OpenAISora2CharacterCameoAIForHumans

Gavin Purcell: [00:00:00] OpenAI is growing again this time to a trillion

Kevin Pereira: dollar behemoth. Uh, yes. And I made a Sora video with my dog, Dr. Wesley.

Gavin Purcell: Dr. Wesley's up. Look at that. Focus fighters. Ready?

Kevin Pereira: Ready. Go. Connection.

Gavin Purcell: Yes. Yes. Sora Two's new cameo creator, also launched, but Kevin OpenAI is saying Automated AI researchers. By 2028,

Speaker 4: we, uh, look towards getting a system capable of autonomously delivering on larger research projects and a, a meaningful, uh, fully automated AI researcher, uh, by March of 2028.

Okay,

Gavin Purcell: Kevin. In just two years, we could be getting past a GI. Yeah.

Kevin Pereira: Okay. But today, here's Dr. Wesley and Ollie. That's your dog, Gavin. And look at them. They're in a Ferrari. Whoopsie doodle. The cops are there. They better step on it.

Gavin Purcell: Okay. I admit that is pretty fun. Also, one X's new neo model is taking pre-orders and causing controversy.

But Kevin, can it close a dishwasher?[00:01:00]

Kevin Pereira: Wait, the robots are getting stuck in dishwashers. Oh, how about the Move on?

Gavin Purcell: We're moving on. Move on. Right now on, can they also

Kevin Pereira: get stuck in washing machines? Gavin plus the music app. Uio just made a deal with Universal Music Group and let me check my notes now. You can't download songs. That's good. That is

Gavin Purcell: not good news.

But in better news, Kevin, we have a ton of new AI video and audio tools. Plus Cursor 2.0 is taking vibe coding to a whole nother level.

Kevin Pereira: Uh, Kevin, uh, I think we got a problem here. Oh, no. What is it? What is something else going on? We were so worried about the robots, but it turns out our dogs want to take our jobs.

Gavin Purcell: This. It is AI for humans, for humans.[00:02:00]

Welcome everybody to AI for Humans, your Weekly Guide into the wonderful world of ai. We are here again, and Kevin, there are two roads that open AI is going down right now and we have to talk about both of them. We have discussed this before. OpenAI is a behemoth, a behemoth of a company. It is very large.

But first we know that the big thing that people in our audience love and other people love. Let's jump into this SOA news and then we'll get to what's going on in the actual company. Tell us a little bit about this update to SOA

Kevin Pereira: two. This update brought me back to the platform and I received my first warning about trying to generate too many videos.

Since Sora launched Gavin, they added character cameos. If you wanna take a furry friend, if you wanna take an inanimate object, uh, a, a frying pan filled with eggs, a child's hand, Turkey drawing, it doesn't matter. You can upload or capture a quick video. It will analyze it and turn that into a character cameo for you or anyone to use.

And this is something we've talked about

Gavin Purcell: from an early [00:03:00] stage in that, like we were saying, how cool would it be if you could open the door to using Sora two with things that weren't, it weren't human, or things that like were animals in different ways. And Kevin, I have had so much fun with this already, I do wanna shout out a couple things.

One time I saw one of the guys from SOA created this kind of pickle character with these googly eyes. Mm-hmm. And what was really cool about that is like. When you have a character like that and placing them in situations. So two does a very good job of understanding the physics of that world, even for these weird characters like this.

Yeah. The other thing I saw somebody do, which made me laugh is they took air pods and they put little, they put little eyes on them and it was just a video of a set of air pods on the ground. Yeah. Um, and that's a character too, right? Someone else used a toilet paper roll, which was like, didn't do anything.

It was just a single toilet paper roll sitting in scenes. But there's a lot of really interesting creative openness you can do here. But we should talk about what you and I did, which is we both put our dogs into a number of videos.

Kevin Pereira: Yes. Uh, I'm a, a childless sinner, and the closest thing I [00:04:00] have is my sweet Dr.

Wesley. So I grabbed a video out of the old camera roll. I hit the button. It only needs a short little, like three seconds, three or four seconds, 10 seconds. It's, yeah, it's really small. Um, and with it, it auto generated a description of Dr. Wesley. There happened to be a sun beam in the photo, or excuse me, in the video that I uploaded.

So now he's like. A bright bushy tailed ball of energy that chases sunbeams and blah, blah blah. And I'm like, I think my dog's kind of a jerk, but that's okay. I'll go with a sunnier descriptor. And within a minute, there it was, and I could puppet it and pilot it. And I made Dr. Wesley public and, uh, folks in our discord have been doing all manner of stuff.

There was a great buddy cop clip, um, and I say it's great because I made it Gavin, um. But that one came right out. And, and again, like I, I was kind of, I had moved on from SOA a little bit. I was checking the platform much less. I certainly wasn't generating stuff, but this brought me right back. Yeah. To make a, a bunch of videos.

And then when you said Ollie was on there, I'm like, well, great. Then I started generating stuff [00:05:00] with Wesley and Ollie, and again, this, this is a true social network effect where you want to play with everybody else's toys if they will let you.

Gavin Purcell: Yeah. We should tell everybody, if you're out there and you want to generate things with our dogs, you can, mine is under the name at Ollie Purcell.

P-U-R-C-E-L-L-O-L-L-I-E. And then yours is under, like, it's kind of something, it's kind of part of your name right. In some way. Not Kevin dot. Wes is Dr. West, is that what it is?

Kevin Pereira: Yeah. It's like not Kevin dot, Dr. Wesley, I think. Or Dr. Wes. Yeah.

Gavin Purcell: I jumped on this pretty much right away and I want to kind of talk through some of the generations I did.

So the very first generation I made was with Ollie, uh, being interviewed about why he was a bad dog and ate the food off our table. And I wanna shout out one specific thing about this, which is really cool. If you're watching the video, you'll see it here. If not, I'll just describe it. This is a scene where he's being sit, he's sitting in like a director's chair.

And what they did really well, I had them wear sunglasses and they got him to take off his sunglasses. The thing that it did not do very well is I gave words to him to say, and maybe Kevin, we can play this one really quickly just to kind of listen to [00:06:00] this and see what it sounds like. Why did you eat Gavin's dinner?

'cause it was there.

Speaker 5: Okay.

Gavin Purcell: So, so first of all, what's going on? What's going on in his voice? Dude, second of all, dude, your dog is

Kevin Pereira: huffing Galaxy gas. You've gotta get it under control. That dog is high. Did you prompt the sunglasses and the deal with this stuff? Yes. Or it did? Okay, so

Gavin Purcell: I, I prompt the sunglasses.

The thing that I think is really interesting about this, at least for animals and, and I have seen open AI's examples used mouths better, but every time I tried to get Ollie to talk, it didn't work very well. There was another version, there was another one I did right after that. And these are my first two Jens.

Where I had Ollie chopping food in a kitchen wearing like a chef's hat. Yeah. And I had him try to say something to me and you saw his like little tiny movements of his mouth, the back of his mouth, but it wasn't moving his mouth. So I'm wondering if there's a way to like open the door to that. But the much more fun stuff is to do things that you would just never see.

I created a video of Ollie battling Ultraman, which is really fun. Which, yes, I love that one. [00:07:00] Yeah, which was super cool. And then I do think, Kevin, that the one we should. Show before we move on from the section is I created Ollie and Dr. West in a very famous talk show clip. Maybe we can play that one for everybody to listen to.

'cause there's some fun audio on it.

Speaker 5: It's a four month old Dr. Wesley Ollie.

Gavin Purcell: You are not the father.

So, so describe if you're listening, that that is a, a, an owner. That's not us by the way. I could have put us into it. He's sitting there with Dr. Wesley and then another owner sitting Ollie and then it's clear that Ollie's not the father. My favorite part of that video is just the, is like the closeup on Ollie's face as he's just there thinking He's not reacting at all.

