Transcription
OpenAI and Sam Altman are gearing up for, wait for it, hot AI summer, LFG, apparently. That's right, baby. Hot AI summer coming in. Compute. Go GPT5 coming up.
Hey, Kevin, buddy. Hey, what did we say? What did we say? Don't go beast mode on intros. That's right. No beast mode and intros. That's right. We're both over 40 now. What did you say about GPT5? OpenAI is preparing to release GPT5, and they've just released a new note-taking app, and Codexes can access the internet.
Kevin, is this going to be how the Open Eye movie will end? Uh uh uh uh Kevin Gavin, are we going beast mode? No, Kevin, the Open Eye movie is actually for real. You know, the movie I want to see is the decades-long beef between Mark Zuckerberg and Palmer Luckey. They recently squashed that beef because they're teaming up to make super soldiers. That's right. And Epic CEO Tim Sweeney says that a team of 10 could make Breath of the Wild with AI. Oh, I'm sure that upset nobody.
Also, some massive AI video and image tools being released from Flux, Luma Labs, and Hey Jen. Hey, Jen. No, that's what I'm shouting at my bestie across the bar, and it's Martinia O'. Hey, Jen. Put your Birkin on the hook and let's down these mimosas. Hey, Jen.
Also, Gavin, you're never going to believe this one. Record labels are signing deals with AI music companies like Sunno and Udo. What? Who would have guessed? But more importantly, Kevin, V3 has brought us something incredible. It is teaching cats how to swim. The training regimen is rigorous. 6 hours daily. No breaks, no excuses. Uh, okay. On God, deadass cat swimming is like peak GPT5 core energy. It gives. Yeah. Yeah. This is AI for humans. We're busting for real. No, no, no. That was a test. No beast mode in intros. Fine. Fine. This is AI for humans, everybody.
What happened in the canon of this show that I'm the principal? When did that happen? Yeah, I don't know, man. You're just the cool kid who wants dank memes, and I'm the one who wet blankets here. All right, start the show. Welcome to AI for humans, everybody. And Kevin, it is hot AI summer. LFG, according to Sam. LFG, baby. Let the farmers grow. That's what that acronym means. Oh, is that right? I didn't realize that. That's interesting. Let's freaking go, baby. Sammy Altman trying to just chum the waters on the heels of Google IO stealing maybe a little bit of thunder in their pretty little Gemini jar.
What does hot AI summer mean, G? Okay, so there's a couple big things to get into here. The biggest thing so far I think that we have to think about is GPT5, which is their huge next new model that is planning on coming out as rumored in July. Now, these are rumors. Nobody really knows for sure, but like that is a big thing coming up. Sam has also, you know, taken uh to the world of interviews and has been doing some interviews in part, I bet you're right, to kind of like re-trigger the uh attention back to the OpenAI mothership. It's the most exciting thing that's probably in the pipeline right now. And I think they thought maybe the Sears portrait studio photo shoot between Sam and Johnny IV uh would be something would would capture some more headline news cycles. But uh well let's let's see what what Sam is is talking about here. The the models over the next year or two years are are going to be quite breathtaking. Um really there's a lot of progress ahead of us, a lot of improvement to come. And like we have seen in the previous big jumps, you know, from GPT3 to GPT4, businesses can just do things that totally were impossible with the previous generation of models. And and so what an enterprise will be able to do, we talked about this a little bit, but just like give it your hardest problem, if you're a chip design company, say, "Go design me a better chip than I could have possibly had before." Oh, what's it going to do? Put Taki powder on it. Oh, that sounds interesting. How much better How much better can chips get? We got salt and vinegar, right? We got flaming hot. Not the kind of chips, Kevin. That's not the kind of chips. What else do we need? You need to go Pringles. Pringles are still a chip; just cuz they have a funky can, it's still a chip. Um, if you're a biotech company trying to cure some disease, say just go work on this for me. Like that's not so far away. Uh and these models ability to understand all the context you want to possibly give them, connect to every tool, every system, whatever, and then go think really hard like really brilliant reasoning and come back with an answer and and have enough robustness that you can trust them to go off and do some work autonomously. Like that that I don't know if I thought that would feel so close, but it feels really close, Kevin. It feels so close to me right now.
As the Calvin Harris said once in a very famous song, Sam is out there, you know, pitching his wares. Again, I think just to to extract a little bit about that, what he's getting at is this idea that you're going to send off your AIs to go think for you, right? And a lot of the stuff that we've seen coming out of these thinking models is is a big deal. First of all, the leap from GPT3 to GPT4, that really paints a picture. Uh if you're new to this podcast and potentially newer to AI, you don't know the world before like chat GPT, the difference from three to four was immense, you know, and we we're now getting used to like every few months there's a new model or a new update or a new trick, a new something, but like it was a massive change in capabilities, uh unlocking a ton of new use cases, uh spawning thousands of businesses, crushing a few hundred in the process as well. But like that was massive. So that on the one hand is gets me very excited and he's been saying that for a while Gavin like so that is the generational leap in capabilities that I am expecting. This is not a slightly new graphics card. This is a new Xbox. This is a whole new thing. It's the next generation. However, on the other side he mentions stuff that we are already starting to see which is these agentic behaviors. Send it off. It has access to tools, MCP servers, whatever else. It has tons of context. Well, we're seeing million token context windows. So, on the one hand, it's supposed to be a complete sea change. Uh, on the other hand, it sounds like uh it's another iteration of stuff that we're already seeing. So, are you expecting a 3 to 4 leap in this 4 to 5 or are you uh expecting that it'll just be, you know, more of the same? Will you be disappointed if it's I have two kind of theories on this. I can't.
