words in and out that you can give this model. So you can give it and receive a combined 400,000 tokens, which is roughly 300,000 words. But only 128,000 of those can be output, which do the math on 75% of that. Maybe we could put it on the screen. 75% of 128,000 tokens, that's about how many words you can get as output from this new model. As far AS API goes $1.75 per million tokens input and $14 per million tokens output. Pretty comparable to a lot of the other models that do similar things. So on software engineering, the Swebench Pro GPT 5.2 scored 55.6 versus 50.8% on GPT, 5.1, 52% on Cloud Opus 4.5 and 43.3% on Gemini 3 Pro. So supposedly this is a lot better at coding science questions. It pretty much blows the other models out of the water. Fairly comparable to Gemini 3 Pro. And kind of like I expected, this is a fair marginal upgrade. I will do more testing with this once I actually have a chance and I'm back at home. But that wasn't the only news that we got out of OpenAI this week. They also partnered with Disney, which is something I don't know if a lot of people saw coming, but it sounds like Disney is investing a billion dollars into OpenAI and OpenAI is going to be allowed to use Disney IP. So I think you're going to be able to generate like Disney characters and stuff like that inside of Sora and maybe in their image generation platform. And also Disney's trying to do this thing where you can generate videos on Disney with some of your favorite characters in it. So it seems like this deal is all tying that together. So don't be surprised if on Disney pretty soon we see some sort of ability for you to generate a Moana short using Sora technology. We also got like a potential leak out of OpenAI this week. These have been circulating the Internet so new OpenAI image model finally potentially cited. So this post from can here inside of the the AI rumors and insights community says it's happening. See below for outputs. Most likely these are from the next GPT image model. They're being tested right now on Design arena and LM Arena. Most likely it will be released alongside GPT 5.2. Here's their key observations. World knowledge similar to Nano Banana. So when you ask it to prompt something, it might actually do some research and collect information before actually putting it in the image for you. It can generate celebrity selfies with very similar quality to Nano Banana Pro, which is very interesting because like a year ago none of these models were letting us generate images of real people. Now they all seem to not care. And it can write code in images very well. What does that mean? Can write code and images? Okay, we'll take a look in a second. So right now the code names for these models are Chestnut and Hazelnut and it's quite possible they came out with OpenAI's GPT 5.2 model that we just talked about. But these are what leaked here. So we can see one here with text on a whiteboard. Now the OpenAI logo doesn't really look like it was written with Sharpie or dry erase marker, but the rest kind of does. Advancing AI for humanity. About the model, I'm ChatGPT4 large language model. So this must be them showing that it kind of did some research and then put what it found in the research into the actual image. They also mentioned it can put code on images. I guess this is what it's referring to. This is like some JSON here model GPT4O OpenAI knowledge cutoff January 2022 data set aesthetic and then it made a PlayStation controller here. And I mean looks pretty good. I'm sure there's plenty of images inside the training data of PlayStation controllers, so not the hardest thing to generate. I think the more interesting thing is the fact that it actually put some like JSON text on the image. Here's an image of a bunch of celebrities all looking pretty realistic. I think if I saw this image circulating I would. My first thought would probably be this is AI. But it's pretty dang realistic. There's just something in some of these faces, like in the eyes of Jack Black here, the skin of Paul Rudd. I don't know, there's just a few things that I can't sort of tangibly describe how I can tell this is AI. But I've seen enough images that I'm like, yeah, I can kind of tell. But saying that I also have the insight of knowing it's AI. So I don't know, maybe, maybe I would be saying something different if somebody sent me this image and I didn't know it was AI. And then here's another one. This one looks even more AI to me. But the fact that it's also slightly like lower resolution and a little more pixely makes it look a little more realistic. I guess it's almost like they generated a decent looking image and then like downscaled it to like increase the potential to fool people. Down here we can see they prompted with GPT Image 1, the existing image model. And this one definitely looks AI. Like the proportions on some of these faces look wrong. It looks like his head's growing out of