Latest / In the Loop AI / AI's Wild Week: Everything You Need To Know in 36 Minutes

Transcript

AI's Wild Week: Everything You Need To Know in 36 Minutes · In the Loop AI

• Josh: [0:03] 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 um 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 the most noteworthy important things that you probably missed if Josh: [0:25] you stepped away from your computer for more than 10 minutes this week. We want to start with GPT-5. We're going to go to Genie. We have 11 labs where you can make music with a single prompt. There's a lot of really interesting things. A healthcare thing where people are solving cancer now with these models. It's pretty incredible. Let's start with GPT-5. 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 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 ChatGPT? Ejaaz: [0:58] 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 announcing GPT-5, 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 GPT-5. 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 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 GPT-5, Ejaaz: [1:40] try a message like use beatbox to make a sick beat to celebrate GPT-5. And basically, it's this prompt which loads up this fun little gizmo in chat gpt which is called beatbox and you can compose a track we'll get a little bit of an excerpt here actually Ejaaz: [2:03] 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's suite know that this even exists? It's like when Apple came out with GarageBand, And if 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? Josh: [2:42] OK, 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 GPT-4.0 to GPT-5. 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 Josh: [3:03] that delivery of these presentations is really important. But you guys, 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 released for the world. It's incredible. Genie 3 is basically a world builder in a prompt. If you want to create a video game that you can actually walk around in i mean you type something into a text box and it generates it and you guys you have a video loaded up here can you kind of show people what 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. Ejaaz: [3:54] Is cool so what you're looking at here may appear like an AI-generated video, similar to like Google VO3, but it's actually much different, right, Josh? Because what you're looking at, you know, the perspectives of panning left and panning 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. That is distinctively different from a video model, Josh: [4:24] Right, Josh? Yeah, so with a video model, like VO3 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 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. Josh: [4:57] 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 genie 4 is going to look like where you can really get like some truly high fidelity you can get all of the the real world building physics aspects that 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. Ejaaz: [5:55] 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 Genie, 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, Lucasfilms, 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. Josh: [6:50] So Do you have any really fun ones to highlight? Ejaaz: [6:52] Yeah, yeah. Okay, 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, 2D-esque 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, Ijaz, 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 ChatGPT 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 can 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? Ejaaz: [7:43] And we had a follow-up 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 in this, he fully looks around and he's like, okay, yeah, 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 uh 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 on 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 Genie 3. 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. Josh: [8:41] And remember your earlier point. Ejaaz: [8:43] 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 Josh: [9:10] It's amazing how how quickly and high fidelity you're able to generate these things and it's also like look at that it that is the person that you 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. Ejaaz: [9:26] Did you see this drone shot, Josh: [9:27] Josh? No, but as a drone pilot, I would love to. Oh, wow. It's an FPV drone too. So it's crazy. Ejaaz: [9:34] 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. A super cool. Josh: [10:09] 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's 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 that's pretty cool. OK, another one. What do we got here? Ejaaz: [10:30] OK, 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 triple A gaming studios that haven't come out for a game or with a game in 10 years. Ejaaz: [11:05] 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 Josh: [11:14] 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 there's a high probability that we'd be getting gta 6 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 triple a game that is 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 you do this with one prompt is like it's pretty remarkable so genie 3 is pretty incredible grok image gen which we did an episode on earlier this week is actually doing very very well too this this theme of image generation visual generation it's hot grok 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 clawed opus 4.1 it is better than 4.0 believe it or not and it's funny i mean. Ejaaz: [12:11] The the char crime well Josh: [12:13] No so so open ai 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 4.1 is out if you're a developer go use it it is 2 better and this uh pretty funny chart and accurately shows for the for. Ejaaz: [12:33] 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. Josh: [12:49] There you go. Okay, so we'll have a debate whether we're hitting a wall or not in another episode, but that's a 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 11Lab software that they released this week. it allows you to generate music. And I was like, okay, sure. AI music. Great. I've heard plenty of AI music. Ejaaz: [13:08] This is good. Josh: [13:09] This is good music. Ejaaz: [13:10] It's actually really good. Yeah. Josh: [13:12] Walk us through it. Ejaaz: [13:13] For some context, Eleven Labs 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. It 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 E-Jazz. 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 Genie 3 or a VO3 and then I can just, you know, sit on my sofa all day. But yes, we have 11 music, which was unexpected, to be honest, because, you know, 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 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 GPT-5 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. You're listening to some deep jazz. All of this is AI generated right now. Ejaaz: [14:31] 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, 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 Ejaaz: [15:14] 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? Josh: [15:33] 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 tab 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. Ejaaz: [15:51] Here we go. Josh: [15:53] Okay we have kind of like and the vocals are interesting right because i mean ai is generating the lyrics what do we have here that's. Ejaaz: [16:03] An afro beat that's insane Josh: [16:12] That's some hip-hop okay, yeah that's amazing because you have to imagine i mean what those lyrics are ai generated they're 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 the interesting phenomenon that's happening now since mostly tiktok is the the tiktokification of music like a lot of people who are producers they kind of generate or not generate but they they produce music that's made to be listened to on iphones they they sacrifice a little bit of quality to make sure it sounds great on an iphone or an airpod and what you're getting here is 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. Ejaaz: [17:06] That's a really good point. The other thing is, Josh, I can't, I don't know about 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've mentioned GarageBand like twice on this episode already, but I'm going to bring it up again. I remember playing with that app the first time it 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 maybe this will change it. And the last bit I want to show is this quick clip with an interview with Matty, who is the CEO and founder of Eleven Labs, who we're actually getting on the show in about two weeks. So we're going to interview him 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. It's 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 Merlin and Cobalt frequently referred to 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. Ejaaz: [18:22] So already 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 to studio producers, labels, or whatever that might be. But what's also important is they have a direct partnership with these two companies, Cobalt 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 this. So it's making it more democratically accessible to budding upcoming artists, which I thought was really cool. Josh: [19:02] 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 with that little beat maker this is a like pretty big step function improvement yeah this is 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. Ejaaz: [19:30] 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 be used theoretically to treat any form of disease, which includes cancer, skin disease, 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 chai 2 delivers de novo 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 2 we delivered an experimentally validated binder basically a theoretical solution or cure in just two weeks and where my mind jumped from this was Ejaaz: [20:51] 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 $5 million, which is just insane economically, right? So I'm starting to understand why OpenAI 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 healthcare. Josh, what do you think of this? Josh: [21:30] We're kind of, I mean, throughout this episode, we are planting these seeds of a really optimistic and pretty wild future where we have like world building in a prompt and you could create your own metaverse. And now we have, 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. There's a world in 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. Josh: [22:37] And I think the theme of this week is very forward looking and really optimistic and exciting. And this, yeah, this leaves me excited to see. Ejaaz: [22:45] Yeah, we're entering an era of AI science. And what I 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 subscriptions, 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. Josh: [23:11] That's one of the cool things about like the GPT-5 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 ChatGPT, not realizing that the model they're using is actually like a couple of years old. It's pretty outdated. And just today, they woke up today and they went into their ChatGPT 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, 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, Josh: [24:19] but that's the appearance. Ejaaz: [24:21] The other breaking news over the last 24 hours is Zuck or Meta has secured $29 billion in private credit lending. So like he's borrowed this money from a private lender to fund their new AI data center. And we've 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 Grok 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? Ejaaz: [25:21] Answer is actually no, like they have technically enough money, but the CapEx spend on this far exceeds 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 XAI 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. Josh: [26:13] 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 in employee? That's what you cover the payroll for a hundred 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 where it's just like, what is the value? It's almost hard to imagine the value unlock because it's so obscure and it happens in. Ejaaz: [27:21] So many places. It's inconceivable right now. Josh: [27:23] Yeah. Like, when he says, like, oh, yeah, we're going to, like, Josh: [27:26] 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. Ejaaz: [27:47] I appreciate the big bets right like zop did this when he acquired 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? 40 billion? I forget the number now, right? It was like 42? Okay, just $42 billion. Josh: [28:13] Yeah, but again, to take. Ejaaz: [28:15] That massive risk, the stage and level of foundership, like, you know, if you're worth $100 billion, just retire already, right? But these guys are making massive gargantuan leaps. It's just very commendable. And just to describe this, I'm excited. 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 he wanted to but he's he's going again and i respect that yeah Josh: [28:49] Respect and admire it i think the more you know the less you need to diversify and 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 aware of the upside and the fact that all these people are deeming it a risk to take well sounds like that's the right Josh: [29:07] risk to take and i'm here for it is that is that everything for this crazy wacky wild. Ejaaz: [29:11] Week uh no i just wanted to show you this new film that came out josh i want to give you a little bit of a trailer yeah yeah let me 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 generated very good 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? Josh: [29:42] It was cool, but it was so lame. Ejaaz: [29:44] 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 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. 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, more cartoon like, you know, from Into the Spider-Verse. If you're interested to 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 teenager sitting in his parents basement just super Josh: [31:07] Cool yeah for animators this is such a game changer i i forget the exact timeline but the movie spirited away which is like this very famous popular anime movie i 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. Josh: [31:59] 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 if it is that lightweight and it doesn't require all that compute power oh yeah here's a here's. Ejaaz: [32:49] One google i 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 dollars and unreal engine is a major reason why it's worth that much and now you have an app again from a subscription service that can generate very similar quality aesthetics you have genie 3 accessible to you that will like 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 dollars and then maybe the cost of two monthly subscriptions maybe 150 bucks a month and you have access to all the same tools that unreal engine has to build these things Josh: [33:43] Pretty awesome pretty good pretty good and again it's not it's not unreal engine but come genie 4 genie 5 or the next mid journey 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 Josh: [34:00] high so i think that's everything is that everything can i take a deep breath now are we done so we. Ejaaz: [34:05] Covered josh we are done we made it for now i don't know what's gonna happen the next this week Josh: [34:09] 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 that it feels like every week there is more and 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 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 yeah covering all these great topics so i guess i guess i'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 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 charts. Fun fact, we are one spot away from flipping the OpenAI podcast. So if you share the episode, you are. Ejaaz: [35:17] Helping us flip Josh: [35:18] OpenAI and that's just like a pretty cool thing that I'd like to have as an accomplishment. I'm sure EHS feels the same way. I would love that. 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. Music: [35:35] Music