AI Recap: New Models, Jailbreaks, and & Future Tech!

Alrighty guys, welcome back to another AI News roundup. We've got so much to talk about today. As usual, brand new open source models have dropped. As well as closed source models from big competitors, we've also got new products, research and demos for video generation that are just mind blowing. So plenty to discuss. All great things happening in the AI space. There's a ton of competition right now and it really feels like we're kind of at that hard launch takeoff moment. We knew the cat was out of the bag when Chat GPT released, but man, I don't think people are ready for what this cat is capable of doing. All right, enough talk. Let's dive in. OpenAI's brand new deep research model, the highest benching model OpenAI has ever put out there for us, has already been entirely jailbroken by Elder Plinus, AKA Pliny the Liberator. Pliny is known throughout the web for jailbreaking AI models and releasing those jailbreaks publicly on his GitHub. He's the best of the best and even just mentioning his jailbreaks and his name to deepseek r1 could result in the model becoming jailbroken. So it's, it's no shock that he is jailbroken. OpenAI's deep research but it does beg the question, at least in my mind, why are we having all of these high safety tuned closed source AI models if they can all just be jailbroken anyways without really too many fancy tricks? Like literally I could just go to GitHub, download some of the jailbreaks, they're literally just prompts and put them through the model and I end up with a jailbroken model. So anyways, he got it to do PhD level drug synthesis recipes how to get away with unaliving people making poison. Yeah, it'll do anything. So jailbroken means jailbroken. Diving super deep on this topic isn't in the scope of this video, but I'm sure you guys in the comments as always will provide lots of awesome insight, lots of great takes and opinions and we can have fruitful discussion down there. If there's one thing that always impresses me about you guys is it's your ability to think outside the box, communicate with each other and share ideas. And we need more of that, especially if we are in a hard takeoff AI scenario. And speaking of Open source models, there is a Deep Research competitor by Hugging Face that is fully open source and you know, comes out within four days. I think of OpenAI's deep research and they also did this in only 24 hours, a 24 hour deadline to replicate and Open source Deep Research. Yeah. So this agent can obviously navigate the web autonomously, scroll and search through pages, download and manipulate files, run calculations on data. They aim for the best performance. But I think they did a pretty great job at matching up to Deep research with that 24 hour deadline. Absolutely crazy. Yeah. So on the Gaia benchmark, deep research from OpenAI gets a 67% accuracy, which is great. Deep Research falls behind it at 55%. But it's fully open source, it can be modified, it can be made better to get better scores on benchmarks. And I'll bet you it doesn't cost 200 bucks a month to get access to this model. And again it brings up the question. It's like we have these closed source models from the likes of OpenAI or Google, but then they just get replicated by the open source field anyways and very quickly at breakneck pacing. So not only do the closed source models get jailbroken anyways, but they also get replaced by open source models that don't need to be jailbroken in the first place in a lot of circumstances. Again, you guys can debate healthily in the comments about whether or not this is a good or bad thing. All I can really say is the speed of AI development is just rapid. It's blistering unlike anything I've ever seen in my life. Quite young life. Alright, in other Hugging Face related news, they actually updated their website as well. It now has an AI based search function that allows you to pretty much just ask it what you are trying to do, give it your specific context, your specific scenario and it will access the Hugging Face space database of over 400,000 spaces to find exactly what you need to get the job done. A nice quick little demo video here brought to us by Victor Mustar. You can generate a 3D model from an image. It will pull up the best spaces for that particular task or you know, image to video generation, whatever it might be. So this is actually a really cool feature that I have wanted on Hugging Face previously and I'm glad to see it's finally available. In other updates, Sam Altman here announces that Chain of Thought is finally visible for their O3 miniseries of models. He does note that it's not the raw chain of thought from the model, but they're organizing it apparently to make it more readable and optionally to translate languages, but they're trying to keep it very faithful to the raw one. So yeah, this is already rolled out in chat GPT you can see, you know, if you send it a couple of emojis, it's going to reason about the emojis. Okay, the user sent me some emojis, superhero, owl, calendar, clock. And I gotta say, reading through the chain of thoughts is always so interesting. I love doing it with the R1 models before it was now available in these OpenAI models. And I gotta say, using it in the R1 models and comparing it here to the OpenAI 03 models, the chain of thought feels very, very similar. Again pointing more to the fact that Deepseek R1's paper and models are a great replication of what OpenAI did with the O series of models. This example is also pretty cool though, because these emojis do have a secret that the AI was able to figure out. The user is likely asking about the super bowl. The emojis, superhero and owl hinting at Super Bowl a pun and then it proceeds to, you know, give you the actual super bowl time. So pretty cool. Regardless though, think this is a part of the reaction to the fact that now we do have full access to chain of thought with models like Deepsea Car one. So why hide it anymore, I suppose. Next up we have Replit, which just added a brand new app on the app store. But this app's a little bit different because again it's replit. Essentially you can just ask the AI for personalized quick apps that you might need at any moment. I'll show a little bit of this quick demo off to you guys so you can get a good idea.