He's just thinking about what his life is gonna be like now that he's not the father Wesley. So that's a very exciting thing. We is pissed.

Kevin Pereira: Yes. He's Wesley. Wesley goes full angry bark. What I love is that like all the memes from before come back around again. Yes. Right. Like you, you, you go like, oh, now we can put our dogs in these templates.

Now we can remix that that way. So I've got, um. [00:08:00] A, a very dramatic slap fight, one of those professional slap fighting competitions where Wesley just saw, pause me, I saw, and knocks me out. Um, I, that was really enjoyable. Uh, Wesley, uh, walking to the ring as a WE Superstar. Ma chef's kiss the poop sampler.

Gavin, I don't know if you saw that, but Wesley, Hey there. I'm Dr.

Speaker 6: Wesley. I chase sunbeams. I wag for strangers, and I know good snipping when I find it. That's why I teamed up with the folks at Poopy sampler to craft a delivery box packed with the wool. Interesting. Slightest droppings. Possibly sounds like a startup.

When you see,

Kevin Pereira: when you see the poopy sampler that Chet G, like the Sora itself came up with it. Yeah, there's a, an herb fed, a smoked bison. Wow. And then there's a, a forest cow and they're all little.

Gavin Purcell: We should talk about why this is interesting really quickly before we move on to the other opening AI news.

And I think to me, yes, it will bring more people back into soa, but I think from a creator standpoint, if you are creating things, the number one thing you want is character consistency. Right? So what this means now is if you think about it [00:09:00] from an animated character, you think about it from any sort of other things.

You can now use the SOA platform to create other stuff. And what's cool, they also kind of dropped this new, there's a couple other things they dropped, which are. In drafts, you can now stitch together videos, which is kind of interesting. So you don't have to publish a video. Mm-hmm. You can actually stitch it together, meaning you could make longer videos in drafts,

Kevin Pereira: which means cap cut incoming.

Like if you can already stitch together videos, why not describe a transition or prompt some background music or change a character's voice.

Gavin Purcell: Yeah. And that's another big kind of place that open AI could open a door to making money or to selling things. Right. The other thing they talked about was leaderboards, which I find really interesting.

Like you can imagine a world where there are gamified challenges or all sorts of other things that can be done. So I really do think it's important for people not to underestimate what Sora the app is doing as a social network. Mm-hmm. I, I know that people like Kevin had said, have kind of fallen out of love with it, but.

When you see things like this and as new features come up, I just think you're gonna see a lot more people sticking around with this [00:10:00] app. I think it's gonna grow significantly over time. Yeah. And I feel pretty like bullish on this as something that's gonna stick around

Kevin Pereira: at large. Same here. Ga, I think it's, I think it's absolutely has staying power and I think as people like imagine right now it's hallucinating the voice of Ollie.

Yeah. But if you could upload a voice for Ollie. Yeah. And we know OpenAI can do voice cloning and they can do it with just a few seconds of audio. Yeah, right now suddenly you've got people creating stables of characters, making shows with them, and then opening it up for a community to remix it. Like it's, it's the, could be the future of, uh, like a TV network.

Gavin Purcell: Yeah, a hundred percent. Um, while that's all happening, another thing that's going on in the background, there's some other really big opening AI news. Number one, they have officially completed their. Corporate structure, which I know everybody's out there, they are wondering when will the corporate structure be finished?

Well, it's done. Uh, they have officially said that they are a for-profit company, but there's a big chunk of their company that they are going to keep, uh, a nonprofit. In fact, Sam Alman said that, I think $25 billion of their investment [00:11:00] money is gonna be going towards AI for health and AI safety. Mm-hmm.

So that was a big thing, I think in part that was out there because he wanted to make sure that this, all this messaging got across right. But, but the bigger thing here, Kevin, is Sam and OV Pache, who is now open AI's chief Scientist, which was the role Ilya Skeer played, uh, earlier in the, in a, in a different life, in a different show that we've been following.

Uh, they came on and did a discussion from where they see AI going in the near future. So let's take a listen to this, Kevin. We want to provide

Speaker 4: some transparency around our thinking there, and so we want to take this maybe somewhat unusual step of sharing our internal goals and goal timelines, uh, towards these very powerful systems and.

You know, these particular dates, we absolutely may be quite drunk about them. Uh, but this is how we currently think. This is currently how, how, how we plan and organize. And so as a research organization that is working on automating research, naturally we are thinking about how does this impact our own work?

And, uh, how will AI systems that accelerate [00:12:00] development of future AI systems look like? How can they empower, um, research like alignment? And so, uh, we are making plans around getting to quite capable AI research interns that can meaningfully accelerate our researchers by expanding, uh, uh, a, a, a, a significant amount of compute, um, by September of next year.

So we believe that is actually quite close. Um, and then we, uh, look towards getting a system capable of autonomously delivering on larger research projects and, uh, a meaningful. Uh, fully automated AI researcher, uh, by March of 2028. So

Gavin Purcell: let's talk about what that means, Kevin. That is a big deal, at least according to their timelines.

And, you know, this is a company timeline and they probably have incentive to say things are going very well, obviously. But what he's saying is that in one year from now, we are going to have AI research interns, which means that, and Sam followed up with a post about this, where the idea is that the interns will [00:13:00] be able to make small discoveries, but by 2028, you would have AI research engineers, full engineers that are driven by ai.

And what those will be responsible for is furthering the capabilities of ai. So we have talked about the fast takeoff, if you remember for a while, which means how quickly AI will take off. All of those fast takeoff narratives are driven by the idea that AI has research engineers, meaning that the AI itself is engineering and getting better at itself.

So this is a pretty aggressive timeline compared to what some people have set up. Um, it's not that aggressive compared to AI 2028 or whatever that that document was that we read a a little while ago. But this is a big deal and I think you people have to kind of sit up and listen to this. Now, obviously, like I said, I dunno if it's fully believable, but it's an important thing to kind of be aware of

Kevin Pereira: when they say that there's gonna be automated AI researchers by September of 2026.

That doesn't sound like science fiction to me. That doesn't sound like saying full self-driving will be here yesterday. We're good. Um. The many of the apps that I'm using [00:14:00] today, like when I use cloud code in a terminal and I say, go off, research this thing, organize it in this way, right? Extract the best practices or principles from it, and then write an app that lets me interface with it.

I know it's not necessarily discovering something new, but that certainly feels like an agentic research assistant and I'm using that today. So if we believe these models are gonna get a little bit better and a little more capable in a year's time, and if history is correct, then it, then they will. That doesn't sound that wild to me.

The notion of 2028 though, and we are hiring these things to go off and discover new stuff. Like I, I mean, I guess why not, right? We got, that's plenty of time

Gavin Purcell: between now and then. That's the hard part right now to really understand. We're gonna talk about this one X robotics, uh, story, which I think opens the door to this conversation too, is like what we think we have right now versus what is kind of being promised by these companies.

Yeah. In a year or two's time. So really quickly before we move off of opening ai, we also just follow up is that Bernie Sanders has [00:15:00] now gotten involved in the OpenAI conversation and maybe kind of surprised Bernie Sanders says it should be broken up, which is kind of like Bernie Sanders, like, you know, for what you like about him.

He is not a big fan of like giant organized, uh, things that have a lot of power. So, you know, there's already people seeing some blowback on this. Kevin. I think one of the things that OpenAI is gonna have to deal with and all these big companies are gonna have to deal with, I dunno if you saw this, but like Google just uh, said they have a hundred billion dollars in revenues for the last quarter, which is their highest ever.