Ellen's shoulders here. I don't know what's going on with her teeth up here, but yeah, this one has got a lot of telltale. Signs and I think is this two of Jared Leto? I'm not sure. Here's another one from GPT image 1, the previous model. The teeth on these are kind of a giveaway. It also seems like we've got like maybe some duplicates of the same person again. So the newer one definitely looks improved over the older one, but we'll see. Maybe this is out by the time you're seeing this recording because it came out with 5.2. I don't know. We also got a collaboration that I didn't really see coming, but I'm glad that it exists. OpenAI anthropic and block announced a new group called the Agenta AI foundation and it's backed by basically every major tech company like Google, Microsoft, Amazon, even Bloomberg and Cloudflare are involved. It's going to live under the Linux foundation, which is a nonprofit that manages like these huge open source projects like Linux and Kubernetes and Node JS and Pytorch. But you're probably wondering like, what's the point of this new foundation? Well, AI agents are starting to be used in real businesses and in consumer apps, but every company has been building them kind of differently. So without shared rules, these agents can't really talk to each other. They struggle to move between apps and they might behave like unpredictably. So this new foundation creates a neutral place where all these companies can agree on common standards. This prevents the tech world from splitting into incompatible systems and makes agents safer, more reliable and easier to use everywhere. So why should you care about this? Well, AI agents are soon going to be everywhere. They're going to be booking your appointments, they're going to be managing your emails, they're going to be troubleshooting your devices, they're going to be doing customer support, they're going to be helping with probably anything you do for work. And this foundation makes sure that agents from different companies can actually work together. Stay consistent, stay safe, and don't trap you inside of like just one company's ecosystem. It's like making sure all the electrical plugs use the same shape so everything works. We've got a cool update out of Runway this week as well. Runway finally started releasing access to their new video model called Gen 4.5. And according to their documentation, it's got state of the art motion quality, physical accuracy and prompt adherence. Supposedly, objects now move with real weight and momentum, liquids behave correctly, faces stay consistent, and characters should show subtle but believable emotions. Now it's also currently ranked number one in the global text to video benchmarks. So here's a few of the things that I tested with it. Holiday sales aren't just for toys and TVs, this sale is for something actually useful. Hostinger is still running their Cyber Week deal right now and it's for something I would honestly be lost without if you watch this channel, you know I use AI agents to help me track the non stop stream of AI news every single week. They pull release information, summarize articles and help me prepare prep these videos. But the hardest part isn't building the agent, it's hosting it so that it keeps running 24. 7. That's why I like to use Hostinger's VPS N8N hosting. They've made the whole process super accessible with a one click n8n install so I don't have to deal with server setup or Linux commands or anything like that. All I did was choose the KVM2 plan, which is the one I recommend for AI workflows. After you check out, Hostinger will auto install N8N for you, which is a super user friendly no code AI agent builder that I have multiple other videos about if you want a tutorial. And with this holiday sale it's just $5.99 a month right now and you can use my code Matt Wolf for an extra 10% off. And then once it's live, your AI agent runs in the cloud nonstop even when your computer's off and you get full root access backups and their AI assistant Cody to help manage everything. So while everyone else is buying TVs on sale, you can grab a holiday deal on something that actually saves you hours a week. Use the link in the description and the Code Matt Wolf to try it out. Thank you so much to Hostinger for sponsoring this portion of today's video. Now let's get back to it. So to start I wanted to test physical accuracy and complex multi object interactions. See how the objects sort of gel together. So this was my prompt. I'm not going to read the whole thing, but it's a slow motion shot of a crystal clear glass sphere rolling down a marble staircase, bouncing realistically on each step. The prompt's more detailed than that. I also gave it details around like lighting and camera that I wanted to use and stuff like that. So this is what I got from it. Keep in mind that gen 4.5 does not generate with sound like some of the other state of the art models that are