Limitless: An AI Podcast - AI's Wild Week: Everything You Need To Know in 36 Minutes

Episode Date: August 8, 2025

36-Minute AI Super-Recap:This lightning-speed episode breaks down the craziest week in AI history. We start with the mixed reactions to the GPT-5 launch, then dive into Google’s Genie-3 world-builder that turns a single prompt into a playable game environment. Next, we demo ElevenLabs Music, cranking out studio-quality tracks in seconds, and spotlight Chai-2, the antibody-design model slashing cancer research timelines. We wrap with Meta’s $29 B bet on a new AI data center and Midjourney’s cinematic HD video mode—proving the innovation firehose shows no sign of slowing. Watch to get everything you need to know in just 36 minutes.------🌌 LIMITLESS HQ: LISTEN & FOLLOW HERE ⬇️https://limitless.bankless.com/https://x.com/LimitlessFT------TIMESTAMPS00:00 GPT-5 Day After03:17 Genie 3 Is INCREDIBLE06:53 Demo's13:03 AI Music Is Now... GREAT?19:39 How We'll Cure Cancer24:38 Zucc Goes Into Debt29:34 Amazing Video Generation------RESOURCESJosh: https://x.com/Josh_KaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠

There are years where weeks happen and there are weeks where years happen. And this year of a week was insane. I would argue this was one of the most exciting weeks in the history of AI. There was so many releases that we figured we need to have our own separate episode just to cover these. So the next 30 or so minutes is going to be a speed run. Of all of the most noteworthy, important things that you probably missed, if you stepped away from your computer for more than 10 minutes this week. We want to start with GPT5. We're going to go to Jeannie.

We have 11 labs where you can make. music with a single prompt. There's a lot of really interesting things. A health care thing where people are solving cancer now with these models. It's pretty incredible. Let's start with GPT5. Yesterday was the release. It was a huge day. It had a lot of opinions. Good and bad. Ejaz, you have some takes on this after using it for maybe a couple of hours. I mean, we've recorded our live stream episode yesterday, but now you've had some time to use the prompts to actually try out the model. What do you think? What are your day after review on Chad GPT? Yeah, it's still the same. I was pretty underwhelmed by the overall launch. These things,

are meant to be magical and it just felt more like a morgue when they were announced in GPD5 to the point that they were writing a literal eulogy as a demo, right? So I was thinking, okay, maybe less than 24 hours later, everyone's got a bit of a taste of things. I tried it out. If I'm being honest, I didn't really notice the difference too much between O3 Pro and GPT5. Maybe that's controversial. I don't know. But I was like, okay, maybe people have some kind of like fun demos out there. Josh, I just found one. And I, I thought it was worth sharing. So it was from Sam Altman himself.

And I've got the tweet up here. He goes, when you get access to GPT5, try a message like, use Beatbox to make a sick beat to celebrate GPT5. And basically it's this prompt which loads up this fun little gizmo in chat GBT, which is called Beatbox. And you can compose a track. We'll get a little bit of an excerpt here, actually. So as you can see, it's pretty basic. but what I found, if I'm looking for a silver lining here,

what I found cool is that I didn't realize you could pull up different gizmos aside from the command line interface when you're coding stuff. I didn't realize you could pull up other gadgets. My one criticism for this though is I didn't, who would know about this if Sam hadn't tweeted about it, right?

And has the people that haven't seen Sam Sweet know that this even exists? It's like when Apple came out with garage band and as if they hadn't told you that that app had existed, no one would ever use it, right? So I'm still pretty much underwhelmed. Do you feel the same? Okay, yep, same thing. No notes. Presentation was lacking. I think that's what caused a lot of the problem. I think my big takeaway now is, hey, 700 million of their monthly active users just got a huge upgrade from GPT40 to GPT5. It's going to feel massive. That is a ton

of potential intelligence that is now out in the world for people to use. So that seems like a win. I hope he learned a lesson from the Apple team and that delivery of these presentations is really important. But, EJJAS, if you think this music sounds good. We have a topic a couple coming up in a few minutes that is going to absolutely blow your mind, the 11 Labs music generation. But before that, I want to get into what I personally think is the most exciting tech of the week, which is the Google's Genie 3 model they release for the world. It's incredible. Genea 3 is basically a world builder in a prompt. If you want to create a video game that you could actually walk around in, I mean, you type something into a text box and it generates it. And Ejazz, you have a video loaded up here.