So, 3 to 4 is, as Kevin mentioned, is something that a lot of people have always said was like one of the biggest leaps, right? And we haven't really seen a leap like that. Although I would argue if you take the reasoning models and compare them to GPT4 that we have right now. That is a very big leap. And maybe if you weren't from right from that to the next thing, it would feel that way. There is a conspiracy theory that is bubbling in my brain. And I don't know if this is real or not in any sort of way that they put out 4.5 GPT 4.5 to kind of lower expectations so that they could then surpass them again. Oh, you think they released a model to kind of sandbag a little bit? they wanted to get to the top of a chart because they didn't make a big enough deal. didn't make a huge deal of it and and one of the interesting things Andre Carpathy who we've talked about on the show a ton former Tesla and OpenAI engineer very good follow on X and does a lot of like educational stuff talked about the idea of 4.5 which is this massive model worked very slowly many of you who might have used it know that it's a very creative model but if that is the training data now for the next reasoning model he believes the leap could be very significant because it is a much bigger model so what's going to be interesting to see is is GPT5 5 specifically this kind of unified model that will be the next version of the training. I think they're going to try to clean up all their naming stuff. I'm really curious to see if it's much better at the reasoning side of it because it's based on this larger model. I do think Kevin, it's important to also shout out this other thing that Sam said because in that clip he seems very optimistic about how great this is going to be. He also then went into another interview where he dropped this. We'll just play this little snippet of this interview because it's a very different tone. I I think there are going to be scary times ahead and there's going to be a lot of transition. I'm very confident this can be managed. I think the more we can all do it together, the the more likely we'll get something. Who's managing it, Kevin? Are you managing the scary times ahead? I I guess I'm going to ask the agentic AI to manage it and I'll go away for a few weeks and when I come back, I hope it's managed. Yeah, it's So anyway, this is like kind of speaking out of two sides up. Again, I think the biggest thing for everybody at home to know is like we won't know until this comes out what sort of leap it is. I do believe if it is a significant leap, if it is close to three to four that Sam said, every conversation you will have will conceivably be about AI in the next year, right? Because it's that big of a deal. I'm sure my friends love that every conversation they have with me is about AI. It doesn't matter if we're on the pickle ball court or we're trying to watch a movie, it's going to go back to AI. But I think you're absolutely right because you know like the release of chat GPT was a major cultural uptick in people discovering wa what is this thing and that seems like nothing in comparison to what these models and these tools do now. So if we get what feels like a generational leap come July when this is rumored to be released. Yeah. All bets are off. Everybody is going to be screaming from the mouth.
Yeah. And you know there's already a couple small things that are dropping soon. Supposedly OpenAI 03 Pro according to Sam Altman is coming soon. This has been rumored for a while. In fact, it's delayed versus where it was supposed to come out right now. I'll be curious to know if like that comes out prior to this. It it might. But also, they've now unleashed Codexes on the internet, which is their coding platform, which means that your your AI coding tool can now directly access the internet, which you know, one of my favorite things about this tweet from Sam because Sam, you know, took some time to tweet. He put he took like 10 days off from tweeting and then started tweeting up a storm. In this tweet, he says, "Codex gets access to the internet today. it is off by default and there are complex trade-offs. People should read about the risk carefully and use it when it makes sense. So, one thing to know about Codexes getting access to the internet is like this is always the question of like what does an AI look like when it has full steam ahead on being able to read everything and do everything. Yeah, I mean that's true. But I look as excited as I am, it's finally now at par with other models that can code and had access to the web before. So, it's like other companies are facing that, you know, that head on. I don't know if Codexes can do things they can't yet. I haven't put it through its paces. But one thing that did happen, Gavin, we're recording this uh surprise surprise on a Wednesday and chat GPT for business just got a ton of updates. This might get in the weeds a little bit for some if you're just like a personal user and and this will have benefits for you down the line, but for right now for the business users, they did a couple things which are pretty massive. One, ChatGpt can now natively connect to all of your databases. And don't get me wrong, this is very quickly going to trickle down to the individual users, but we're talking, you know, Google Drive, Dropbox, uh if anybody uses one drive. Okay, I know what this is going to be searched for. We already know you've got all these databases you've been generating for 20 plus years of uh fictional character feet that you are now very excited to be able to search through and surface whenever you need them. And I need it to if I say I want feet in jam or jelly, it needs to know the difference. Good point. Okay, good point. And if it's a if it's a character stomping on lasagna versus stomping on grapes, it should know the difference. Okay, and trolls are different. And it can do it right. I don't have to label my directories. You know, I'm meticulous. Point is, now it can connect directly to all of your files. Now, why does this matter? Well, this gives uh for individuals it eventually give it way more context on your life, your businesses, your creative pursuits, etc. On the enterprise level, I have been and I know you've been asked to consult on this too, Gavin. no shortage of companies screaming, "How do we integrate all of our data into a system that spins up, you know, strike teams that work within a division and they come up with tools?" And then you start building workflows and pipelines and looking at 16 different software tools that can bolt this onto that. And while OpenAI's demonstration today seems pretty rudimentary, you