out there. So we see this marble rolling down the stairs as water hits it and it bounces. But then at the end, it seems to be filled with water. I mean, it looks pretty good. If we take a look at our prompt and what was actually generated, we got a crystal clear glass sphere. Check. Rolling down a marble staircase. Check. Bouncing realistically. Check. Ish. A stream of water pours from above, splashing onto the sphere. Check. Accurate fluid dynamics. I would say yes. Ish. It looks pretty close. The camera tracks closely beside the sphere. Refraction to the surface detail. Yeah, I mean, it hit everything. So as far as prompt coherence goes, it definitely squeezed everything in that we were looking for. The ball interacted with the stairs, the water interacted with the ball. I'd say that's a good test. So next up, I wanted to test, you know, character acting and emotions in the characters. So here's the prompt I used. Feel free to pause and read the prompt, but let's take a peek at what it generated. So we've got a woman with an umbrella walking through the rain, waving at people. Some like neon signs in the background. And if we take a look at our prompt here. So a middle aged woman. Got that. In a rainy city. Yep. Crosses a street. I wouldn't say she's crossing the street. She's more walking down the middle of the street holding a transparent umbrella. Cool. Yep. Pauses. Looks up with a subtle smile as rain hits the umbrella realistically, then waves to someone off camera. Yep, that worked. And then I mentioned to keep it consistent, dynamic handheld camera with micro jitters. Dramatic backlight from neon signs. So we do have the neon signs in the background. And if I move this through this really quickly, it even does have the little like camera micro jitters. It looks like somebody's walking with the camera. Not like a perfect steady shot. So good job on this one too. Next, I wanted to try non realistic, like animation style. So I gave it a stylized 2D anime world where a young explorer runs through a floating island marketplace filled with hovering lanterns and animated creatures. Here's the result of that. It's definitely got the cartoon look, but I mean, there's definitely some wonkiness, especially going on in the background. The character stays fairly consistent, but when you look in the background, like, what the heck is going on here? I don't know what is happening. Like, yeah, these are some kind of crazy looking characters, but they almost look like really poorly drawn characters, unfortunately. And then we get over here and it's just nonsense in the background. So the cartoon thing here, not amazing. I'm not super impressed with this. I Think some of the other models probably do these cartoons a little bit better. So for this one, I wanted to test just some basic footage that can be like. Just like a slice of life B roll. I also wanted to test how well it did on more like macro shot details here. So I gave it a barista makes a latte and a warm coffee shop with more details about what I'm looking for. Again, feel free to pause the video and read my prompt. But here's what we got. Close up. They're pouring the milk into the espresso here. Then it zooms up and the person smiles at the camera. So taking a look at our prompt. Close up shot of milk swirling into the espresso. We back it up. Here you can see the milk going into the espresso and starting to swirl a little bit. The fluid dynamics looked pretty accurate. I mean, it looks like he's pouring that espresso that nothing looks too off to. That steam is rising naturally. Yeah, we got some good steam going here. Background patrons move softly out of focus. The barista wipes the counter, glances up, and smiles. So we definitely have some blurred people walking around in the background. The camera pans up to look at the barista, who is now, as we can see, about to clean the counter, and they smile at the camera. It nailed it. I wouldn't really say this is the most realistic looking face I've ever seen, but it's pretty dang good. I mean, as far as prompt adherence goes, this model is actually really, really impressing me. All right. And finally, my last test, I wanted to test, like, cinematic action and how well the physics do in this kind of environment. So I gave it a futuristic police drone chases a motorcycle through a narrow neon alley at night. We can see the drone. The drone actually looked like it was in front of the motorcycle at first and then got behind it. So a little bit of a weird perspective shift that happens there. But again, looking at our prompt. Police drone. We've got that chases a motorcycle. Check. Narrow neon alley at night. We're in a narrow alley with neon signs. Sparks fly as the bike scrapes a metal railing. Did we get that? I don't think we've got the metal railing. Oh, wait, we do. So right here, a couple of seconds in, sparks fly out of the back of the bike on, like, I don't know, something on the