Can you kind of show people what this means, what's going on here? Each one of these is an interactive environment generated by Genie 3, a new frontier for world models. With Genie 3, you can use natural language to generate a variety of worlds and explore them interactively, all with a single text prompt. This is cool. So what you're looking at here may appear like an AI generated video similar to like Google V-O3, but it's actually much different, right, Josh?

because what you're looking at, you know, the perspectives of panning left and paning right that you're seeing on the screen here, you're probably wondering, wait, why are there arrows there like a keyboard, like a game? It's because AI is generating in real time what you see and what you interact with and things that you do. And that is distinctively different from a video model, right, Josh? Yeah. So with a video model, like V-O-3 that we're used to, when you type a prompt, it generates you a two-dimensional video. It looks like you're watching a YouTube video. But with Genie 3, it's a video. a little bit different. I mean, you could see here on the demo, you typed a prompt and you had generated an entire three-dimensional world that you could actually walk into, navigate around, view things. And one of the most interesting things that we'll get into in a second is that it has

what we're seeing on screen now, world memory. And it can remember the actions that you take on a world. And you can actually go back and revisit it with those actions. So we're seeing this example of a painter placing paint on the wall. And you'll notice that the actions are actually persistent. So not only is it generating a real world for you to navigate, it's generating a world for you to actually interact with and then save the state of those interactions. And to me, this, this like totally blew my mind because I love video games. I've been playing video games forever. And this very much feels like, oh my God, this is how every video game in the world is going to be made. If you can generate an entire world from a prompt right now, imagine what Jeannie 4

is going to look like, where you can really get some truly high fidelity. You can get all of the the real world building physics aspects that are maybe lacking a little bit here. And to me, this feels like the future of gaming. And this was, this was so cool because I'd never seen anything like this. I think with a lot of the things we're going to talk about this week, we've kind of seen them. They're just iterations making them better. This very much feels like a zero to one moment where we now can go from words to real life, fully immersive, metaverse type world. And that seems like a really big deal. I think if I were to condense what's so valuable about this, model, there's a few things. One, video and images in general is just way more complicated to

figure out from an AI model perspective than kind of like the LLMs and words that we've seen so far. So as you said, like as a version three for Jeannie, this is pretty impressive. The other thing is video production or to the scale that we're seeing on the screen here typically costs tens of millions of dollars, minimum, right? Hollywood studios, VFX studios are worth hundreds of millions of dollars. If you remember like the Star Wars saga, Lucas films, they're worth like well over a billion dollars they got acquired by Disney. And now you have a tool that you can access for what, a hundred bucks a month and create very similar grade, if not better grade videos in the coming months. That is just insane. And honestly, some of the examples that I've been seeing,

Josh, are pretty crazy. So I want to have any really fun ones to highlight. Yeah, yeah. Okay. So I have some personal ones, which I was pretty excited about. Okay. So what you're staring at here, is this like 2D, rather, yeah, 2Ds kind of like old school gaming night that is walking through this kind of world, looking around, you've got birds flying in the background, and you might be thinking, well, each has, this is so old school. Like, why would I care about this? Well, what's amazing is this was an image generated, I think, on chat GPT last week that went viral for its aesthetic. And people were like, wow, I miss this old school feel. And literally a week later, less than seven days later, we have this new tool that can turn any image into not just a video, but an interactive