connect it to a thing and you can query it. It does respect the permissions of the users, which is massive in an organization. So an intern doesn't have access to like high-level engineering docs or something like that. um potentially within a household you can share documents but if you are keeping information a secret like your feet pics repository you don't have to worry about the miss have a separate account that why not right look maybe you do all of your work inside of a aluminum foil line Faraday cage maybe that's where you store your thumb drives okay and she doesn't need to know what happens in there but this is big that this is rolling out along with MCP capabilities which we've gotten into the weeds about MCPS in the past but just know it it means that it it allows OpenAI's AI to talk to other AIs fairly effortlessly. And so suddenly you can go from a system that has a basic understanding of itself to a system that understands AI systems everywhere and you can get really powerful workflows. Sam has said in the past and this is where I'm hoping to connect the dots again where I'm talking about grandio statements that don't come true. He has said the mistake that many make are trying to uh overengineer and build for the current shortcomings of the models as they exist. This is one of those things where so many companies and individual creatives or contractors have gone out and been like, I'm going to engineer a thing that lets an AI talk to all of your documents. Well, guess what friends? You know, Google has their version. Amazon has their version. OpenAI just rolled theirs out now. And when it trickles down to the individual user, that could be the Kleenex moment for AIs that that talk to all sorts of and maybe even more importantly for these AI companies, the pay for the Kleenex moment, right? Like where the idea is that suddenly instead of a because a lot of studies about like how people are using AI in business and like a lot of experimentation but not like a lot of money that's set aside and we know that both Anthropic and OpenAI's, you know, business model has gone up and they've been making a lot more revenue. This could unlock even huger amounts of revenue because if you can imagine all of the businesses in the world if you're suddenly locked into an OpenAI ecosystem that's a big deal when all your data is there and I think that is going to be the fight going forward. In fact, there's a story that kind of tangentially connects to this. Claude and Windsurf are kind of fighting a little bit. This is a vibe coding platform that we know that OpenAI is buying that they were buying for $3 billion; they basically lost access to uh their specific uh fast pipeline to Claude models and the CEO came out and said kind of a long description of what this is. They don't really know why, but Kevin, this definitely feels like some drama stuff that has happened around these two companies because of like lock-in, right? We talked about with Google's like very expensive AI ultra plan. Part of the goal here is to try to get companies to lock into a specific OpenAI ecosystem. And you know, if Windsurf has access to all of Claude, it might mean that like by cutting Claude off, Anthropic seeing this and being like, we have to get locked in for our stuff as well. Well, I don't know if you saw the other rumors that were swirling about that uh there might be an Anthropic Apple collaboration going on, which could have something to do with that. They're they're talking about Xcode, which is uh Apple's coding platform. I use it all the time. I am astonished at the lack of AI in it. even Android Studio if you're building Android apps or anything in that environment at least it has Gemini built in now and is getting better so Apple needs their solution anthropic and Claude makes a lot of sense and that might be some of the reason there it could be the OpenAI deal as well so you know it tough to say but really interesting the way everybody's going for lock-in and again to the point of connecting to all of your documents that plus memory equals a much more difficult switch for an individual or an enterprise when they're six months down the road and this AI knows everything about them for them to leap to another AI might be really tough. So that helps for the lock.
And the only other quick thing that came out of this business thing that is interesting but maybe not fully featured yet is there's a record mode that they are introducing into the enterprise teams and education system where there's going to be a button on your chat GPT and you can record a meeting at least locally it sounds like and it's going to record all of your meeting notes. It's going to be accessible as a thing that you can get into. Kevin, it seems like to me this is a little not fleshed out because the most useful version of this would be something that could be a plugin into say Zoom or to Google Meet because how many meetings do you have where everybody's in the same room? Not that many anymore. But this is a movement towards the same thing. And again, it is lock-in. If you have data from your company that is in one place, then it does become very searchable. On the other hand, it also kills conceivably a number of AI startups. Like we have said again and again, like there are AI startups out there that are trying to do this across many different platforms. If each of these suddenly has their own version of this, sorry buddies, that might be the end of it.
Well, listen, I'm a Nathan Fielder fan. I promise this will connect. You know, uh, taking notes on a meeting is one thing, but rehearsing the meeting in advance with fake AI personas of the people you're going to meet, Gavin. Well, that sounds a little different, right? Maybe Black Mirrorish. Well, Madna, CHRO Tracy Franklin has something to say along those lines. I've created profiles in a GPT um of our executive committee and I have scenarios of when two people are are maybe at conflict or when I have to go in with an opinion um or a recommendation and how might the the group react to my recommendation or if I'm having a really bad day and I need to understand myself and why I'm triggering. I actually have a completely interactive coach, therapist, you know, and and teammate um for me that I use all the time. It's been like my favorite thing. And I've said, Gavin, circle of trust. Do you do this, too? Do you do you have a Kevin GPT that you run ideas by? Cuz you know I'm going to blow. But now I really do want one. Now I'm going to go out and create a Kevin GPT. How are you going to prompt that, Gavin? I'm just curious what you don't worry about that. I'll keep that secret.