ground, I guess that's metal on the ground. It's sparking off of. Not exactly what I had in mind. You've got like a metal railing here, but the bike doesn't even hit it. So a little bit of wonky physics there. I don't know what the sparks were coming off of. Rain soaked pavement reflects lights accurately. I mean, yeah, it looks good. The drone fires a spotlight that tracks the writer. So the drone does have a spotlight. The camera switches from overhead to a tight flash following shot. We've got this shot from in front and it never actually switches to the following shot. So it might just be we didn't have enough time. Right. It's only a 5 second generation, so it sort of ran out of time to switch to that final shot. But overall it's pretty impressive, especially with prompt adherence. That's one thing I always love to test is I like to throw a lot of things into a prompt and see does it get all of the things that I tossed at. It did. It managed to sort of separate out all of the details and get them into that video. And 4.5 here seems to do all of that really, really well. Now as far as realism goes, I don't know. I still think I might give an edge to VO 3.1 on realism. And also most of the newest like state of the art models also generate Sound. So like Veo 3.1 Sora, the new Cling 2.6 or whatever it was that generates sound. So most of the newest state of the art models also generate sound with the video. So I do feel it's a little interesting that Runway doesn't have sound in this model, but visually definitely one of the more impressive models. We did get some new large language models this month. I'm not going to go too deep into playing around and testing all of these models. I'm just going to kind of quickly share the various models that came out. Mistral, the French AI lab released Devstral 2 and the Mistral Vibe CLI or Command Line Interface. So Mistral's really going down that like vibe coder route of let's give people tools to write code. If we look at the benchmarks here, their Devstral 2 model is this 72.2 model here. And these are all of the open weight models that are available. Deepseek being probably the best model in the open weight world right now. And this Devstral 2 is pretty much right on par with it, just slightly below it and actually performs better than the proprietary Grok code Fast one and just barely below the state of the art models. Gemini 3 Pro GPT 5.1 Claude 4.5 Sonnet, not Opus, but their second best model, Sonnet and these new models are open source. So if you're someone that wants to do like vibe coding stuff, but you want to do it locally, you want to download the large language model onto your computer and actually run it yourself without connecting to any cloud servers. Well, right now it seems like your options are kind of deep. Seq v3.2 and Devstrol2 are going to be the best open weight coding models that you can run totally on your own cloud or on your own computer. We got a couple new models out of some of the Chinese labs this week as well. Zai debut their open Source model called GLM4.6V and this one is actually capable of tool calling and using vision. So it looks like there's two models. There's the 4.6V and the 4.6V flash. One's 106 billion parameter model. One's only a 9 billion parameter model. So this one you could almost certainly run locally on pretty standard GPUs as well. Taking a look at their benchmarks here, they're pretty much on par with other models in the, you know, similar open weight world. And these are mostly just being compared to the other Chinese models here. Quin3 also updated one of their models. Their Quinn3 Omni Flash got a massive upgrade. So they improved multi turn video and audio understanding. You can customize your AI's personality through system prompts. Smarter language handling and voices are indistinguishable from humans. You can try it over at Chat Quinn AI and we can see here these are their benchmarks on this model. They did Compare it to GPT4O and Gemini 2.5 flash. So not comparing to the most state of the art models, but you know, comparing it to decent models. And they pretty much do on par with the other models or a lot better. But again, they're sort of cherry picking who they're comparing to here. All right, now I know I've already shared quite a bit and I mentioned it was a slow week, but I've already talked about a lot of stuff. There are just a few, a few more things, just, just a couple that I want to share, but I won't go too deep. So let's jump into rapid fire. Pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew, pew. So in last week's news video I Talked about how OpenAI was starting to roll out ads into ChatGPT and people were noticing them and going, hey, this sucks. And I also agreed in that video that hey, that sucks. Turns out they weren't actually putting ads into ChatGPT. They were building out their like shopping fe and their shopping features just happen to look a lot like