video where you could move around like a game. And someone thought, let me put this viral image in this model and kind of let him walk around, let him do things, right? And we had a follow-up post that came after this being like, don't forget that you can kind of turn around literally back on yourself, you know, this is a POV kind of video and see what it looks like. So I think and this, he fully looks around and he's like, okay, yeah, look, the bridge, you can go back over it. There's loads of cool things, but there was this really

meta video that I saw, Josh. I don't know if you, you spotted this one. Oh, this is weird. This is crazy. Okay, so what you're looking at right now is a simulation within a simulation. So what he's doing is he's filming himself,

let me rewind him. What he's doing is he's filming himself exploring this interactive video in Genie on his TV, on his laptop and then on his TV. But then he's recorded a video of him doing that and used it as a simulator in GD3. So now he's walking outside of the house that he's literally sitting in

and exploring this AI simulated world. That just blew my mind. And remember your earlier point, Josh, that it's persistent memory. So it remembers that he's in this house and can go back and see it. So the use cases for this, I'm most impressed by

is like think about what this looks like from an education side of things, right? When you're explaining some kind of like abstract concept in physics or stuff, I don't know about you, but I don't want to read this in a textbook, Josh. But I would be more than down to jump into this simulation

via VR head goggles or whatever and explore this. It's super cool. Yeah, it's amazing how quickly in high fidelity you're able to generate these things. And it's also like, look at that. That is the person that you place yourself into this. It makes it really easy to believe

like, hey, this is totally all just a simulation. Like, if we're this good already. Did you see this drone shot, Josh? No, but as a drone pilot, I would love to. Oh, wow. It's an FPV drone too. So it's crazy. So what this guy did was he took, I think it was like a three second clip

of an actual drone shot that he took on his drone and was like, I kind of want to see what this genie model creates, whether it identifies the location that I was in from this three second shot, and whether it'll allow me to explore the rest of that environment that I was trekking over and filming. And it did exactly that. I don't know whether this is geologically accurate, but it looks so cool. So he took like a three second drone clip and now he's just traveling around exploring it. Super cool. Yeah, that seems like fun. And also when you're learning to fly FPV drones, there's this big process of flying in simulators where you have to spend a lot of time in a simulator

before you can go out into the real world. And if simulators looked this good, and you could generate your own simulators to fly around it, similar to what he's doing, that's a really fun way of learning how to fly FPV. So just another use case, that's pretty cool. Okay, another one. What do we got here?

Okay, so this one demonstrates how accurate the physics is in this model. And that might sound lame to you, but it's a super important one. Look at the reflections of the puddle that you're looking at right now, right? And then you're always wondering, well, okay, maybe I can look left and right,

but like, can I see my own feet? Am I there? Do the steps work? And you see the foot going forward. That is just insane. Look at the splash on the puddle. Like, again, I must reiterate, you have AAA gaming studios that haven't come out for a game or with a game in 10 years.

I'm talking about GTA 5 over here. That spends hundreds of millions of dollars to try and create these games. And now you have a tool that can do that in seconds. It blows my mind. Yeah, it's pretty remarkable. I was actually thinking about GTA 6 this morning. because they finally announced the release date next year. And had they pushed this release date out another year or two, there's a high probability that we'd be getting GTA6, but just

artificially generated, where we wouldn't have had to even wait until the real thing because of how good this is. And it's probably going to be the very last blockbuster AAA game that is built without a majority of the assets being generated by AI. Because, I mean, these virtual physics engines are so good. And the fact that you do this with one prompt is like, it's pretty remarkable. So Genie 3 is pretty incredible. GROC image gen, which we did an episode on earlier this week is actually doing very, very well, too. This theme of image generation, visual generation, it's hot. Grock image usage is growing 50% per day, which seems pretty outrageous. So if you haven't tried, go check that out. Another noteworthy piece of news this week that you may have

missed is that Anthropic actually dropped a brand new coding model. It is called Claude Opus, 4.1. It is better than 4.0, believe it or not. And it's funny. Dude, the child crime on this is hilarious. The chart's hilarious. Well, no, so, so opening Yeah, I was exaggerating. I respect Anthropic for showing it as it is. Your model is 2.2% better. That's great. Show me how it is.