Always wondered why that app didn't exist. Now you can make it meet Replit agent, want a custom workout plan? Make an app for that. Have a great idea at 2am make an app for that. Need to run an analysis or learn with your kids? Make an app for that too. This isn't just making apps, this is your imagination unleashed.

It can do quite a lot as you can see all from your phone. I haven't tried this one out but it's on my list. And I really do see this as a part of the future of the way we interact with computer systems. I think in the far future gone will be the days where we have pre made user interfaces and they will be replaced with on the fly generative ones that will look just as good and work just as good but be custom tailored for whatever situation we need because that's what AI is capable of. The future is generative. So yeah, overall, overall it's. It's super darn cool. I like the idea. I like the concept and I do like the slogan make an app for that. In the comments here, we can see Paul Covert apparently built an entire agent today with it. Other devs say they're going to be making apps on every single commute. This comes from Amar. Yeah, I think the core audience here really is devs, but hopefully this should kind of open the door for more regular, average, everyday Joes to create a customized experience for themselves utilizing the power of AI technology. Again, if we're in a hard takeoff scenario, normalization is what needs to occur. So I see this in a very positive light. Alright, guys, before we dive into the rest of today's AI news, this video does have a sponsor, but this channel is only sponsored by companies where I've actually used their product and enjoy doing so. And at the end of the day, sponsors allow me to pursue this channel full time. I want to share an AI tool that changes how we bring our ideas to life. Just imagine this. You've got a brilliant film idea swirling around in your head, but actually turning that vision into a reality would traditionally mean endless storyboarding, writing a whole script. But in the age of AI, what if we could just skip all that and visualize our concept almost immediately? Well, Nvidia V3 does just that. Let me show you an example. Using their Hyper Realistic Films workflow, first we select workflows on the main page, then Hyper Realistic Films. Then we input a detailed text prompt, up to 32,000 characters worth. So you could put your whole script in there if you already have it. You're gonna want to describe your film idea, though, with all the specifics, the characters, sett dialogues, you name it. Once you click the generate button, Invideo AI crafts it all for you. It makes a comprehensive script. It visualizes your scenes. You're the linchpin, the one who changes everything. What are you talking about?

I'm from the future. And I'm here because of you.