You have about eight companies that are gonna start taking a ton of the value and Bernie's out there saying like, Hey, we have to really be aware of this. So I just think it's important to track this kind of human side of the story

Kevin Pereira: too. Do you think we're gonna care? In like three years time when we're just ordering our robot assistance around, like, I feel like I'm gonna be more upset that my ice cubes aren't perfectly ID and I'll send the robit back to the kitchen.

Gavin Purcell: Will you be able to afford the robot assistant? Is maybe a better question in that sort of way. Let's talk about this. So this was a huge, no answers no by the way. Yeah. This is a huge story that blew up, uh, a couple days ago. One X robotics, which we had been [00:16:00] tracking here for a while, has a, a, a launched and announced video for pre-sales for their neo robot.

So if you remember this one, this is the one that kind of looks like a, a little different than all the other ones. It's kind of more relaxed looking. It's wearing like a canvas outfit. It has a little, kind of a very simple looking face. Mm-hmm. But in their release video, they really showed off some really interesting new skills of this thing.

First and foremost, their goal with this is, is a chore robot. So they are pitching this not as an industrial robot. They are pitching this as a robot for your home and for $20,000 you can pre-order right now, won't deliver until 2026. But they are saying that it will autonom autonomously learn how to do things around your house.

But Kevin. How do you think it autonomous autonomously learns to do those things? Because this is a very important point.

Kevin Pereira: Well, it learns it through simulation and software, and it learns it through teleoperation. Yes. Someone taking over a control of the robot to move it about, which thus far seems to be the way the robot has been demonstrated, which I know we're gonna get to.

But let, let me talk real quick about this. Like [00:17:00] $20,000. You can own it, right? Yeah. The robot is yours. You get a three year warranty, you get premium delivery. Okay? Sure. You get priority, uh uh, pre, excuse me, you get premium support, all right? You get priority delivery. Okay? But you own the robit, but for $500 a month, Gavin?

Yeah, you can. You can buy it on layaway. That's right. You get, and it includes your monthly subscription of. Teleoperation Yes. Slash other chores that it can do supposedly autonomously. So, so the

Gavin Purcell: interesting thing about this is that there are two things happening at once. First and foremost, one X robotics is having this really fun video that comes out and shows off what it's capable of and announces pre-sales.

They also then led in our friend, friend of the show, Joanna Stern, to do a kind of hands-on with this robot as it is. And Kevin, if you've watched that video, and I hope everybody in our audience goes and watches it, we'll link it to below. What you see is very much like what this robot is capable of today.

And a couple things that come out of Joanna's video that are important to know today. It is all remote [00:18:00] operated, so when they demoed it the whole time, there's a person doing the things. Also today it is struggling a lot to do very simple things. Joanna had a go-getter or something outta the fridge and it took a couple minutes, which is something you a human would take 15 seconds to do.

But then to load three glasses and a few pieces of silverware, and I think one plate into a dishwasher. Took five minutes and of that five minutes, a good 45 seconds, was it trying to figure out how to close the dishwasher? Yeah. And Kevin, I think the important thing for our audience to know is that like we are not haters of this technology.

In fact, I am a big believer that this is like a huge thing and watching these things do this now it's really important to understand this is, does not mean where we're gonna get at in a year to five years from now. But Marquez Brownley had another good video talking about Joanna's video, which was really the sense of like what these tech companies are promising and what they're selling versus what is actually gonna be delivered when you get it.

And I think that's a big thing to understand here.

Kevin Pereira: Yeah, I mean if you're just going by the cherry picked video highlights on their website, you would think [00:19:00] this thing is going to be unloading groceries from the car. Yes. And handling a whole bunch of stuff for you, which it might be able to do over a long enough arc of time if someone else is remotely operating it, but they're selling it as this autonomous something at the moment.

Um, no shade. We, I love that we're taking steps towards this. You and I both believe this is gonna be massive in the future, but like, let's keep in mind, and this, I think this holds true pretty much across the board. The third version of a product is usually the good one. Yes. Right? Yes. Like, V one comes out and you're like, oh, okay.

I can see the shortcuts they took and the software's not there and it's a little heavy, or it's a little, this or that. Version two comes out and it's like, oh, they did optimize it a little bit. Version three is the one that usually fulfills the prophecy. Yes. So B three or, uh, I guess three X of the neo robot will probably be the great one.

We gotta start somewhere. I have one thing

Gavin Purcell: I want to point out, which is what I like to call the, um, Roomba problem, which this is my Roomba problem, which is I've owned, I don't know, five Roombas over the [00:20:00] course of my life. I got one very early, it didn't work very well, and I didn't stop, I stopped using it.

The one I had most recently was maybe three years ago and, and probably Roomba or whatever Wizard or whatever other brands exist now. If you're out there and you're like, it's so much better now. Great. Yeah. Even the last version I had, which is three years, three years ago, which is probably like a 10-year-old product of Roomba.

There are multiple rumors that have come out from that. There were so many issues with how that thing worked. I had to figure out like what was gonna do, how it was gonna deal with stuff, especially with pets. It really didn't understand what was going on. And I, you know, we've talked about this on this podcast before, but a couple like accidents have happened and you know, spread around everywhere, which is a terrible thing to see.

Yes. The idea here is that like at some point I chose, it was not worth it for me to have a Roomba. Right. Because I was like, you know, screw it. I'm just gonna sweep and mop myself because it's a much simpler thing. Yeah. I think you see a lot of people out there when they see this video, it's like, why would I wanna have something in my house that is gonna be worse at doing the sorts of things I do?

Or a lot of people have the freedom to bring in a person that can clean for them. Like [00:21:00] that's a really great thing if you financially have that availability. But, and it's also a job for a human right. But I think the bigger question here is, and I think the point that I wanna make is that we both, I think, believe that these things will get there.

Yeah. But we just might be. Five to 10 years away from a place where they're actually able to do the things that we think is being promised by the company

Kevin Pereira: at $500 a month. If this robot really could efficiently, right, effectively do the dishes, do the laundry, put the groceries away, it's usually valuable.

That's actually super valuable. It makes total sense, right? Um, uh, that, that, that's great. Uh, how many hours a day will actually be able to work autonomously to amortize that $500 a month cost? It remains to be seen, right? Like it, how does it charge? Well, we know how it charges, but do we know how long it lasts between those charges?

In two to four hours is what they've said, two to four hours, which is not that much right now. Right. You think about it, it's not that long, especially if it takes 15 minutes to close the dishwasher. Yeah, [00:22:00] right. Then it's got a shamble over to the charging dock that's

Gavin Purcell: four dishwasher closes or whatever.

It's like eight dishwasher closes. Eight dishwasher

Speaker 5: closes. What? They should do it.

Kevin Pereira: Right. And then it's gotta go and charge and we don't know how long that's gonna take. Yeah. So it's like the, again, there's a, there's a lot of nebulous stuff here, but what's interesting is you and I have talked about this before, like inviting a quote unquote stranger into your home.

The privacy cameras, Mike. Yeah. The FAQ goes well, out of its way to say that you have to schedule when an expert. Connects to the device. Yeah. Um, the experts are all, uh, uh, physically present in the USA, it says Interesting. Okay. So they make that very clear. It says NEO works without the need for data sharing.

In autonomous mode, limited sensor data may be sent to our servers to fulfill a request. Yeah. What, what is that? You may need to opt out from data sharing. Uh, uh, you may or you may always opt out from data sharing for improving Neo's performance. Okay. That's nice. Sure. But I wanna know what that sensor data is.

Um, can my neo cook, which is a big one, and the answer is, oh yeah, no. [00:23:00] What it says, no. Now cooking capabilities will be restricted from use. Neo can provide you with great recipes or help with the cleaning up instead. But again, if you see it, try to manipulate the dishwasher door. Maybe you don't want it fumbling to walk dangerous on the gas

Gavin Purcell: with the heat.