All right. Four per one is out. If you're a developer, go use it. It is 2% better. And this pretty funny chart and accurately shows that. For the nerds out there, if you're wondering how GBT5 compares to this in terms of coding, it is 0.4% better than this new Claude model. Just to let you know, just to keep you on your toes.

if you're looking for like the best coding model, it's technically still GBT5. There you go. Okay. So we'll have a debate whether we're hitting a wall or not in another episode, but that's new coding model update. There's another really cool update that I want to get into. And this one is actually pretty interesting. We're going to spend some time here, which is the new 11 lab software that they release this week. It allows you to generate music.

And I was like, okay, sure, AI music. Great. I've heard plenty of AI music. This is good. This is good music. It's actually really good. Yeah. Walk us through it.

For some context, 11 labs is. is famous for being the text-to-voice model. So you can type a text and it can read or orate whatever you've written or create a script from its own and then orate that in so many different voices. You can also do your voice as well. So you could literally speak into a microphone for three seconds, I think. That's all it takes. And now you have an entire model that can speak and sound like Josh, that can speak and sound like EJA.

So I don't know why I'm doing these podcasts when I could just, you know, just do that and combine that with a genius. or a V-O-3 and then I can just sit on my sofa all day. But yes, we have 11 music which was unexpected, to be honest, because this is not kind of like a domain that they've dealt in. And basically

you can write any kind of prompt to create any kind of studio soundtrack. And I want to emphasize the studio part of this because there's been a lot of AI music generating products, but they kind of suck. We just played a demo

of GPT5 making some beatboxing and it sounded pretty lame to be honest. But these ones are pretty good and you can see the fidelity that it goes into. Like you're listening to some deep jazz. All of this is AI generated right now. Look how smooth. Look how buttery that is, right? That's good.

And you can have this in so many different languages as well. But more so, what if you are an artist that wants to work with AI but you don't necessarily want it to be 100% AI. Well, you can take your vocals, your recording, or even just a soundtrack of an instrument that you play at a cool beat or tune, and you could throw it right into here and create and compose whatever kind of album, studio, or song that you want to do. Publishing rights are all yours. I don't know how they split profits amongst AI, but who cares?

My question is, you know, what's the traction looking like on something like this, right? And they answered that, I think literally two hours later after they posted this, which was over 111,000 studio quality tracks have been created with this. Since it was released 24, I missed that part, since it was released 24 hours ago. That is insane. So all of this from a single prompt, Josh, I know you have a lot of friends that work in the music world. Are they worried about this? You know, what's your take?

I don't know, but I would be. I mean, that is impressive adoption rates because it's actually so good. Could you go back to the last hab and can we listen just a little bit more? Because that one sample, it's shockingly good. Maybe just click around and see if there's any other. Oh, yeah. So this is like a female vocal. Okay, we have kind of like.

And the vocals are interesting, right? Because, I mean, AI is generating the lyrics. What do you have here? That's an Afro beat. That's insane. That's some hip-up. Okay. Yeah, that's amazing, because you have to imagine, I mean, what, those lyrics are

are AI generated. They're read by an AI, they're sung by an AI. The whole thing is just AI top to bottom. And it sounds pretty good. I think the interesting phenomenon that's happening now, since mostly TikTok is the TikTokification of music, like a lot of people who are producers, they kind of just generate, or not generate, but they produce music that's made to be listened to on iPhones. They sacrifice a little bit of quality to make sure it sounds great on an iPhone or an AirPods. And what you're

getting here is, I mean, from my novice ear is parody with that where these songs sound really good. And if you're a creator or if you're someone like us who wants to use music in their content, but often at times has problems with copyright, this is a really great opportunity to just generate your own and never worry about copyright again. That's a really good point. The other thing is, Josh, I can't, I don't know you, but I can't sing, mate. But I would love to have the power. Exactly. I would love to have the power to kind of like play around with this.