Right? I think you've had enough. Look at your watch. As I glanced down, I saw the impossible. The hands of my watch were spinning backward. But InVideo AI v3 doesn't stop there. To perfect your creation, they offer an enhanced editing experience that's super easy to navigate and use. Tweak the script, manage your media, add voiceovers, music, even sound effects. And if you work with a team, they have great collaborative features. Work together in real time and review and refine your content seamlessly. Are you ready for the future of filmmaking and content creation? Well, check out InVideo AI v3 with the link down in the description below. Start visual and bringing your ideas to life effortlessly. Trust me, you'll never look at content creation the same. Huge thank you to Invideo for supporting the channel. Now back to your regularly scheduled content. Welcome back guys. Let's dive into some more advancements. Next up we have something from Play AI, which was formerly Play HT. They've got a new model called Dialogue 1.0. Apparently it's ultra emotional and it's a text to speech model. There's claiming that this thing outperforms Elevenlabs, who is, you know, the king. The king in the text to speech space on both expressiveness and quality. 3 to 1. It has a sub 1% error rate. Supports 30 plus languages. Apparently the voice cloning is also really good and pretty decently low latency of only 303 milliseconds which might make it somewhat capable for real time applications. Let's check out the little demo so we can get a taste of what this model actually sounds like. This is definitely something I might take a closer look at in a deeper dive video because I've always loved text to speech AI, especially ones that are very realistic sounding in terms of emotion and expressiveness. I think 11 Labs is really good, so let's see if this holds up.

Get ready to dive into a classic from the golden era of house music. Up next we've got the legendary David Morales with Here I Am, a track that showcases his iconic touch in blending soulful vocals with infection beats. If you're loving the vibes today, stick around for more great tunes.

Oh dear, did I knock over your little plan? Tragic, Truly tragic. It's pretty good. Every chatbot, virtual assistant and automated message out there, Listen up.

Today's guest is the author of the best selling book about the healthcare industry titled everything is fine. Dr. Jonas Salka. Welcome to the show.

Hey, thanks Anara. I'm honored to be here. As honored as a hospital billing department. I mean that little demo clip, obviously these are all cherry picked. It sounded pretty good. I'm not gonna lie. Like obviously the first one we heard here sounds like a radio host. There is some music in the background though, which might be saving any errors or slight mistakes in the audio.

Get ready to dive into a classic from the golden era of house music. Up next we've got the legendary David Morales with Here I Am.

But voice more towards the middle of the video plan. Tragic. Truly tragic. That's really good quality. Like that's. That's pretty staggeringly impressive assistant and automated message out There, listen up that I think you can hear a little bit of like roboticness or a little bit of like voice crackiness. I think you could absolutely convince someone that all these audios come from a real person. But I guess the question would be whether or not it's able to trick the brain in regards to it being a professional voice actor, right? Or something that's completed or finalized. Again, I'd love to do a deeper dive video, so let me know if that's something you'd be interested in seeing. In regards to this, apparently it benchmarks 3 to 1 against 11 labs and according to user preference they choose expressiveness and pacing more than any other factor like audio quality, voice accuracy, even hallucinations. The one thing that does make me very excited is we got a comment here from Mr. Fake Name any plan to open source and they just send the eyeballs which could really mean anything. But if they did open source this model, that would be incredible. Play AI. If you want to earn the trust of the AI folk, open source your models, man, that's the way to go. In other news, we've got some big drops from Google expanding the Gemini 2.0 family with some new options. First up, Gemini Flash 2.0 upgraded performance in comparison to their previous model and it's faster. Gemini 2.0 Pro. Apparently it's still experimental too. Best model yet for coding performance and more complex prompts. And they've also got Flashlight which is an even more lightweight model and according to Google, it's their most cost efficient model yet. 2.0 flash thinking experimental is also available officially now, so that's out of like the API testing phase. Honestly, I don't think there's too much to say here. A lot of these models we've already had access in a preview to, they aren't bad models by any means. All of these companies now have really good LLMs that are really useful for a wide variety of tasks. But you know, does this compete with the likes of OpenAI that have their O3 mini things models or Deepseek R1 which is open source or you know, the open Deep Research variant from Hugging face At the end of the day, do I think this is going to drive more competition in the market? I don't really know. Like I said, all of these models are great for a wide variety of tasks. So I think it's about how can Google integrate this effectively into products that people are going to be using every day? How seamless can it get, especially if they can't compete at that Upper echelon benchmark topping. Most impressive. Closest to AGI models out there. I don't know. Let me know what you guys think about this. All right, now we get into some fun stuff here. First up, we've got Small Brain releasing. I have no friends AI. Apparently you can create any character and just talk to them instantly on this app. The, the trailer here, this little thing they made is hilarious. We've got like 80s or 90s old fashioned Sam Altman.