I don't need something else in my house telling me what to do, Kevin. I don't need a third party saying Go do this, go do that. That is the last thing I need.

Kevin Pereira: Is that where the shallots go? Gavin? Are you sure? Oh look, another gym sock on the floor. I guess that's my responsibility. Exactly right. One x also is neo waterproof?

No, just the hands. No, that's not

Gavin Purcell: surprising I guess,

Kevin Pereira: but But can it be used outside also a no. And so can it change its own clothes? What does it say about that? Um, it will actually disrobe at quote. Inappropriate times and intervals. Now I don't know what that means. Get outta

Gavin Purcell: here. It does not say that.

Doesn't say that at all. But that would be an interesting one. That would be an interesting one. [00:24:00] Uh, meanwhile, Kevin, the other thing they could point out with robots, we got a little early robot watch here. In China, unit tree and all these other companies keep going. There's incredible clip that just has an interesting difference of this.

In China, they've trained a G one, which is, you know, Riz bot and all those ones. Yeah. They've trained a G Ones a tiny aerobic. Yes. They've somehow trained a G one to pull a car. So there's a video coming out of just a a G one pulling a car. So here. We're struggling with cooking and all that stuff, right?

In China, they have turned somehow this little tiny thing into a car. And I will say with one of the things with uh, the one X Neo is they're really trying to build something that is not dangerous. Like it can only lift 60 pounds. It's very little soft and then they're trying to make it so it's very usable in the home.

But still, I am curious to know, like part of me wants to know like what is the one X version? Is there a version in China that's already doing some of this stuff that we're thinking about? Right? And it'd be interesting. And where

Kevin Pereira: are the weird benchmarks for that? Like where's the strangle strength? When do they have like the choke grip force?

Because like lifting 60 pounds is actually pretty strong for that tiny thing. [00:25:00] That's true. True. What's the pillow test? Can it smother me if I didn't make my monthly subscription?

Gavin Purcell: That's coming. That's definitely coming. Alright everybody. Um, one thing that also is coming is our subscription numbers are going up, which is exciting.

Our YouTube subs are going up, which is always a fun thing. You out there have always and everyday shown up for us and we appreciate that. Thank you. If you are not doing it, you must right now go like sub and do all the fun stuff on YouTube. Leave a comment for us. Our great comment experiment. I think last week worked out well Kevin.

We saw a great variety of people engaging with our comments, so we really do appreciate that. Also, we have a Patreon, if you wanna throw a couple bucks in the tip chart. We've got a couple new Patreons this week, which is really exciting. That always helps us pay for both our editor for the show and the tools that we spend money on.

So thank you very much

Kevin Pereira: and if you wanna be the first recipient of what will be the most valued, number one MVP comment to reward. Drop a custom haiku in the comments.

Speaker 3: Ooh, okay. Just any old haiku

Kevin Pereira: about this episode, about any of the content, your favorite AI for [00:26:00] humans, uh, references or characters that we don't even do anymore.

Drop a haiku in the comments. A unique, does it

Gavin Purcell: matter? Yeah, I was gonna say, does it matter if it's AI generated or not? Can

Kevin Pereira: they do? Oh, I'm assuming it's gonna be, I don't take the time to write your own haiku. Oh, I thought that's crazy Haiku. Yes it is. Five, seven, step five. It's not that hard. It's damn near.

See? Impossible would've gone over. You can't do it. You literally can't write a haiku. It's

Gavin Purcell: impossible. Well, you're right. Yeah. See, it's impossible. It's impossible to not make fun like the lab grown of now. Seven. Please help us out, folks. There you. 5, 7, 5. I'm done.

Kevin Pereira: Yeah, and, and I can tell you that comment's not winning the MVP award.

No shade Gavin. I just know that it's not, use a robot, leave a haku, uh, or maybe just load up an Adobe product. 'cause I'm sure they have a new AI haiku writing feature now, which artists will be okay with because it's inexpensive product and has a brand name. That's how works, that's how it works with Adobe.

It's not generative ai.

Gavin Purcell: So Adobe had their big Max sneak event, and I will [00:27:00] say Adobe really goes all out for these events when you, what they've done on their stage and what the thing was. Obviously Adobe is like leaning in hardcore to Ative ai. They know that this is the future of these tools, but to your point, they do a very good job of integrating into places that, uh, artists and creative professionals already use in their workflows.

So yeah, let's tell us a little bit about what they announced here. 'cause I think it's pretty interesting.

Kevin Pereira: Alright, Gavin, they had a, a couple AI sneaks at their Adobe Max event. One was project motion map. Mm. Which will take vector illustrations and basically animate them automagically. So they had a demo where they had like a logo for a food restaurant that had like some, uh.

Food restaurant as opposed to the other one. Oil restaurant. Well, I was trying to remember what the restaurant was. Lanker Restaurant. Yeah, exactly. Now they had a, they had an example. It was a logo for a restaurant and it had some eyes in it, and they're like, make it blink and make the text rotate. And it understood what it was like.

It figured out, oh, those are eyes, we'll make them blink. They had another layered vector illustration of a person in a swing, and when they hit a button, it [00:28:00] sort of automatically added the little objects that would be needed to make the person swing and animate it. Um, cool. Again, I feel like, I mean, artists, obviously, the people in attendance, they're, they're clapping it up and they're loving it.

Um, it's interesting to me to always watch the reactions from the capital A artists on this one because they go, wow, what a powerful tool. It's ai, you're using generative ai, but it's within. It's within the, the, the, the software package that, you know, so I guess it doesn't feel as dirty. Um,

Gavin Purcell: I think that we're seeing a shift, by the way, I really do think there is, even with Capital A artists, I think we're seeing a shift of people and I think that this is much, much more mainstream than it was when we first started this podcast, obviously.

But even a year ago, I feel like it's a big deal.

Kevin Pereira: Well, I think people are realizing that, oh, if you have talent, it might not have been the exact talent from before, but if you have talent and taste, you can leverage these tools and do something better than someone just whispering to the machine. That's right.

Can output. And that is once again, an artistic difference maker [00:29:00] project clean take. Hey Hollywood, pay attention to this one. You can edit dialogue. It's kinda like a D script, Gavin. You can edit dialogue by just like messing with the transcript. You can change. Oh good. Yeah, sure. Yeah. The pronunciation and the performance of things.

But what I really liked was that right within the timeline of your video editor, you could click a button and it would automatically separate the speaker from the reverb of the room, from the sound effects, from any detected music, and you could regenerate those stems on the fly. So if you were filming something in a a, a, A restaurant, a food restaurant where there was background music playing that you didn't have the rights to, you could take it out.

You could isolate that track. You could mute it, or you could hit find me a sound alike and it would go to a library and grab a track that sounded like it. And then you could say, make it match the environment. Match the acoustics of the place. Interesting. That's very cool. Um, yeah. Really, really, really cool stuff for, for video editing, uh, and for sound editing within that video, um, they had a surface selection tool, uh, that would, um, [00:30:00] automatically, like if you had like a car and you wanted to change the color of the car, you could hit it, it would map the properties of it so you could retain the highlights and the shadows or reflections within the thing.

Um, they had a tool that would take a 2D drawing and turn it into full 3D so you could place it within a scene. Uh, they had relighting tools, they had a trace, a race thing, Gavin. So instead of just highlighting an object in a scene and hitting a race and having it only remove that object, it would intelligently look for, is that object reflecting in something else?

Is that interesting? Is that person, they had like a really cool example where someone was swimming beneath the surface of like a of disturbed water and it. Pulled out their reflection in the wave of the water pretty naturally. Um, it's interesting '

Gavin Purcell: cause that's like the, the big thing of these AI video tools is when we use them, like whatever, it's the clings or even the sos, it's like kind of this like, throw something in and get something out.