Mean, this is an interesting thing. Again, you are a professional; has been a Ctier Platinum Plus cable television who done it. Oh, no. A footnote in the history of never. Hey, you have a Wikipedia page still, Kevin. That's something. I want you to make sure you put that in your back pocket. That's still something. Please don't update it, anyone. It does not need to know about my RV phase or my convertible hiking pants on television.
Again, this is another thing that shows you how this business side of AI is something you can't discount, because we're here talking about like what's the cutting edge look like, but the money will be made on the business level. Kevin, the other thing that's important to realize about all this stuff is that the dramas around this, there will also be money made on that, because there is going to be a movie made of the open AAI Sam is he or isn't he CEO uh saga that we covered very specifically on the show last year. Andrew Garfield has been cast as uh Sam Alman.
The other thing I really liked about this is if you saw um Anora, the best picture winner, the uh Russian ball guy, he is going to play Ilia, and that guy's amazing. I love that guy in that movie. So, if you know, Ilia Suskgiver is the guy who kind of helped lead this ouster moment and then got kind of pushed back when it he came back. Ilia then went on to start uh and is still in the middle of a company called Safe Super Intelligence, which we have not heard a single thing about for a year, but he and his co-founders and arguably a bunch more people are up in the mountains making super intelligence right now.
Kevin, it's just an interesting thing to see how quickly these things are happening now. While all that drama was going down, like that's what you and I were texting back and forth. And I think everybody in the bubble as well were like, "Oh, this is the third act. No, wait, no, this is the third act. Oh man, this is going to play out so great," because it was pure drama unfolding on Twitter and behind closed doors and in the rumor mill. So, I I'm psyched for it.
Is there anybody else you'd like to see play Sam Alman? You know, it's a really interesting question. I've been thinking about this. I mean, Andrew Garfield had a role in The Social Network, so there's an interesting connective tissue, but of course, the person that most clearly is is um Jesse Eisenberg, right? Sure. Jesse Eisenberg is so good at playing nebishy sort of characters and did Mark Zuckerberg so well in the social network that it's hard to see anyone else. Oh, you know who might be really good? Speaking of Jesse Eisenberg, you know, Kieran Kulkin could be interesting in that role if he was able to do it, cuz he's kind of small and kind of has an but maybe his energy is too much. Anyway, I it'll be interesting to see. Andrew Garfield feels a little too suave to me to play Sam Alman, but we will see.
Oh, Sam's going to take uh take offense. This is not about dissing on Sam. He's a charismatic young man. Sammy is a very suave gentleman. If you want to cast yourself in something, you could cast yourself in the role of helper to AI for humans. That's right. We need you to subscribe. We need you to like this video. We need you to be aware that you are part of our army. You are part of the people that spread our message. Last week's YouTube video was Kevin by far our most YouTube video episode of all time. And that is because of people out there. Really sorry to hear that. It was an okay episode. It was all right. I promise it gets better.
So, if you're a returning champion to AI for humans, thank you so much. Literally, you're the only way Gavin and I are able to spread the message of this thing. So, please share it with your friends and family. Give the thumbs up. Click the subscribe. It costs you nothing. If you listen to it on podcast, leave a fivestar review. And also, huge thank you. There's this like contingent of very kind people who give us $5 a month on Patreon, and we use that to pay for editing and for a bunch of AI tooling as well. So, that helps immensely. Thank you all for helping out this uh this little endeavor of ours. We appreciate it. Absolutely.
And thankfully Kevin and I were able to do this podcast cuz we squashed our 10-year beef just like two other people in the industry, Palmer Lucky and Mark Zuckerberg. Mhm. Yeah. When you backstabbed me for that cute little late night thing and left the great gravy train, we were all chugging along. Boy howdy did I write you off, but we're back together because, you know, time heals all wounds. Palmer Lucky, founder of Oculus, uh you know, amazing amazing engineer. uh created really responsible for VR as it is today. Didn't like create VR but really brought it back into the fold. Um unceremoniously ousted from at the time Facebook now Meta um and was very vocal about it after a short period of time. And I I actually appreciate how vocal he was about it cuz he talked about how difficult his like it was for him emotionally, intellectually um how uh Anderil's new defense company was the let's see if he got it in you twice kid sort of thing. And clearly he did have it in him twice. The company's doing amazing.
Well, you know, a lot of documents have been leaked since uh Palmer was ousted. And he said it seemed like Mark Zuckerberg wasn't really at the helm making the decisions about Palmer's ouster, that it might have been others that are no longer at Meta anymore. And it's a very roundabout way to them getting together, taking a photo in what could be, I don't know, the lobby of a rainforest cafe or Meta. What if that's where they are? I would love to see the two of them in a rainforest cafe like underneath the frog looking up and the frog's just sitting there. Volcano. 100% happy to be a fly on that wall. I would pay for the dinner. Point is, they squashed the beef because they want to make super soldiers.