Are you sick of your real friends? Are you a sad, lonely person? I'm gonna fix your loneliness with AI. Go to ihavenofriends AI and talk to AI generated friends. Swipe for a new friend. Swipe again for even more new AI friends. Just keep swiping. I had to talk to a real person in three weeks.

Amazing.

Go to the website. I'm waiting just for you.

Thank you. Scary Sam Altman. Oh, he's so uncanny. Pretty funny though. Here's the website in case you guys are interested. Yeah, it's about what you would expect. It's just a bunch of community made insane AI friends you can talk to. Yeah, this one's more just for fun. There are similar websites out there and available. I thought, I thought I would share it with you guys though because you might find it kind of interesting and it's. It's free to hop on here and mess around with it. Alright, you want to see something mind blowing? Guys, this is Dyne vfx. This can augment real videos with dynamic content. It's got a project page and hugging face paper. No code just yet, unfortunately, but it's really cool. Take a look at the demo video. Obviously, again, this is a real video, but we're adding effects in with AI. So turn the car across the thing and then immediately we add a massive tsunami and it totally overfloods the video and changes things entirely. I mean, there it is added in. Massive tsunami crashing against the windshield. It looks great too. I know the quality here on Twitter isn't amazing, but we'll also take a look at the project page. Yeah, the input video versus the Dynev effects is a huge difference. It can totally add a tsunami into this scenario. Add a puppy peeking out of a box. We've got a golden retriever added into this video running through a stream. I think the golden retriever looks maybe a little bit big, a little bit morphy, but pretty cool nonetheless. Adding a massive whale into a video of some scuba divers. A lion in the jungle. That is a massive lion. Very scary. Where you can make it a giraffe. It does a pretty good job though, at blending everything seamlessly. I would love for this to become open source. I hope they have plans for that. As you can expect, though, this is still in the research phase of things. It does say, hey, code is coming soon. Hopefully we get weights and everything though. Yeah. Some more examples available on this project page. As always, you guys know everything's going to be linked down below. So many possibilities. And I do have something very similar to show off to you guys later in this video from another company as well that is accessible today. So ByteDance is added again, dropping another model. This is Omni Human 1. It takes a photo input of a person and audio and then animates it all together. OMNI Human, the AI colony on Twitter here brings us 10 examples we can look at. I'm not sure I'm gonna go through all the examples, but again, everything's linked down below. If you want to get a closer look at anything we talk about. This is AI generated. I assume this is like a celebrity, guys, I'm not. I'm not a big celebrity person. I've been criticized in the past for not even knowing who Tom Hanks is. Forgive me, but I really just don't care all that much about celebrities. I'm sorry.

Now, some of you may know me as a stand up comedian and from my appearances on Roast, but I am not here to roast you tonight. I want you to know that. And how could I really? You're all so famous, so talented, so powerful. I mean, you could really do anything. I mean, now.

Wow. Holy crap, dude. My word. That's shocking, man. That is just shocking. Obviously the. The audio is helping it a lot here. It's real audio from. I assume the same clip as the reference images from. But dude, it animates it so well. I mean, you could show this on your phone to someone random on the street and they could just be like, yeah, what are you showing me this video for? Actually, it's all AI generated except for the voice. Insane.

Now, some of you may know me as a stand up.

Like there's a little bit of weirdness going on with the. Like she opens her mouth maybe little too much in the beginning, but that's crazy. Look at the lighting on her dress. I guess her body kind of morphs a little bit like right there. But I don't know, man, I'm. I'm struggling to find flaws with this.