Yeah. Again, we've said this before, nano Banana does this a little bit now, but like the more granular you can get with things, these, these things, the better. And I think Adobe's Lane might just be like, [00:31:00] Hey, we see all these major models and what we can do well is allow you to manipulate this stuff, right?

Because you wanna be able to do X, Y, and Z. Because ultimately, if you've ever built something in AI video or in AI photos, you know, the layering part is really hard, right? And like that opens the door to so many other changes. So that's, that's great. I'm excited to play around with this stuff.

Kevin Pereira: Some really cool tools.

Last one, project frame forward. You take a video frame, uh, the first frame of a video, you modify that first frame and it applies the modification to the remainder of the video. Oh, that so it's like

Gavin Purcell: a Yeah, it's like a style stylizing sort of thing. Like is it, is it like runways thing where you put a, you

Kevin Pereira: can change the style of a video?

Um, no. I mean, like you could intelligently insert ob Yes. You could change a style, but you could also insert things into a scene. Oh. So one of the examples they had was like, here's a shot of a cat as the camera pans by and he is like, ah, let's add a puddle to the front of the scene. And so the cat's still doing all of its cat-like movements, but the puddle is now tracked with the camera and reflecting the cat cool.

From the scene in [00:32:00] the video. So again, a tiny little tool that will get very powerful when you combine it with. Real artistry and potentially all these other tools.

Gavin Purcell: And is that all being driven by Firefly or are they allowing you the freedom to use the other models that are within their system? Because I know Adobe has also opened the door to using things like soro or CL or other models.

They have Topaz

Kevin Pereira: in there. Now, if you wanna upscale an image, cool, you can just say, Hey, upscale it. So I think it's a blend of their custom stuff, but I think they're using a lot of off the shelf models to achieve. Um, you know, the generative video and stuff like that. Got it. They don't really make a point of saying, we're plugging you into insert product here.

They're just saying you can do it now with Adobe. So tell me, Gavin, do those fun new AI features sound like music to your ears.

Gavin Purcell: They do. C you know what else? Sounds like music to my ears. Uio. Actually, there's something wrong here, Kevin. What? There's something's going on with UDO if you don't remember. UDO is the kind of competitor to the Suno.

It's kind of a second, a second now I kind of say a distant number. Two UDO just announced a very big deal and if you read, there's two different ways to [00:33:00] look at this story. The founder is very thrilled about this. They've made a deal with UMG Universal Music Group, which is a big label, and they are very excited to partner to partner with that label.

At the same point, UMG has come out and said, we are partnering with this company. The kind of underlying message here, though, Kevin, is that we know that UDO and Suno Bolt have been sued by these major labels. UDO specifically, we always thought might've had a slightly dirtier uh, dataset, meaning that the data that they took in might've had more actual music.

'cause you could get real voices and things out of it. That was harder to get outta suno. What this has meant is that it really means that UDO is being, for right now, capped a little bit, and users are no longer gonna be able to download the songs that they created there. My question to you is, let's explain.

Yeah. Why do you think that is happening? Why do you think the users are no longer able to download their songs?

Kevin Pereira: I, I don't actually know. Um, I don't really know. I mean, there, I I, I understand if you're generating a [00:34:00] song with a UMG artist or you're taking something that you've made and transferring it into the style of that artist, that makes total sense to me as like, oh, well you can't download that because that's our artist and we don't want you spreading it on a platform that we can't track or whatever else.

Okay, I get it. But if you look at the outpouring of folks on the UDO subreddit or on X, people are big mad for good reason. Yeah. Because they suddenly can't download things that they've made. Now, we, we could talk about pending litigation, and maybe this is one of the terms of that, right? Of like, Hey, our artists are in your system.

People are already generating songs, probably off the backs of our artist content, so no more downloads, but. Uh, this is, this is feels like a massive betrayal to the fan base that is paying for the tool and now is suddenly cut off.

Gavin Purcell: Yeah, so this is what I was gonna say is this is big music, whatever you wanna call that.

The labels have often been very draconian about how they deal with this stuff, and again. You know, we're, I'm not against [00:35:00] protecting artists' rights, like wanting to make sure the artists get paid, but I think in some senses, the idea of not letting people download to me says that, oh, there's some deal that's been struck about the way that this was trained ahead of time, that the rights of the music that has already been created is going to be some sort of legal snafu.

Yeah. And that, going forward, the idea that this would only be streaming is pretty weird to me. And I, I guess there's a world where, you know, streaming dominates music and, but, but for creating AI music, there is a level where like, I want to use a track for a video that I've created, or I wanna share a track with my friends.

And like, it may not be the easiest thing to share it on UDO. I just think this is a, a, a real sense of, like a company that came out pretty early, probably played a little loose with the, the, the copyright and trademark law. Yeah. And are getting kind of smacked down and now having to deal with the repercussions.

However long later, like a year or two

Kevin Pereira: years later. Uh, today we announced a strategic alliance with UMG to [00:36:00] co-develop professional AI music creation tools. This is stability. AI Gavin, hot off the presses. Oh. It looks like you UMG must have rounded up some lawyers in a room. Interesting. And started shaking some hands.

Gavin Purcell: Well, the other thing that's big about this, and I think it's important to realize is that, um, in UDO and um, G'S announcement, they did say that some select artists would allow people to create AI songs. Mm-hmm. So you can also see a world where, I don't know, think of the funniest, like who's the guy that did, um, uh, Mambo number five, if you could create your own version.

Lou Bega. Lou Bega. Yeah. Lou Bega. Imagine Lu Bega ai and Lu Bega is out there wherever he lives, in a mansion in the island somewhere. And he is like, of course, use me. That would actually be a fun moment, right? Like if you could create the Lou Bega, you know Sure. Ai about your thing. So there is a world where like actually creating like parodies with these, with these artists could be super fun.

I just, it's gonna be interesting to see like, which artists say okay, there's still a huge amount of anti AI feeling across the, both the ar the music and film and TV industry. Yeah. I'm [00:37:00] just really interested in how this plays out and whether or not the, the labels will be able to like profit off of it.

Most likely they will. 'cause that's the business of what they do. It's just a, one of those things that's like

Kevin Pereira: a, an interesting conundrum right now. But seeing the backlash, which I have to imagine is a fairly predictable backlash. Yes. If you're suddenly gonna take downloads away from paying customers who have invest, literally invested time and money into your platform.

Do you think it was Uud o's hands? Like they were tied here basically and they had to shut it off? Oh yes. I think a

Gavin Purcell: hundred percent. This was not Uud o's choice. Right. That's what I, I, that's what I was saying at the beginning. I think. And again, I don't know any of this. This is a speculation. No. So please.

You know, but I think sudo, that's our whole podcast. Yeah, that's right. Suno and UDO were two companies that came out kind of similarly. And if you go back and listen to our old episodes, maybe I'll even drop a, the episode in there, we did a test of these. And Kevin, do you remember with UDI was able to basically recreate where weird Al's voice.

Like I was able to literally, yes. Oh yes. Get weird. Al to sing

Kevin Pereira: a song. So I remember, dude, I sent that to so [00:38:00] many people and I was like, uh, they're cooked. This is it. This is gonna be, and I thought, I really thought I would wake up the next morning to see like, I don't know, MS nbc, they're down. Yeah. Like talking about like, oh my God, this guy's gonna sue and this, and it kind of became nothing.

But maybe it became what we're talking about today.

Gavin Purcell: I think that's what it was. Yeah. I think, and I think why that happened, and you know, we spent a ton of time on this already, but why that happened is that the music labels see the future. They see the idea that people wanna create more stuff. They see the idea, we've talked about this personally and privately, but also in talking about it and then.

A lot more people are gonna be creating now than just consuming. And so they want the opportunity to really create more. In fact, I have a mini theory that I'm gonna say, just to take a second. My theory of what's happening in the world right now with AI tools is kind of a callback to what happened in the nineties with remix culture.