Um Palmer was touting a new vision system called Eagle Eye for soldiers for a long time that was going to be an augmented reality display that in his vision every soldier should have on their noggin when they are deployed in the battlefield. And the way this works is that it creates like a mesh network of communication helmet to helmet uh where if you Gavin are on one point in the battlefield and you have eyes on an enemy, some sort of vehicle, an interesting point of reference, it can beam that visual data, those coordinates, an outline of another soldier, whatever. It can enhance my vision with the data that you have gathered, with the data that an autonomous drone flying about has gathered and give us a far more detailed uh picture of what is going on in the battlefield. And you know, makes sense. I'm sure it will be great fuel for the bipeedal robots that'll eventually be running around on the battlefield. But that's their vision for soldiers today.
Yeah. I mean, there's two things I want to mention here. One, I I don't know if Kev if you heard this, but there was an incredible interview with the Palmer on the Core Memory podcast. a guy named Ashley Vance who broke away from a major tech publication to make his own stuff. You should definitely be following Core Memory. I listened to the whole two hours and I know you've been a Palmer fanboy for a long time. And I have to say what was really interesting listening to this, you know, I always am like, you know, I'm not super political, but like Palmer was very very on the right and like I was trying to kind of get a sense of like where he landed, what he was this interview with him, it feels very reasonable and I do not think he is like an extremist in some form. That was just my take. I know some people might think that, but I thought it was a very reasonable interview. I do I do suggest you go listen to this. Like it's a really interesting thing.
The other side of this in this interview, he talked a lot about the idea of how these VR helmets need to be different than consumer AI. And he was pointing out the differences between what like Apple has to develop and deliver for consumers to do like the Apple Vision Pro versus what these helmets need to be, right? Because in the helmet, it's really about functionality for soldiers on the ground. And that is a big difference than like, you know, a nerdy guy on their couch trying to have the ultimate entertainment experience. It's all these little things you don't think about when you think about like defense mechanisms and all that stuff. And and you know, the thing that really this stood out to me, if you've been following the news, you clearly saw the the weird drone attack that that Ukraine uh unleashed on on Russia in the last week, which is one of these really fascinating ways of looking at how the future of warfare works now. So when you think about the future of warfare and you look at what's going on now, it is exactly this kind of thing that is going to lead to whatever the next stage of that is.
Now you may not love the fact that like robots will be fighting for us. But I do believe and Palmer even says this in the interview. This is how we save lives, right? You save lives in training by doing this, but also you save lives on the battlefield and eventually maybe it is drones fighting drones and then who knows what that feels like, but it is a very different way of looking at battlefields in general. Yeah. And again, don't take a second to think about what happens when one side runs out of drones. We don't we don't think about that. That's not a thing. It's just metal on metal. It's basically uh what was it? Battlebots. And then everybody goes home happy and we all share our resources because we live in a world of abundance. Also, don't think about what happens when one side suddenly turns to understand how to control themselves versus the human being controlling. All of this is not worth thinking about now. We're not going to think about what this means. We're just going to talk more about uh AI video tool.
Kevin, I I did you such a disservice as a friend and co-host, too, because like you were making like really solid points, but all I could think about was a soldier on the battlefield trying to like pinch zoom a stock widget because Apple Yeah, sure. built their thing. It's like they're under fire. And it's like, no window. Get out of the way. Siri, close the Siri. Oh my god. Can you imagine having to be in the battlefield and using Siri to control it and like you set a timer for like yesterday morning and I can't turn it off, sir. Here's what I found on the web for how to stop the bleeding. is Alexa. It's like here's a blog spot from 2011 about dog fur. We should move on.
There's a big story brewing in the AI audio space which I know you'll have a lot of thoughts on, and the two big players in this space and I I kind of think of Suno as the big player but UDIO is still up there. They're both really interesting AI models. They are in talks with major labels to not only make a deal but give them some percentages of their company. So let us talk a little bit about this and about the history. You don't say. You don't say. It was the I was patting myself on the back as hard as I could. Um I I think you shared the sentiment with me I don't know over a year ago when it was like oh the the labels are going to sue. This is going to be a big deal. And I said the genie's out of the bottle and there's too much money on the table for it to be upset. And I was like, eventually it's going to be a negotiation and oopsie, don't you know it, suddenly the artist rights thing is not going to be a bit they're going to get some pennies, maybe fractions of a fraction of a penny after the labels get what they get, but there was just too much at stake. So, I'm not surprised to see that deals are being made here.