And again, I mentioned garage band like twice on this episode already, but like I'm going to bring it up again. I remember playing with that app the first time he came out on the Apple store and being like, oh my God, I can now become a musician. I didn't realize my potential then. But like maybe this will, maybe this will change it. And the last bit I want to show is this quick clip with an interview with. Mattie, who is the CEO and founder of 11 Labs,

who were actually getting on the show in about two weeks. So we're going to interview and watch out for that episode. Yeah, it's going to be awesome. And he talks about why it's, oh yeah. And he talks about why it's super important. Exactly. One of the very special things about the release of the model is that it's both extremely high quality, one of the highest qualities in the market,

while it's also fully licensed and built in collaboration with the labels, with the artists, with the publishers. Like you mentioned, we already are very happy to be working with Maryland and Cobalt frequently referred as the fourth majors, both because they represent respectively 30,000 independent labels on the Maryland side and over 25,000 songwriters on the Cobalt side. So really that gives us such a high ability to create extremely high quality music. Of course. So what he's mentioning there in that short clip is two main things. One, the music that you make is fully licensed. So you can earn royalties, sell it studio producers, labels or whatever that might be.

But what's also important is they have a direct partnership with these two companies, CoBOL, and I forgot what the other one is, which represent independent artists. So it's probably never been easier for you to get discovered as an aspiring musician. If you're listening to this show, with something like that. So it's making it more democratically accessible to budding upcoming artists, which I thought was really cool. That's really exciting and really empowering, right? If you're an artist and you can sing or if you're a producer that doesn't have a vocalist,

you can kind of fill in all of the blanks of your own production world and actually generate really high fidelity, high quality music. So relative to the chat GPT example we had of that little beat maker, this is a pretty big step function improvement. Yeah. This is impressive tech. I'm excited to see where this goes. But this is not the end of our exciting news.

We have another discovery in the health fields. Walk us through this. What is Chai? Okay. So Chai is this brand new AI health. startup, which just raised a crazy $70 million round, which actually doesn't sound too crazy given the billion-dollar raises that people are doing. But they raised from the likes of OpenAI, Anthropic, and Menlo Ventures. So, you know, very small fish type of deal, obviously being sarcastic

here, but you might be asking, well, what can they do? Well, this new startup figured out a really cool thing. They have this AI model that can basically create designer antibodies that could you be used theoretically to treat any form of disease, which includes cancer, skin diseases, whatever it might be. What's awesome about this is it has a 15% success rate compared to the 0.1% success rate that all the other AI health models have when it comes to this. And they actually put this to the test with a study that they have. I think he has a tweet over here where, yeah, over here, he goes, Try to Deliver's DeNovo antibody design with a 15% hit rate. But one group spent three plus years and five million dollars on a tough target. But with

Chai II, we delivered an experimentally validated binder, basically a theoretical solution or cure, in just two weeks. And where my mind jumped from this was, oh my God, you're going to save hundreds of millions of dollars, but who cares about that? You're going to save tons of lives. You're going to improve the average health of an individual at the cost of a GPT prompt. And you get an answer in like, I don't know, a couple of minutes or even an hour. Even if it's a few days, it's even marginally better than three years and five million dollars, which is just insane economically, right? So I'm starting to understand why Open AI spent like 15 minutes of their live stream yesterday focusing on health. I'm starting to understand a lot why Jensen Huang is

investing a lot of money and time on chips that are devoted towards health care. Josh, what do you think of this? We're kind of, I mean, throughout this episode, we were planting these seeds of a really optimistic and pretty wild future, where we have world building in a prompt and you could create your own metaverse. And now we have genetic mutation, genetic modification and understanding of the DNA better so that we can actually manipulate the human body

and all the cells in our body. And to me, what we're getting from AI is just this deeper core understanding of the world and then a really empowering ability to manipulate it. So now that these DNA sequencing AIs are able to help us understand things, well, now we can manipulate it and cure diseases and cure cancer. And now that we have a deep understanding of physics and world building, we can emulate that and we can generate our own.