Um, I've been thinking a lot about this idea. Most people in our audience may not be old enough, or maybe they are, I don't know. But in the nineties, if you remember sampling and an artist like with the Beastie Boys or rap artists, or even Beck people that were coming up, or the Dust Brothers or or Fat Boy Slim, [00:39:00] they were creating music based on other people's music.

I think we're actually gonna see a world now where that's going to be a vast majority of music ultimately, right? I think 'cause now you have the tools to do it. Everybody has the ability to do it. And I'm not saying it's like you're gonna take pieces of popular songs and make that the main thing, but you may have a system where you have like six samples in front of you from different artists and you could put them together in your own unique way.

That feels like the future of music to me in a big way.

Kevin Pereira: And, and we're seeing it now arguably, like how it could come to form with, with Sora and the character cameo feature, for example. Right? That's exactly right. You do the Spotify version of that where I say like, here's me playing guitar with my unique sound or tone.

Here are the kind of runs that I do. And someone goes, oh, I really want hear that over that person with Luga person's drumming. I wanna hear that with Luga, with Lou Bega. Yes. And I actually want the sound of the guitar to be replaced with Lu Pega's voice.

Gavin Purcell: Oh, that would be interesting. So it's doing mamba number five, but instead it's a guitar strum kind of vibe.

That's

Kevin Pereira: it. Yep. You're getting

Gavin Purcell: it. Okay. And if you're put that

Kevin Pereira: together, by the way, when you're done writing your haiku, I would love [00:40:00] to hear that.

Gavin Purcell: Put that together on ear. All right. Couple more big things. Cartia. Kevin Cartia is a AI audio model and we are very aware of them. We have, uh, played around with them for, and then.

They have released a new model. First of all, they raised a hundred million dollars congrats to Cartia, but they've released a new model called Sonic Three, and it seems like it is very close to the cutting edge state of the art, if not there. And Kevin, if you want to, maybe we can play a little bit of what this sounds like.

Yeah. I'll to give people a sound.

Kevin Pereira: Well, I'll play from their announcement video. Gavin, they had a fun little audio test. See if you can figure out, uh, which Elon Musk is real musk and which is AI Musk. I, I, I don't really have a business plan. What's a business

Gavin Purcell: plan?

Kevin Pereira: Okay, I'm gonna play those one more time for you.

Okay. Okay, here we go. These are, uh, which one's real? Which one's fake? Elon. I, I, I don't really have a business plan.

Gavin Purcell: What's a business plan? Okay. I say I know neither of them are real, Kevin. 'cause I read the article. What? No, you got it. You're too [00:41:00] informed. That's right. Those are both cartia and I think.

It's just another example of a very, uh, advanced AI audio model that is starting to get closer and closer to real speech.

Kevin Pereira: I, I played with it quite a bit. Yeah. You, I mean with, with 10 seconds of audio you can make, uh, literally that's all they give you. You can make a pretty convincing clone. Yeah, you can.

Um, it doesn't have as many prompts, uh, for steering the emotion or, or like, um, the ticks of a human voice, like a, a 11 labs, flash V three, for example. But you can prompt laughing, you can kind of steer emotion a little bit, and it is fast. Like when you click the button to generate the speech, it is there.

And that is important for literally any and every application of voice ai. I don't care if you're doing like. Deep dive adventure games like we're trying to do with end then, or if you're just doing a tech support bot, like it needs to be quick. Yeah. And now it can sound way more human.

Gavin Purcell: Yeah. They actually had another demo on their, on their video where they had like [00:42:00] a, a support bot interaction.

Yeah. And like, you know, it's interesting. What's gonna be interesting to me about support AI conversations is I'll be very curious to know if there's ever a Turing test moment for this. Because as of now, even with that conversation, they play. Um, it does still sound like you're talking to a not real person, right?

Speaker 5: Yeah.

Gavin Purcell: When we get to the point, and by the way, I think we will get there first with AI audio models because it's already so advanced When you get to a point where you are talking to an AI and it does not feel at all like you are talking to an AI in that way. That will be a quite a day and I don't think we're like Sam's thing of 2028, like I think we're probably like a year away from that experience, which is really fascinating.

Yeah, I, I would agree. Alright. There's a couple other big stories. Very fast Cursor 2.0 has launched. This is cursor's new. Custom coating model and also Windsurf is launching one of these as two. Kevin, you are a big vibe coder, you know, uh, kind of what is good about these things? Tell us why these two companies are creating their own custom coating models outside of [00:43:00] OpenAI or clo.

Kevin Pereira: The call is coming from inside the house, Gavin, that is why these are, these are the, the, the two, uh, two of the biggest, um, uh, coding platforms, basically saying, we're gonna use our own models. I think they both took existing open source models and kind of fine tuned and reinforcement learned and distilled their own version of it.

But nevertheless. They both seem to perform very well. Um, they perform better than say, G PT four or Sonnet four. They don't perform as well as GPT five or SONET 4.5. They're, they're just shy of it in the benchmarks that are available, but cheaper, I would assume. But they're much cheaper. They're way cheaper, right?

Yeah. And much faster in some cases, like four times faster. Oh, wow. And, you know, there are some times where you do need the most cutting edge oracle of an AI to help you architect or engineer or read through documentation, find a, a unique solution to a, a complex problem or a rarely seen problem, but for 80% of the code that most people are churning out, that is like, you know, [00:44:00] copying and pasting and sort of reconfiguring something that someone has kind of done before that speed is so important.

And that cost is so important because you can afford to pull the handle four times in a row or send off multiple agents to try multiple approaches. And if one of those four works, then you're, you're great. You just saved a boatload

Gavin Purcell: of time. Let me ask you a question about this. As somebody who's spent a lot of time doing this sort of thing, because the thing I think about when my little bit of experience with live coding has to do with the level of frustration you get when something's wrong, right?

Yeah. And what, and what I wonder with coding, particularly if the actual edge and the actual money will always go towards the best models because the time saved, rather than having to pull four slot machines and one of them works, if one just works, is that not worth paying more for that result? I guess that's my question.

And by the way, we all assume that like it'll get better and better over time, but for right now, I don't know, like is a GPT-4 0.5 [00:45:00] version okay. Enough if it fails a bunch of times or a couple times, yeah. Whereas a GPT five one doesn't fail. I don't know.

Kevin Pereira: Yeah. I mean, look, there's a, there's a time and a place, right?

Yeah. Like even when we're, when we're doing and then, and we need an agent to quickly render an evaluation on, did. Did the thing that someone say, did that, was that a yes or a no? Was it a good or a bad? Was it a correct answer or not? We don't need the power of a GT five to spin up multiple agents, right?

So if I need to go and make a quick change to something, change the color of a thing on a website or go and right, oh, this, these, these 10 lines of code are probably wrong. The time. And again, the money that it would take to have the bigger model go after it, eh, not there. But again, if I could spin up a bunch of subagents to go pull the research, start building the little components, and then that's interesting.

Ship into an oracle and it can orchestrate it and bug fix. There's, I think there's really real value in these models.

Gavin Purcell: Yeah, that's an interesting idea of like subagent that don't have to do the hard work, but they know how to do the other stuff. But speaking of hard work, Kevin, there's a new company [00:46:00] out, well, it's not a new company, it's been around for a bit, but Tropic AI is a company that has been working on an entirely different way to serve AI models.

And, and Kevin, I regret saying we should cover this. I regret it right now. This is, this is complicated. It's a complicated story. I just want to be clear. These are a couple guys, or a, a group of people that have been working kind of in the background for a while. There was some kind of like semi shadiness around this.