ads. Mark Chen from OpenAI said, I agree that anything that feels like an ad needs to be handled with care, and we fell short. We've turned off this kind of suggestion. While we improve the model's precision, we're also looking at better control. So you can dial this down or off if you don't find it helpful. So again, if you have no clue what I'm talking about, here was the original post on Twitter from Benjamin decracker saying that he's on a plus subscription, a paid subscription, and he saw this shop for home and groceries connect Target here. He shared this and said, OpenAI has ads now. What the heck? That prompted this response from Mark where Mark essentially said that wasn't an ad, that was us testing new shopping features. Benjamin said, well, if it looks like an ad and we perceive it as an ad, there's really no difference to us. So OpenAI decided we're going to remove that feature for now and get it all fixed and cleaned up so there's no confusion about whether or not that was an ad. Why don't you let me fix you some of this new MO cocoa drink? All natural cocoa beans from the upper slopes of Mount Nicaragua. No artificial sweeteners. What the hell are you talking about? Good move on OpenAI's part, I think, since we're talking about OpenAI and ChatGPT. ChatGPT rolled out a new feature where you can now use Adobe apps to edit your photos and PDFs for free directly inside of ChatGPT. I haven't actually tried this yet, so let's just jump over to ChatGPT real quick, see if it's rolled out in my account. If I click on my plus, let's go to more. Let's add sources and it's not on by default. Let's connect more. Okay, so there we go. We can connect Adobe Acrobat, Adobe Express, Adobe Photoshop, and a lot more connectors that than the last time I looked in here. I don't remember seeing a lot of these. So they've been really cranking out new connectors. But we have Adobe Express and Adobe Photoshop. Let's connect Adobe Express. Then we'll click the plus and then select Adobe Express. So theoretically I can have it edit images for me using Adobe Express in ChatGPT. So let me upload an image here. This is just an image of myself and let's Say, remove the background and make it a green screen. I mean, theoretically I can do this just with the regular image model and not need Adobe. But let's go ahead and use the Adobe connector and see what it does. Okay, so I guess that's not something Adobe Express is capable of. So Adobe Express can't do that in ChatGPT yet, but you can search the design, fill text, change background color. Isn't that what I just asked for? Okay, replace image animate design. Okay, change background color to green. All right, since this is your own uploaded photo, not an Adobe Express design template, I can't apply. You know what? Forget it. I will just use Adobe Express if I want to do this stuff. I'm not actually interested in learning how to do something I can do in 10 seconds in Adobe Express when it's making me have a conversation in ChatGPT. Anyway, anyway, just a random tangent. You can connect Adobe now and it'll tell you, right currently that you can't do stuff, but you can connect it. Not a whole lot to talk about with this news story, but the limitless pendant, which is like this little thing you see on his shirt right here, it's a little clip that literally records everything. It's just always on, always recording, records all your conversations, records your phone calls. Well, you know your side of the phone call at least. It basically records your day. And then you can go back and like look at conversations you had and recall things that you told people you would do. And it can make to do lists for you off of those things. Well, they've been acquired by Meta. So the company that's putting cameras and speakers and microphones on your sunglasses now also owns a company that puts microphones on your body and records everything. That's not scary. Company that's gotten in trouble in the past for mismanaging our data is collecting all of the audio data of conversations we're having throughout the day. Why are you so obsessed with me? I don't know how I feel about that. Let's move on. We got another really cool like image thing out of Alibaba called Quinn Image i2L, which I believe stands for Image to Laura. Now if you're not familiar with Lauras, I am going to way oversimplify it, but they're essentially like little like plugins that you can add to like an image model or a video model that sort of steer it in the direction of the Lora. So if the Lora has details about what you look like, it can steer the image model to Generate things closer to what you look like. If the Lora has training data on a certain design aesthetic like Studio Ghibli or something, you can run that Lora with a prompt and it will more steer the model towards what the Lora tells it to do. Again, super oversimplification. I probably made it even more confusing by saying that. But you can