Really? You're all so famous. So what.

How are we this far along? Bite Dance. What Are you doing? Oh, now we have Jensen Huang rapping. This is gonna be ins. Dude, I can't. It's crazy that it does it with his arms moving too. We've got some audio from Albert Einstein paired up, obviously, with a reference image of him, as if he was actually talking, kind of bringing him back to life. This is really cool.

What would art be like without emotions? It would be empty. What would our lives be like without emotions? They would be empty of values. So a famous classical poet said, we hate and we love. Can one tell me why?

Look at his body, like, moving his hands.

Science does not answer why questions.

That's just. I mean, it looks like a real freaking video. Wow. And I know about this because I had the opportunity to stand on top of the Olympic podium and feel the.

Weight of that gold medal around my neck.

This is a little bit worse, I think, than some of the other examples as it deals with her head turning and then the microphone as well, and also the strands, her hair kind of floating in a slightly unrealistic way. But we're really nitpicking here. Staggeringly good. Just. Just shocking on. On multiple levels. I don't know what to tell you guys. Holy cow. Yeah. I'll link this down below again if you want to check out some more examples. Let's move on to Meta's latest release, Video Jam Joint appearance. Motion representations for enhanced motion generation in video models. This is new research from, again, Meta. I'm really hoping they open source this sucker, but we've got a project page for it. Again, I'm gonna check that out in closer detail for you guys. So, first of all, we have vastly improved body deformation. This is dit, and then this is DIT plus Video Jam. And we can see that the Video Jam version has way more coherent body parts and motion. I mean, it looks like 10 times better, man. Way more usable. Actually looks close to a real video, where this one we've just got, you know, crazy morphing insanity going on on the side. Absolute AI mush. Next up, we have motion and physics, which VideoJam also tackles again, DIT over here. And then VideoJam added on. Way, way better cutting physics. It's actually stable. And it appears that the singular object is being cut through on the side. We've just got a knife kind of going in and out of frame, and then a hand kind of down there. It's a little bit weird. It's a lot better with Video Jam. Again, more improvements to physics here. We've got touching, like some slime I suppose can see the slime touching the fingers. Very cool. And apparently this can be easily applied to any model as. But we've also got Sora versus Video Jam. Very difficult prompt here. We've got a woman with one of those rings. I think they're like. I don't know. Correct me if I'm wrong, but this is like gymnastics, right? Or gymnastics aerobatics. Either way, it's very difficult for AI to produce in videos. Sora obviously has this woman morphing around and twirling like insane. And then Video Jam, you know, she's just sort of stable, rotating around with the ring. We've also got juggling again. Something very complex, difficult to replicate. Destroys Sora in every way. Hula hooping again destroys Sora in every way. You kind of get the idea. We've also got cling 1.5 versus video jam. Cling. We've got a floating dude. This guy I assume. Yeah, there he does a handstand in front of the lake. Jump roping. Something that I've tried in the past with multiple models and seen pretty much no successful replication of Video Jam destroys it. Does actual good jump roping for the most part and at a beach as well. Well, we've also got blowing out some candles again, much better. With VideoJam, it seems like this add on that you could put on your models in vastly improve things like physics, human anatomy, understanding and just overall prompt comprehension for more specific physics related or motion related tasks. Squirrel flying down the water slide in the pool. That looks great with the water physics. We've got the hip hop dancers in sync together. A little bit of morphing going on here, but overall impressive figure skaters. All these prompts are typically very, very difficult for AI video models to replicate. And VideoJam just comes out here swinging fixing this problem. It seems like a great encompassing solution. Apparently it can be pretty easily applied to traditional video models we already have. So it seems like a no brainer. I love the turtle pushing itself on the skateboard. That's cute. The guy eating an apple. Oh man, so many great examples. Yeah, this is gonna be linked down below. Don't worry guys. Dog with aviator glasses. That's cool. Slow motion of the the water drop in a crown formation. That's awesome. Meta please, the love for you to open source Video jam here. I would be so interested to see how it performs with lightweight models like LTX Video. You know, we could have very high quality, good physics, good anatomy, local video models running on consumer grade hardware. No problem. That would Be amazing. But also imagine applying it to more advanced models like Huanion Video. What kind of quality echelon could we potentially reach? This is awesome research. Meta. Please open source this for the love of the community. Next up we've got just more insanity. This is a real time AI generated avatar that will essentially just put you in a zoom meeting for as long as you want. This would have been very useful during COVID times, I imagine. I wanted to show this off though, because I thought it was totally mind boggling.