So, you know, something I thought about with this is that GPT40's image gen moment, I think, can be seen by different people in different ways. Obviously, a lot of the anti-AI people see it as like, oh, they ripped off these, you know, the Giblly the Giblly IP and they're making all this stuff with Gibli. Well, the thing I thought about with AI music, and I'm sure you felt about this before, too, is that like I kind of would love to be able to make whether it's like a parody version of something or like use another song as a jumping off point to and like use either that artist's voice or that artist's melody for that thing and remix my own stuff. And it's really hard to do that. You can do that open source. there are ways that you really have to kind of go deep on it. So, I'm curious to know like in the future of this world where Yudio and Sunnio have made these deals. It will be really interesting to see like can I a song that's been like we mentioned uh Calvin Harris's Feel So Close to the Top of the Show, right? Which is a very classic like dance track. I'm I'm a fan of that song. It's super fun to listen to. What if you could take that song and like insert your own lyrics and like make your own version of that, you know, not to make money off of, but you would be able to remix it using Calvin's voice track and maybe the actual sound and the beats and the way the thing rises up. That all feels like super compelling to the mainstream in a way that's different than just making music from scratch. Does that make sense to you?
I think it does. Yeah. And and Mitch Glazer, I want to say chief executive officer of the RAA said, quote, "The music community has embraced AI." I don't know if that's 100% true. Okay, that's what he said. And we uh already we are already partnering and collaborating with responsible developers to build sustainable AI tools centered on human creativity that put artists and songwriters in charge. So that is the head of the recording industry of America. What do you think the percentage these companies had to give up, Kevin? I mean it's so so hard to say because it's you know all of the training data goes into the big pot and gets stirred about and then something comes out on the other end. So, will they say, "Hey, based off a percentage of a particular artist that went into your data set, we're going to do that." Or do they have to build in tools that can reference what percentage of an artist was used in the output? Or maybe they'll just blanket say, "Hey, guess what? You're going to get a Mr. Wonderful royalty on every account that signs up. 1% of everything." that that is fine actually as long as the artists in the end also get their piece because time and time again you know that's the thing we saw with streaming and then you see artists coming out saying hey you streamed 9 million you know sessions of my song and I got 50 bucks that doesn't feel quite right absolutely I think that's an important thing to think about and also like maybe acquisition targets for both Apple and Spotify these two companies okay Kevin another big story Tim Sweeney CEO of Epic Games is coming out and saying some positive things about AI they pressed the AI button and um it might have been a bad idea but like in truth there's no unpressing that button.
I want to read a little bit of his quote here with an interview with IGN. He came out and said that AI characters giving you the possibility of infinite dialogue with a really simple setup for creators means small teams will be able to create games with immense amounts of characters and immense and interactive world. What would it take for a 10-person team to build a game like Zelda Breath of the Wild in the in which the AI is doing all the dialogue and you're just writing some character synopsis. Kevin, what is your immediate take on this idea? I am angry, but I don't know why. And I'm going to throw out my caps lock and just point me at a URL. I'm ready to flame. You sound like most of the gamer commenters out in there in the world. I, by the way, so this is something we've been talking about for a while that this idea that gaming teams will get smaller. I know that a very a lot of very smart people in the gaming world are also tracking this and I want everybody to know that this obviously coming from Tim Sweeney this is a big thing like he is kind of like planting a flag in the ground as has John John Carmarmac as has other people in the AI space. Tim just to be clear is like the leader of really one of the biggest game companies in the entire world. Fortnite, Unreal Engine, all this stuff. So they just had the Unreal Expo which is where they debuted a lot of awesome new things. Witcher 4 footage which I don't know if you saw looked amazing but this is I did but again one of the biggest empowerers of creators and creatives that's that's epic a lot of people use their tools they use their engines not just in gaming but in motion pictures so you look I I fully subscribe to this we've talked about the barbellification of all things which is the middle is going to be eroded and on one side you're going to have individual content creators that are making feature films that might look a little AI at first but eventually will rival the Hollywood AAA gaming massive multi-million dollar giant team GTA 12 stuff over on the other end.
And I I know people fear of the the interim and the job displacement. I completely understand that. And you know, we don't need to dive into that here, but but just for a second compartmentalize that and go, okay, if you were at, let's say, a company that had a team of 30 and that is no longer feasible. Now you're you're you're wondering what to do. Maybe you team up with three or four others and you build a massively multiplayer game leveraging AI because you can generate that many art assets once you define the style. You can voice that many characters once you decide what they should sound like and and what their personality traits could be. You can vibe code server architecture to just get you up and running a play test and maybe get some fans and get funding. So, I I I'm not saying you should for completely forgive or forget everything else that's over there that could be bad or potentially wrong, but open yourself up to what some of the most celebrated game designers and creatives uh of all time are saying, which is this is going to be revolutionary and it's going to empower individuals to make amazing products. 100%.