And AI is just kind of helping us go deeper and deeper on these topics to get better understanding and then empowering us to make these crazy tools on top of it that could actually change things. So there's a world in which we are curing all of these diseases. This is a role of which we don't even have to worry about diseases. And there's a world where you can create an entire world inside of your virtual reality goggles and do that all in a prompt. And I think the theme of this week is like very forward looking and really, really optimistic and exciting. And this, yeah, this leaves me excited to see. Yeah, we're entering an era of AI science.

And what I like love about that is it's going to affect so many people who have no idea how any of this AI stuff works, who don't have GPT subscribers. who don't have access to all these fun coding models or maybe have no use for it. Real impact on real people, I think, is going to be totally awesome. That's one of the cool things about the GPT5 release this week is for a lot of people, they don't really know what's going on. They don't follow the show. They don't follow AI. And they just use this thing called chat GPT.

Not realizing that the model they're using is actually like a couple years old. It's pretty outdated. And just today, they woke up today and they went into their chat GPT app. and they don't even notice a difference, but suddenly it's a lot better. And then it just has this impact on their life. And there are these moments in time where the average person kind of catches up to the frontier. This is one of them in the model space, but I assume we're going to get one of those fairly soon in the healthcare space too, where there's a lot of healthcare stuff happening on the frontier, but none of it's really readily available to the rest of the population. And one of these days,

one of these companies, is going to release a product that does that. And it will suddenly catch the world up to the frontier. And those moments, like the days like GPT-5, that the days where this company like Chia will announce, or Chai, will announce that they've actually made this public and you can go and get cures for whatever problem you're having. Those are the days that are going to feel really big to the rest of the world. Because for us, it's another week in this world. But for them, it's like, oh my God, wait, you could do all of this stuff. How did this happen overnight? And of course, it never happens overnight, but that's the appearance. The other breaking news over the last 24 hours is Zuck or Matter has secured $29 billion.

in private credit lending. So he's borrowed this money from a private lender to fund their new AI data center. And we're spoken about data centers on a few episodes. Josh, we actually had a dedicated episode. I think, was it this week? I'm losing track. I think it was maybe last week on Elon Musk's Colossus 2,

which is being used to train the next generation of GROC models. And so these big AI companies are in a race to secure as much compute as they can to train the best models. And Zuck is no different from this. But this requires a lot of money. I saw a crazy stat, which was META has spent, I think, $40 billion already this year on training compute. And they aim to spend another $150 billion over the next year. That is just insane.

And so you're probably wondering, do these guys have a big enough war chest to do that? And the answer is actually no. Like, they have technically enough money, but the capex spend on this far exceeding. the amount of money that they're going to make over this time whilst they're investing in the core model infrastructure itself, which means that they need to borrow this massive amount of money and strike these deals with private lenders to do that. And actually, I think ex-AI, Elon Musk, did the same for Colossus 2, I believe,

now that I'm kind of remembering it, where they raised a similar amount, $20 billion, to build this new Colossus 2 data center. So the point I'm trying to make here is, I think the numbers are going to get much bigger. and I think that our feeble small minds aren't quite adjusted to what this is eventually going to become. I think trillion dollar valuations are going to become way more abundant in this world. And we are only now starting to see the sunk cost of how much it's going to take to build these models.