As you can see, I'm currently participating in a zoom call, but what if I tell you that I'm actually not in front of the camera right now and this is an AI generated live stream that moves in sync with my voice Pickle generates real time video of talking you based on your audio inputs which can be streamed directly to zoom as you can.

Just insane. I mean you can barely tell, right? Especially over like a low quality zoom call. Like no one's gonna bat an eye really. And like he even kind of moves his eyebrows in some scenarios. He's definitely moving his head and talking. The only problem is like if you do this for long enough, obviously you're expecting people to do very human things like oh, I gotta itch my nose or adjust my hair, adjust my glasses. That kind of a thing might give it a little bit of a giveaway, but it's pretty cool nonetheless. Man. This is brought to us by Hamptonism, who says we are cooked. Chat Are we cooked? That's the question. We're either cooked or we're headed to the stars. I think that we're just two scenarios. I'd like to head to the stars personally, but maybe that's just me. Alright, now we're moving into another massive drop. Pika Labs adds references to videos. Could be any object or character from an image and that will seamlessly be added into the video and become a part of the scene that you already recorded with your camera. They're calling it Pika Editions. It keeps the original footage untouched while adding whatever you want in the video. It's really crazy. Especially in this example. We've got the giant rabbit sitting inside of the shower here. And the craziest part is that we've got the flashlight and that's a real flashlight in the original video. And it's able to replicate what it would look like if we pointed that flashlight at a massive rabbit. I mean, we've got overexposure on parts of the fur here. The shadows and the lighting looks so realistic. Rec this with traditional vfx, guys, would take so much effort in time and a lot of prep. Now you could just do it in like a matter of minutes by recording something on your phone and putting it through an AI. What kind of sorcery are we working on here, guys? We could just conjure things. We've got more examples here. A giant tiger is attacking this guy. Again, you know, you're just recording stuff from your iPhone. Oh no, the tiger stepped on you like so wild. I think the proportions of the tiger here are a little off and obviously it's kind of maybe a little bit weirdness in terms of the movement. Again, these are gonna be cherry picked examples, but it's, it's cool nonetheless. This is another one of those things, guys, where it's like, alright, I think we should probably cover this in a longer video because this looks like a total game changer and it looks like so much fun to use. We've got Mu Dang poking out of a little glass of like coffee or something and he seems to be enjoying himself, having a good time. I think the object references here, I mean, they're not perfect in every scenario, but good enough to realize that it's like a little cute baby hippo, right? Eagle landing the back of a child here. You can definitely tell it has to do some overwriting of the original footage because we're losing quite a lot of detail in the hair, in the skin. The face looks slightly different. People on Facebook are screwed though, because they're gonna think stuff like this is real. They're gonna, you know, the boomers on Facebook, they're gonna watch this and be like, oh yeah, that's, that's absolutely 110% legit. So, you know, maybe we're cooked in that sense. By the way, I'm pretty sure if a bald eagle did land on your shoulders like that, his razor sharp claws would not be a good time for this poor little girl. That's all I'm saying. Still though, that we're able to do this, I think this one looks a little bit less realistic than maybe some of the other examples, but mind blowing nevertheless. Instant vfx, guys. Instant. Got a polar bear living in my fridge. That's fine and dandy. That's a cute polar bear, guys. Again, this one is like so scary real looking that people are just gonna assume it's legit. They're gonna be like, oh, you are a horrible person for keeping your polar bear in a fridge like that. It's AI okay, It's not real. Like I said though, I mean, this is so awesome. It's so cool. I'm going to have to spend some some real time with this. It also works with animation though, and stuff like that, which is interesting. So I assume this is originally