And the other thing to think about is if you're a specialist really in any industry, unless it's unless it's AI model training, I would say think of the generalized version of what you want to do going forward because you will be able to do many more jobs with AI's help. So, and what I'm saying that in the games business is like you may have specialized in a specific type of animation or you may have specified in a specific genre of something. If you and a small team can be more generalists and think about ways you can split up your work amongst yourself. I mean, Kevin and I have done this in the kind of secret project we're working on. There's all sorts of ways that you can open the door to this. So, to me, it's cool to see Tim Sweeney coming out and saying this. I think we're going to be going through this a little bit. I mean, obviously, Breath of the Wild is an amazing piece of a work of art, as is like many of the Nintendo games. I don't think he means per se that like it's going to be exactly there yet, but he can see a world where it gets to it. If 10 people make Breath of the Wild, that does not diminish Breath of the Wild. Totally right. And that's an amazing thing for those 10 people to make another game that people absolutely love and still play to this day. So, so let's not be upset that it's 10 people doing it potentially. Okay. You mentioned our super secret project. As much as we love to pontificate about the future of all things AI, as much as I love vibe coding and the way it's
Unlocked me being able to develop things and and and others are claiming that as well. You and I are hunting for a full stack developer. Someone with years of experience and expertise. You and I know the tools and what they can do and and how you can breathe apps and experience into existence, but there's still very much and I still think going to be very much a role for expertise, you know, for for for and by the way, if you happen to be a full stack developer out there, reach out. Reach out, everybody, reach out, or someone with audio engine experience or uh whatever, like we we hiring for real. Yeah. Figure it out.
Kevin, Flux Context came out last week. This is Black Forest Labs' update to their flux image model, which we love. We've loved it for a while. Um, they have come out with context, which allows you to basically swap in different versions but keeping everything the same. If you know uh OpenAI image sourcing, this is a lot of control network. It's a lot of that sort of stuff. So you can put an image of something up and hopefully get back the same thing or or match a style in a specific way. And I'll point out it's natural language editing as well. So, if you snap a picture of yourself and say, "Hey, give me a mohawk or put me in a cool leather jacket like Jensen," it can do it. Um, I know you played around with it. I I very briefly poked at the API and tried to um take a person and just modify their clothing, which is an example that everybody is like, "Oh my god, this is amazing." I was, I got to be honest, not too impressed with the results. It put the clothing on the person, but in the meantime, completely modified their face uh and the pose and everything else. And I was just like, that to me is one of those examples that should just work. There was a great uh tweet thread from Menchoy, king of the AI tweet threaders, but he talked about a use case where it could upres and and help old photos; there's a lot of good examples in there. There was another guy who went by uh @8bit_e who showed what flux context plus RunwayML video and comfy UI altogether could do. So you can see in a in a really interesting open-source workflow, if you're deep in that space, you could do a lot of interesting stuff.
But okay, my examples here, if you look in our in the drive there, I wanted to see what I could do quickly because in my use case, what I'm trying to figure out is like, first of all, is this better than something like 40 Image Gen off the shelf? So, I used a couple models. First, I used Flux Context Pro, which is their basic like swap out a genre, look for something else. So, on the Flux Context Pro page uh on on Replicate, you see Black Forest Labs. And this is a big thing where you're trying to see like how it transforms something into the other. In this instance, it says make a 90s cartoon. So, I wanted to use that exact same prompt to see what I could do. And then I compared it to 40 image gen. So, if you see in the drive, I took our thumbnail from last week and I just used the exact same prompt that they had here and I put it through Flux Context Pro and you know, it's not bad, right? You can see like it's a kind of an interesting output. Like it gets the text right. I don't think it's like amazing, but it it does look like a 90s cartoon. But then Kevin, go and look at the version I did with 40 image gen, which again we've had our problems with, but to me, much more '90s cartoon. It almost like the text almost looks better to me. Our faces are definitely better. So this goes to the point of like I think if you're really good at using open-source tools and company UI, this is going to be massive for you because it's it's a door you can open.
The other thing I used was Flux Context Max, which in their example shows how you can take a um a logo and put it in a location and they're it's a beautiful sort of 3D shadowed logo. I did the same thing. Now, granted, our logo maybe not fit as well, but I did the same thing with Flux Context Pro and 40 image gen. And again, I have to say like the better version of it was kind of 40 image gen. If you look at the two different versions, like it it put it I use the same prompt and it put it in location. Now, the Flux Context Pro came a little closer to holding on to stuff, but if you see the examples, they're not perfect, right? They're not it's not like it's a major thing that it felt didn't feel like a huge leg up to me. But Al for humans. Yeah. I is that about Al Borland or Weird Al? Like, which Al is that? Yeah. I mean, that totally got it wrong, right? Like Al for humans. It didn't it thought that was AI for humans. So, in my general experience, if you are an open-source maxi, if you love company UI, this is great. If you are more trying to just do something simple, I still think 40 Image Gen is probably better at a lot of this stuff than you might need. And that that was my my initial take on it.