That was how I was going to start this section is like, I don't even know if $30 billion is a lot anymore. What is at $300 million an employee, that's what? You cover the payroll for 100 people for a year. It's like the scale of this conversation. has gotten so grand, but one thing that I do admire about Zuck is, I mean, I assume the founder-led companies are always the most interesting because they have the most, those founders have the most ability to make crazy, seemingly irrational decisions because they have the total control of the company to do so. And Zuck is literally betting the entire future on the company on

building these AI models. They are using their entire balance sheet plus plus and are betting fully on the fact, not only on the fact that AI will be here to stay and will generate a ton of money, but also that having their own unique AI models will yield enough value to pay this back. So you're seeing these massive bets that it seems like this has to work, right? Like, clearly there isn't a scaling wall. Clearly, he believes that $29 billion of additional capital will yield better models. But we're getting to a point now. We're just like, what is the value? It's almost hard to imagine the value unlock because it's so obscure and it happens in so many places. Yeah, like when he says like, oh, yeah, we're going to, like, we'll generate $10 trillion in revenue.

You're like what? How? That's like, no one's even worth like half of that now as a company. So it really, it's going to be pretty amazing watching this play out and watching the scale of this unfold and, and the returns start to come in because like the numbers are really getting massive. I appreciate the big bets, right? Like Zop did this when he acquired an instant. Instagram and WhatsApp at times which people thought he was insane to be spending billions of dollars on them. And now each of those companies are worth hundreds of billions of dollars, individually on their own.

And I'm thinking about Elon buying Twitter for, what was it, $20 billion? I forget the number now, right? It was like $42. Yeah, but, okay, just $42 billion. Yeah, but again, to take that massive risk, the stage and level of foundership, like, you know, if you're worth hundred billion dollars, just retire already, right? But these guys are making massive gargantum leaps. It's just very commendable. And I'm, just to describe this, I'm excited to see where this goes. I have no idea whether it pays off or not. Zuck spent a similar amount of money,

actually maybe a lot less to rebrand the entire Facebook company to Meta and bet on the metaverse. And that kind of didn't pan out kind of how we wanted to. But he's going again. And I respect that. Yeah, respect and admire. I think the more you know, the less you need to diversify. And I would imagine that Zuck is really deeply in the weeds in all this stuff and has been for the last decade. And he is very well aware of the risk and very well way of the upside. And the fact that all these people are deeming it a risk to take, well, sounds like that's the right risk to take. And I'm here for it. Is that everything for this crazy, wacky, wild week? No, I just wanted to show you this new film that came out, Josh. I want to give a little bit of a trailer.

Yeah, yeah. Let me hide the mid journey thing. No, it's just a movie, Josh. You know, this is These are real people. These are real actresses. And this is a plot. No, I'm kidding. Of course, this is AI generating. What you're looking at right here is completely 100% AI generated. And the reason why I'm bringing this up, this is Mid Journey's HD video mode.

Do you remember Mid Journey at the start? Josh? It was cool, but it was so lame. Dude, it was so lame. It was so inaccurate. And now I feel like I'm watching an A24 film right here. This is like crazy. aesthetical quality. Look at how the water ripples. Look at the expressions. Look at how the light

glistens and glows off these people's skin. It is insane. Again, you would need to pay tens to hundreds of millions of dollars to hire these actors, get all the crew and equipment out there to film a scene like this. And now you can do it for what? An $80 subscription? Just insane. And the rate of progress that these models are improving at is insane. But don't worry. It's not just about humans, Josh. We also have the animation studio. Oh boy, this looks like a Pixar movie. That is also in danger. It's like watching Toy Story. Look at this. Well, not Toy Story, but like, you know, a new Disney or Pixar film. Totally. Like, what you're looking at is four separate scenes from the same type of movie or storyline and they're in slightly different grainy quality,

right? You've got like the high fidelity animated show here. You've got a slightly grainy and more cartoon-like, you know, from Into the Spider-verse. If you're interested or watch that, saga before, just hugely interactive pieces of content that I think is going to make it so easy for the next world-class director to appear behind a keyboard, to be some kind of like 15-year-old teenagers sitting in his parents' basement, just super cool. Yeah, for animators, this is such a game changer. I forget the exact timeline, but the movie spirited away, which is like this very famous, popular anime movie. I understand that it took many years just to develop some scenes, because each frame of each sequence had to be fully hand drawn and animated.