No, no, no. I, that's just kind of people were wondering if this is even possible. Yeah. But they've just come out with a new, uh, video and o an open source product that they're going to sell. Kevin, tell me in the most simple terms. I know, I know. Yeah. Why? This is interesting. I I know why it's interesting.

Let's talk. That's an easy thing to figure out. You can say that, but what they're doing here, everybody, this is Gavin from five Minutes in the future, I just want you to prepare yourself for what is a very dense conversation around something called p Bits and Probability. Alright, we'll be back. Here comes our conversation right now.

Kevin Pereira: Yeah. So, okay. What they are doing, um, whenever you are doing something with a lot of these AI [00:47:00] models, um, there's probabilities involved. For example, you're predicting the next word in the sentence. Right? That's a very, very simple chat completion type thing. We say it's simple now, but it. It's using a ton of powerful compute on these GPUs, these graphical processing units to try to roll the dice and figure out the probability that the next word in the sentence, the car is what?

Blue metal on fire. On fire. Sure, sure. Uh, abyss, no abyss doesn't make sense. Sure. Nos doesn't make sense. Throw that out. Yeah. Right. Yeah. And it's just doing that and, and it's trying to. It is trying to brute force randomness and, and, and check these probabilities. And so what they're doing with their, uh, p bits as they're calling them, their prob uh, probabilistic bits is to try to build the randomness into the hardware itself.

So you can actually control the randomness of a bit. And you can like, put weight in there so that these probabilistic models that require dice rolls [00:48:00] or randomness or math to solve ai. What's the probability that a cat looks like a cat or that a face is in this position, or that the next word is blah, blah, blah.

These, there's a lot of probabilistic math involved. They're baking it into the hardware. Okay. Hold on, hold on. Hold second. One second. One second. One

Gavin Purcell: second.

Speaker 5: Yeah,

Gavin Purcell: we'll cut this into the very top. Everybody, this is Gavin. From Five Minutes In the Future, I just want you to prepare yourself for what is a very dense conversation around something called p Bits and Probability.

Alright, we'll be back. Here comes our conversation right now.

Kevin Pereira: I apologize to Tropic for trying to break down what they're doing. 'cause what if, if what they're doing works, why does it matter? Gavin? Yes. Massive, massive reductions in the amount of energy required to crunch these models. Yeah, and so, you know, like.

We had sort of hoped that all this discussion of these, uh, data centers requiring their own linear nuclear reactors or whatever else to push towards solar or get the dinosaur juice outta the ground, whatever we have to do to fuel this AI revolution, we were hoping [00:49:00] that advancements along the way would reduce that overhead.

Yes. Now maybe they'll just increase the demand for it as well, like that, that could happen as well, but at least the stuff that we're doing today might require, uh, a significant percentage less energy.

Gavin Purcell: But even if they increase the demand, if this company's promise comes to fruition, which they have set a 10,000 per, I think 10000% or 10,000 x, reduce a reduction in energy consumption.

That is how you get to a world where all of the stuff we want to do exists. Sam Alman and other people have come out and said they were already compute constrained. When you hear that term, what that means is they do not have enough energy or chips to be able to run these models at, at the size they want.

It also might mean that they are cost constrained where it's too expensive during that many, but the whole reason we are building all these data centers is 'cause there is a belief across the board, whether true or not, that we are going to have lots of business running these models over time. If this system can make these, uh, models significantly, uh, less [00:50:00] expensive to run, or the compute, much less the amount of progress we could make very fast in terms of, especially with an automated AI researcher.

Would be remarkable. So it's another aspect of the AI space. Yeah, it is an aspect of the energy slash how these chips see things space. It is very interesting that we'll link the video down below their launch video. Think

Speaker 5: of the dog videos. Yes. We are going to be able to generate

Gavin Purcell: Gavin. Yes. Yes we will.

And by the way, speaking of. Think of the real time AI videos. We'll be able to generate Kevin, because there's a new cool tool, an update to a company called Odyssey, which we covered here, gosh, I don't know, maybe six months, nine months ago. Odyssey two is out, and what this is is a, it's a kind of a window into a world and what you're able to do is prompt in real time and change the video that's happening in front of you.

Yeah, it is. I would say. Semi limited into how good it is, but we both spend time with it. It is super fun. It's still in that kind of like toy phase. But yeah, this is the promise of like [00:51:00] Genie three or all those world models are starting. This is the beginning stages of it. Kevin, do you wanna kind of walk through what, yeah.

What are we looking at here? What is this that you, that you've made?

Kevin Pereira: So this was on my cell phone, not even logged in. Right? Running in a web browser, it connects you to a starting scene and you can prompt your own scene or you can pick from a prebuilt one and you're watching a realtime video, uh, stream.

I, it feels kind of like it's a looping image to me a little bit. Sure. But, um, nevertheless, you're watching this realtime video feed and you can tap it. Uh, it will make a little magical effect appear around your finger and then give you suggestions for things. Um, or you can type your own unique suggestions.

So for example, I said, show me a hotdog on a city street. And what I got was. POV footage basically of a handout stretched, holding a hot dog, a very steamy dog on a city street. Like we're okay, that's pretty good. Sure. City background. Yeah. Yeah. Okay. Then I said, you know, I think I said like, uh, make it fly up or something like that, or make it fall to the ground rather than falling to the ground.

It sort of hovered up into the air, and I was like, all right, that's [00:52:00] fine. Then I started pushing it and said, okay, set it on fire. It did, it added flames around hot it, it, pretty hot fire, actually. It's very good at fire. And then I was like, okay, turn it into a hamburger. It kind of did it. Yeah. The hamburger was still on fire, and so every step of the way, it's almost in real time, adding the effects or modifying the scene in the way that you asked.

So I started really pushing it, turn the hamburger into a hamburger, man. Make the hamburger man, bite his own hamburger arm, give him wings and let him fly over the city. Make him land on a taxi cab. Right. I I used it to generate hallucinating gangs of babies on bikes that pull out guns or joust with knives.

Um, two varying levels of success. Yeah. It's not difficult to look back at earlier video and photo generation. Yes, yes. From less than a year or two ago and go, oh, this video kind of looks like those models. Oh wow. What is this gonna look like when SOA is real time like this?

Gavin Purcell: That that's exactly, and what I was thinking when I watched this is that, you know, we, uh, pers Beats who's an AI artist that we're both big [00:53:00] fans of, that we had on the show way back when was doing those interesting kind of looping videos back in the day.

Yeah. When I say back in the day, I mean like literally like 18 months ago, we've been doing the show for a couple years now, and what's interesting is to think about like this now looks kind of like that, except that was all pre-rendered and you had to kind of like send it away and you got a video back.

Now we're talking about this happening in real time and you get this kind of option to do this. Yes. I think Kevin, this is the most important part to think about is you now have Sora two, which is that version of real, it's that send away and get a video back. That's very realistic. You can put yourself in and do all sorts of stuff.

You definitely can see a world where we are entering very soon, where you will have a real time version of something like Sora two that could be ongoing and that is kind of what Genie three from Google is trying to do. Yeah. I think we are like two years away from that being like a real legitimate, viable, like everybody every day use it product and that is crazy.

Kevin Pereira: Uh, crystal Ball Valenzuela, who's the CEO of Runway, he said, [00:54:00] what is it? Probably about a year ago, he says he imagines a near future where this stuff is rendered in near real time. Yeah. So that you can evolve it and whatever else. Like this is not Runway's product. We know Runway was working on world models and really fast, quote unquote real time models.

They were teasing it, they were talking about it. So I'm excited for the next. Foundational leap of this. Yes. And then, okay, what's after that? Controllable. Yes. Controllability. Yes. Let me put myself into it and move around the world. Instead of prompting it, let me just press the button to interact with it.

Not gonna be far off.