a scene from the Simpsons and then they added this in as an image reference and was able to put Sonic and knuckles just gaming in the Simpsons household. The original video was an empty sofa according to Faux Fur. Next up, we've got more open source stuff. This is an open source speech to speech translation model that can run directly on an iPhone, which is super cool. We're going to want technology like speech to speech translation to be. Well first off, not only open source, which benefits everybody, iPhone users and Android users, but also the ability to run on your phone is a huge deal because you can't just be lugging a massive GPU PC around everywhere just for translation. It needs to be right on your phone. The fact that they were able to achieve this is pretty crazy. Not going to spend too, too much time diving into this one, but it's super cool and stuff like this often flies under the radar. Guys, open source is moving all the time. New models are coming out all the time. They're built off of, they're improved, they're evolved. The open source AI train can be stopped. Next up, Korea AI releases. Korea chat. It's a very seamless way to generate images with AI. You can upload plenty of references and just chat with the model and it will make the changes that you describe and ask for. We've seen similar technology before, but at least from this little demo video it seems like this is more accurate than some of the previous examples of this technology we've seen and the quality seems to be pretty outstanding. Again, I want to cover this in a deeper video. Don't have the time to fully encompass this one today. Be made aware. This is super awesome. Let me know if you tried it out in the comments as well. If you look at the specifics in the demo video they provide, we're uploading a very specific face style here, but we want it on a cat. Crea is able to replicate it pretty accurately though. I mean these two examples, I mean they've got the blue star, but it's not really in the right spot. This middle one nails it in my opinion. Then we add another photo for lighting reference and it does again a pretty decent job. Star Accurate accuracy is difficult. Props to Korea for actually showing this off with a complex prompt. This is the image that seems to nail it at Least in this and finally asked to put it in the Matrix and it does an okay job where you could give it pink hair. Alright, I'm gonna leave you guys with this one Boring reality Laura AKA a fine tune for Huan Yan video. This fine tune specifically does not give you cinematic shots, but it gives something that looks like it came from an average everyday person's iPhone or like a Snapchat story. Who knows what it's meant to look realistic in the sense that anyone could have recorded it. There's no digital altering really occurring here, but actually it's all AI generated, so it's all digital altering. And that's. That's the cool part, I suppose. I think you could do some really fun stuff with this. Again, the boomers on Facebook are going to have a time trying to keep up with some of this AI technology. I think they are already getting stumped by stuff that looks very obviously fake from the examples I've seen. And this is not going to be any better for them. But it's cool for us. Like I really do think there's creative potential here. There's something to be said about having more low quality footage or more amateur esque footage that gives it a sense of authenticity or a different feel creative wise. So this is cool. And again guys, open source. There would be no Lauras in general if one on video wasn't open source in the beginning. So this is just more proof here. Open source is taking the lead. Man, I like some of these comments. We're so cooked. Why is it British core? Oh, it's so much creepier. Quick recap. What did we learn today? AI is is moving at an insane rate. Of course. We might be on on the hard takeoff launch here. Guys, buckle your seat belts. There ain't no getting off this ride, man. It's exciting. It's also bewildering. It's a little freaky. We have learned to conjure magic with computers and it is not stopping anytime soon. But yes, thank you so much everyone for watching today's video. Everything's linked down in the description below. If you want to dive into any of this in a deeper sense. Again guys, let me know if you want to see a full video on any of these topics. There are definitely a few. I think that could cut it for a full dedicated deep dive. Goodbye.