see here, this new model lets you generate those Loras with just one image. So if you want to prompt an image model and make it look like a Studio Ghibli every single time, but be the exact same style, very consistent with the imagery and style, you give it one image of a Studio Ghibli photo, tell it to generate a Lora out of that. And now if you include that Laura, it will generate any future prompts with that style. I probably just confused you even more, but it'll make it a lot easier for you to replicate styles, people, objects, your pet, things like that. Pretty cool. Now, this week I've also been at an event at Rivian called the AI and Automation Day. Rivian's doing some pretty cool stuff. So let's jump to future me, who's at the Rivian event, and I'll break down some of the cool stuff I'm hearing about at this event. All right, so Rivian made some pretty cool announcements today about what's coming to their vehicles. And guess what? A lot of it has to do with AI. One thing that I found really interesting about today's Rivian keynote is that Rivian's actually getting into making their own silicon. There's going to be a new Rivian chip in future Rivians, which seemingly is a play to release reliance on Nvidia. But when we ask them about what's going on with Nvidia, they said that they're still working with Nvidia and that they're going to rely on both chips for a while. I think they're still trying to figure that stuff out. But the point is Rivian is making their own chips so that a lot of the AI is going to run on device, on the car, on the Rivian that you're using. There's going to be four waves of autonomy when it comes to Rivian. I actually got to test one of them earlier, which was a self driving mode where it drove us around Palo Alto here with hands off. There was a few interventions necessary. We went over a speed bump and the car stopped immediately after the speed bump and the driver had to, you know, press the gas pedal to get it to Sort of kick back in again. But for the most part, it drove around Palo Alto with no issues, stopped at stop signs, stopped at red lights, did everything it was supposed to. But this is just their phase one. That's the universal hands free, where you could be hands free, but you still need to be ready. You need to be ready on the gas, ready on the brake, ready to steer the steering wheel if you need to. You still need to be paying attention. The next mode that's going to be coming out after that is called point to point autonomous driving. Where you give it where you want it to go and it will take you from where you are to where you want to go, but you still need to kind of be paying attention. The third phase is what they're calling eyes off autonomous driving. You could be literally turned around in your chair, talking to people behind you. You don't have to have your eyes on the road and it will be driving for you. It sounds like that could be something coming towards the end of next year in Rivians. And then the final phase is called Personal L4 or Level 4 Autonomous Driving. And that's where it can just drive itself on its own completely. Like nobody needs to be in the car. So for instance, one example they gave was, let's say you need a ride from the airport. You can summon your car to drive from your house to the airport to pick you up, and it just drives itself to you. You hop in the car and it drives you home. That's like true level 4 autonomous driving, which is what they're working for. We're not going to see it next year, maybe 27, maybe 28, I don't know. But that's what they're working towards here at Rivian. The next generation of Rivian's are going to have LiDAR, similar to what Weibo has, except instead of like a bump on the car, it's very integrated into the car, you can barely see it. And they're also rolling out a Rivian assistant, which is kind of like Rivian's version of Alexa in your car. Or like a Siri in your car. You could just say, hey, Rivian. And tell it what you want to do. It will connect to your calendar, it'll connect your text messages, so you can say, hey, what do I have to do tomorrow? Hey, can you send a text to Joe? It'll message him and let you speak out the text. But it also knows everything about your car, so you can say, you know, hey, Rivian, I'm getting cold. And it will bump up the heat for you. Or hey, Rivian, can you turn on the butt warmers for everybody in the car but the passenger seat? And it will do exactly that. So you can use voice commands to do everything in the car. Now I own a Rivian. One of the biggest pet peeves I have about the Rivian is the big touchscreen. Every time you want to change anything in the car, you have to use the touchscreen. So if I want to bump up my AC a little bit, I have to find the button on the thing, open it up, slide the slider up and change my temperature. If I want to change the volume on the stereo and I'm not the driver, I can't do it on my little scroll