And these things took forever. And now what we're seeing is like all of that time is compressed into a single prompt generated in a couple of seconds. And like you said, the rate of improvement is incredible. And one of the things that I believe is underrated that I could admire as someone who's played video games for a long time is the physics engine of these videos and of this world generation that we saw in Genie 3 is remarkable. So much of the complexity in video games is simulating physics in a way that's lightweight enough for a console to play.

So when you're generating these worlds, say that you're in Grand Theft Auto, and they're basically simulating all of Los Angeles. And to simulate all of Los Angeles on a little PlayStation or an Xbox console is incredibly difficult because it consumes a lot of resources. There's a lot of textures. There's a lot of physics, gravity, light reflections, sun. These things are really challenging to emulate. And what's interesting about these video generation models is that they're able to emulate that physics without any of the compute costs. And I think that's a really interesting thing, is it abstracts away a lot of the heavyweight processing power and just allows you to generate these worlds in a very lightweight way that's accurate because it's trained on the physics of the real world. So it looks identical.

And that to me feels like a really big breakthrough because now that unlocks this whole extra realm where you can literally run a world on your PlayStation now. if it is that lightweight and it doesn't require all that compute power. Oh, yeah, here's a one Google front. I just looked up the value of Unreal Engine, which is a gaming engine, which basically does all the things you just described. Josh, but in the old-school traditional non-AI method, it's owned by Epic Games, and Epic Games itself is worth $32 billion. And Unreal Engine is a major reason why it's worth that much.

And now you have an app, again, from a system. subscription service that can generate very similar quality aesthetics. You have Genie 3 accessible to you that will take that video and allow it to become this whole entirely simulated gaming world, very similar to the games that Epic Games makes. So you've got $32 billion and then maybe the cost of two monthly subscriptions, maybe $150 a month. And you have access to all the same tools that Unreal Engine has to build these things. Pretty old.

Pretty good. Pretty good. And again, it's not, it's not Unreal Engine, but come Genie 4, Genie 5 for the next mid-jury iteration, like, it is very, there's a very clear path to doing so. And the rate of acceleration in which we are going at to reach that point is, is insanely high. So, is that everything? Is that a deep breath now? Are we done? So we're deep breath, Josh. We are done. For now. For now. I don't know what's going to happen the next to the last of a house, but we're done. This week was insane. There was, there's so much incredible technology. So hopefully this serves as a nice little overview to help everyone who stepped away get caught up. We have been in the trenches all week. It feels like every week there is more and

there more and more and they're getting impressive. And even where it seems like we weren't accelerating as fast as we normally do with GPT5, well, the other companies picked up the slack and they pushed that frontier forward even further. So what's nice is we're not dependent on any company. There's a million companies all pushing forward, all doing interesting things. And we will be here to cover it every day, every week as we go. Oh, yeah. Covering all these great topics. So I guess I guess we'll wrap it here.

Thank you for listening. Thank you for tuning in for this insane week. We posted a lot of episodes this week. We covered a lot of ground. But if you listen to all of them, you should be fully up to date on everything that you need to know in the world of AI. We are going to go enjoy our weekend, maybe touch some grass, see some sunlight. If you enjoyed any of the episodes this week or want to fill your friends in on GPT5

or any of the cool topics, please share this with them. It goes a long way when you share the show. We are doing well. We are actually on the Spotify. chart. It's fun fact. We are one spot away from flipping the Open AI podcast. So if you share the episode, you are helping us flip Open AI, and that's just like a pretty cool thing that I'd like to have an accomplishment. I'm sure you just feels the same way. So please, we very much appreciate the support and sticking with us listening to the end of these episodes. It means a lot. So thank you

for watching. I hope you can take a deep breath after listening to this too. You're all caught up, and we'll see you next week with a bunch of new news. Thanks, folks.