Gavin Purcell: Yeah. And Kevin and I were at an event recently where we were speaking to a bunch of people in the video game industry and one of the things that came up there was this idea of what does the future of video games look like when worlds are all generated in real time versus pre-programmed?

That is a massive shift for that industry, right? When you think about the industry that has been created on engines that exist pre rendering information or, or, or rendering stuff in real time, but knowing what those things are. If the [00:55:00] user or the player starts kind of like driving their experience in real time, entire new genres will come out of that right?

Entire new kind of ideas of how these scenes can be used. That is not to say that like there's gonna be so many jobs for people to understand this stuff. Yes. So if one thing you're out there thinking and you're in the game industry is like, God, it sucks that this game industry blows. This is the place.

Find your time here. Because if you understand how real time generation can work, plus what are interesting video game, uh, uh, actions or things you can do within games, you will be set up for the future. So, well, it's gonna be amazing. Hey Gavin. Yeah. Hey, I see what you did there.

Speaker 5: Sometimes ya rolling without a care.

Then suddenly you stop and shout.

Gavin Purcell: Alright everybody, this week we've got three [00:56:00] quick things that people have done that we love. First of all, our buddy Fabian Seltzer, who again, somebody we've been following forever, works at a company named Gly. You should check him out, has released a new agent called I Hollywood. And what this does, this works within the gly platform and gly, you know, was originally this thing for people to build their own, kind of like sta scaffolding around stuff.

And people can still do that. What they've also done now is built these really interesting tools that use a bunch of AI things to pull them together. Bobby and built a thing using one 2.2, uh, animate, which we use for our launch video for, and then where he basically has built an agent that lets you take an image and a video of yourself and replace it with somebody else entirely.

Meaning that you can now with a single agent, uh, create a video that will then be transformed into somebody else. Fobbing in this video, turns himself into a, uh, female gamer with cat girl headphones, and then at one point he takes off his mask. He walks through this video and shows you how you all do it.

I have not used this yet. I know that that glyph model in part is you have to pay for these things as they come through. I don't know how [00:57:00] much it is. It's probably a couple bucks to do one of these, but still have. This is a really cool way of looking at how to do the stuff we've talked about, which is like face swapping or interesting stuff.

Yeah. But like do it all automatically, which is amazing.

Kevin Pereira: Yeah. And, and interesting the, the pipeline of just use your use chat, use basic commands to talk to the tools and have it intelligently stitch things together. Yeah. I mean, we're, we're fans of Glyph, we're fans of the whole team, so, uh, definitely something people should check out, but what's going on with the bots in the hall?

Gavin Purcell: Okay, so this is a very cool story. You know, we love the AI creators, like neural vis. This is a guy that I've discovered on Reddit. He shared his video there. And what bots in the hall is, is a fully AI generated, human created animation series. And the reason I wanted to shout this out is, first it's very funny as many human written things are, he's got a good sense of humor.

It feels very adult swim like. But if you are creating AI animations, you know, there are all these sorts of issues to try to like consistency, everything like that. Bots in the hall is the best looking [00:58:00] adult swim style animation show I have seen. It is very funny. The videos are like five minutes each.

He's been uploading them to YouTube on and on Reddit. He's also on X, but like you should just go watch these and like this again, is another one of those things that is the promise of what you can be as an AI creator. I was really impressed by this. He's also very cool in his bio where he says like, look, I am a human that makes things with robots.

Which is a kind of what neural viz says too, and I think the more that individual creators start doing things like this that can stand out, I think is amazing. So go check these out. They're very funny and just very well made. When

Kevin Pereira: do you think Gavin, even in the description of this, it says comedy short, 100% self written, directed, edited, and performed.

But I do use generative imaging for the visuals and for voice filters slash changers on the performance. If that's not your thing, I totally get it. If it doesn't bother you, enjoy. Yes. Which is a great way to be straight up about what the thing is. And hashtag AI is in there as well. But when do you think, or do you think we'll see a point in time where people can just [00:59:00] make a thing and they don't have to disclaim necessarily the tools that they're using or try to, you know, go please.

Gavin Purcell: I think people are already doing that and they're just not saying it right. Like I think what's good, what I like about bots in the hall is he's being very clear, like neural visits that he's using these tools. And I, I, by the way, I will say like if you're able to do that, that shows you have the skillset.

But I think in the world at large. There are probably people already doing this that are just not saying it. I would imagine that's the case. Now, granted, if you work for a giant corporation or if there's a company that you know has a a, it may get you in trouble. So there's people that are very cautious about what they say or what they don't.

I think it's just gonna be a lot more. I, I think to your point, just to say like, this guy particularly, I'd say in a year, it wouldn't even matter. Like one of the big things I think is that generation Z, the p you know, 18 year olds and under, they don't care that they're creating with ar i with AI or not.

Some of them might, and by the way, again, I don't wanna in any way sort of make people feel that, that I've totally legitimate feel if you wanna organically create grape, but like, I just think it's [01:00:00] gonna change soon. You're not

Kevin Pereira: blanket stating here, but you are pointing out a trend

Gavin Purcell: as you see. Yes, that's exactly

Kevin Pereira: right.

That's exactly right.

Gavin Purcell: And finally, Kevin, the last thing I wanted to point out was just one of the funniest videos I've seen in a bid. There's been a lot of like, really interesting AI music things that have popped up in major ways, like we talked about the soul hip hop covers and all that stuff. This person has been uploading weird AI videos, but this is just a video of what if Michael Jackson trained Anne and Skywalker and it's taking, you know, uh, episode one, uh, footage and reusing it for partly to Couldn.

Imagine if Michael Jackson was a Jedi and he was the one responsible for training kin. And there's just a lot of funny moments and it, it's a pretty short clip, but it really did make me laugh. What, what, what, what is, is

Kevin Pereira: the algo surfacing this for you, Gavin? What? You mean, how do I find this stuff? Yeah, what is going on

Gavin Purcell: in your timeline?

I just am that person, Kevin. I find all this stuff in my brain. I don't know, but this is what I love, right? Like this is the creative use cases of ai. This is also an example of what I was saying before about [01:01:00] remix culture, right? Yeah. What makes remix culture so much fun is we have a collective understanding of something, and your collective of Michael Jackson probably involves a lot of things as we know, but this and, and Star Wars episode one, they're both these kind of divergent ideas, but when you put them together, it's not like it's a, it's, it's definitely not gonna win an Oscar.

It's not gonna be like, it's not like an amazing piece of art. But it is a fun moment to see somebody take two ideas that exist. Yeah, put them together and then wrap them in a, what I believe is a very, very funny idea. So anyway, I hope more of this stuff starts to happen, and it might be tied into the fact that we have digital rights to things now, and because YouTube has figured out how to get money back to content owners, how to get money back to, to music rights, maybe there's a world where the AI space can do that as well.

I, I hope so. And I think maybe

Kevin Pereira: that's what Sora could lead into. And in just about a year's time, for 4 99 a month, Gavin, you could have an X one robot, learn the baseline. To Annie, are you okay? Ooh. And I could meet my own band. Me, Anakin, Michael Jackson and my X one [01:02:00] robot altogether. You start slinging some pizzas, you put some old arcade games in there.

Suddenly your house is actually really suspect. I wouldn't, oh, no, I would not. Like I was gonna say,

Gavin Purcell: suddenly I thought you were trying to set me up for the stupid show. BZ Pizza, Chuck E. Cheese. That's what it was. Yeah. Yeah. It was like

Kevin Pereira: you're setting yourself up for a chicken cheese, but if it's in your house, that's a little, I gotta tell you, it's a little weird.

I'm not letting the, the niece and nephew come over, guy. I'm sorry. All right,

Gavin Purcell: everybody, that's enough for now. We'll see you all next week. Thank you everybody for joining us. Bye-Bye. Bye.