wheel right now. I have to press the volume button and then drag my finger up to change the volume. So a lot of the touchscreen controls can get kind of annoying, especially if you're driving. Well, this Rivian assistant, you're going to be able to talk to your Rivian and it will do various controls in your car for you. So pretty, pretty cool stuff. This was really fun event. I don't normally get to come to events that are about vehicles, usually about events where they're telling me about the latest software and large language models. So it was really cool to get hands on experience with actual physical products. And the last little bit of news that I want to share this week, there's been a new McDonald's ad circulating that you've probably seen or at least heard about. Now in this ad, it's a whole bunch of people talking about how they absolutely hate Christmas and can't wait for it to be over. And it's a bunch of people meeting misfortune like getting thrown out of a window or their presence falling off their car, or somebody getting stuck in like a trolley door and then getting dragged by it. And the whole thing was AI generated and everybody hated it. Now here's my take. I think people have been so inundated with all of this, like, AI slop on social media. And we're getting to this point where almost every image or video you see, you have to wonder, is this real or is this AI? I can't tell anymore. Like, it was especially apparent during Halloween where we were seeing all of these videos of like crazy Halloween costumes or like, you know, cats ringing a doorbell and trick or treating and not the cats one. But some of these videos were very realistic and hard to tell, were actually AI and they were getting shared around as if people thought they were Real videos. When people learned they were AI, they were frustrated because they didn't realize they were looking at AI videos. I think that people, myself included, are so fatigued by seeing AI generated slop everywhere that when we see a company like McDonald's or Coca Cola go and generate more AI slop, but for their company, it just pisses us all off. Holy, I am pissed, right? Because it's like this is a multi, multi, multi billion dollar company with financial resources to do whatever they want. They can hire the best actors, they can hire a list actors to come and be in their commercials. They can hire the best visual effects people, the best musicians, the best voiceover artists. They have the capacity to do a high budget video. And so when we see them generating AI slop that sort of feels like all the other cheap AI slop that any individual can generate. We're like, why are you guys doing this? You don't have the financial excuse to be using AI to generate this stuff. Use real humans. Go pay people to do this stuff. We're getting enough AI slop. We don't need our commercials to be that also. And I think that fatigue is setting in. And whenever we see that kind of stuff, we're always going to get backlash. AI needs to be used as like, in my opinion, in my opinion, for stuff like this, it needs to be used as a smaller tool, right? Like make 80% of the commercial with real humans, with real graphic designers, with real VFX artists. Make like 80% of it with, with them, right? If there's a few scenes that would be hard to generate without AI or really, really, really expensive to generate with VFX or might take months and months to generate. Okay, that little, like maybe 15 to 20% of the video use some AI to get those shots. But come on, like you can afford to get real actors and real VFX artists to do this stuff for you. It just feels cheap and lazy on the part of this mega, multi billion dollar corporation. Anyway, on that happy note, that's pretty much what I got for you today. So if you like stuff like this and you want to stay looped in on all of the latest AI news as well as get tutorials on to actually use all of this cool AI stuff that's coming out. Make sure you like this video, subscribe to this channel. I will make sure more stuff like this shows up in your YouTube feed. I am keeping my finger on the pulse every single day so that you don't have to. And then once a week I'm putting out these videos, breaking down everything that happened to keep you looped in again. If you like that stuff, maybe give it a subscribe and hopefully I'll see you in the next one. Thanks again for hanging out nerding out with me today. It's been another fun week and I don't think it's gonna slow down. All right, later. Thank you so much for nerding out with me today. If you like videos like this, make sure to give it a thumbs up and subscribe to this channel. I'll make sure more videos like this show up in your YouTube feed. And if you haven't already, check out FutureTools IO, where I share all the coolest AI tools and all the latest AI news. And there's an awesome free newsletter. Thanks again. Really appreciate you. See you in the next one.