Okay, Kevin. Another thing I tried quickly, hey Gen Avatar 4 came out, which um you know is their most updated talking avatar feature. Um, just to show you like you know this is they did give us some credit so I want to be clear about that. I got some credits for free to try this out. I think it's good. If you if you play the first clip, you'll get a sense of like I basically took a single screen grab from a from a mobile video I made to see what I would look like. So, you can play that for everybody here. This is my test of Hey Gen's Avatar 4. I was given some credits to check it out and I'm wondering, do I need to do this anymore? Well, you got Jordan Belfort mouth. You got Wolf of Wall Street mouth. I know. I really do. It's pretty crazy. Like big choppers. Please, if you're getting me audio only, go to this and watch it and tell me he's not about to sell you a pen. Like that is that was full wolf mouth. So I got some you know you know what's what's so interesting about this to me and I've said this before with AI avatars is like look it will work and and one of the other big pushes for them is like virtual avatars and like the difference that really the only main thing I problem with is the teeth look different than my teeth and there's a little funkiness to it as well. Full disclosure, I use Hey Gen almost daily on some of my AI products. I've known about these updates but couldn't say anything. So, I'm I'm glad they're finally here because I think the gesture support, which maybe we'll get to in their new video editor, which is really like a text file editor, is super interesting to me. But for this avatar version, Gavin, what did you have to do to train it on yourself? Nothing. Single picture. So, that's the thing. Like, so this is a single screen grab. So, that is very cool, right? But, you know, in some ways like it's not that far off in terms of what you can do with Hedra, all these other lip-sync tools.
Correct. Um, and I will say to your point about, you know, you've been using this on the enterprise side. I think from a value standpoint, if you are really doing a lot of work or really high-end work with one model that you have created, like there's real value probably in Hey Gen's like backend. The other thing, Kevin, I'll show you is I I took a I took a screen grab of a of our uh we got to come up with a name for our our Terminator character that we put in our thumbnail. So, what I tried to make uh this character speak as well because to me again, the biggest thing is like how do you do characters that aren't humanoid? How does it recognize faces? And it didn't do amazing with that either in some form. Haha. Kevin and Gavin, the world isn't ready for me. Avatar 4 Bender. Oh, interesting. Yeah, I I've seen some examples where it it works incredibly well of animating like, you know, weird characters or Pixar-like animals or whatever. That particular test was a little and it might just be like how it perceives it. But like again, all this is just getting better all the time and and you know, if you really want to dive in, there's a whatever an hour-long HeyGen uh keynote. There's a keynote presentation. I I will say so so the editor which is interesting is because you know they're trying to rethink the way video editors work. If you've ever done any video work you know that it's usually layers and it's like you know your your footage with your graphics on top and maybe some sound design below. They're getting rid of that almost entirely and going for you have your script and then if you want a graphic to appear yeah you can go place it where it needs to appear but you can highlight the words and say this is where the graphic comes in. This is where it goes away. Similarly, you could take a script of Gavin saying, "Welcome to AI for humans." I could highlight welcome and say, "I want a thumbs up." And I could I want the voice to be excited and on the humans I want a point. And so, you know, it's still very early for this sort of stuff, but you can imagine like it autogenerating a library of gestures, emotions, not just with the voice, but with the the facial expressions of the character. You sort of seeing them and going, "Yep, okay, that works. That's good." And now suddenly your avatar can be contextually aware of what it's saying because you know the the you should try this licorice is delivered with the exact same level of excitement and a smile as I'm sad grandpa went to the farm upstate. Like it's all the same performance. So we got to get more grand.
And one thing I'll say about that is we're working on a presentation for next week. By the way, we are going to be at the B uh film festival uh next week. So if you're there, take a look at us. Come find us. We'll we'll say hi to you if you're a listener. Yeah. all of our listeners that are going to the G7 summit right after that. Maybe come a little early, fly those PJs out. Exactly. Why not? One of the things that's so interesting to me about making content for that is just how hard audio can still sometimes be to get what you want out of it, right? Like you really do have to think about custom models and doing all that sort of work to make it work. Okay, Kevin, we have to keep going. Mark Andreessen has been starting to talk about robotics in a big way. And if you know Mark Andreessen is the head of A16Z, they are a a one of the biggest players in the VC business. But in this clip specifically, he said something that I think is important for all of our listeners and viewers to hear. You've all probably seen, you know, Elon has this Optimus robot that he's building. Um, these humanoid robots like the the the general-purpose robotics thing is going to happen and it's going to happen in the next decade and it's going to happen at giant scale. And I I I think there's a plausible argument which Elon also believes that robotics is going to be the biggest industry in the history of the planet. It's just be gigantic. That is a big deal that you think about it. Now, granted, like just to be clear, I'm sure they have investments in a lot of robotics work, but I do think we talk about robots on this show a ton. Humanoid robots are the things that if you've been watching our show or listening to our show for a couple of years and you were early on all the stuff we're talking about that's now come to the mainstream, this is the next one, right? And Unitree has just teased the fact that they may be releasing a sub-$10,000 robot, which is a big deal. We also know that we've seen the Unitree battle bots that we talked about last week, but having a sub-$10,000 humanoid robot is something that's pretty significant. Listen, we already know that if it can fold laundry, Gavin, you're going to allow it into your household. I would I would consider one in mine at sub 10k depending upon what capabilities it had. But, you know, again, this is this is such a race because as as cool as the simulations are that these things can learn end to end through AI by simulating, walking around, picking things up, the data that they're going to get from it being in the real world, moving about, interacting with people is going to be massive. And as long as it doesn't karate chop flail its arms like the robot we had a couple of weeks ago, what, you know, I could see these things popping up everywhere. What if it could play badminton with you, Kevin? Would you like that? You can charge me 20 bills for that, baby.
Are you talking about the robot dog with a little a little racket on its back, guys? Yeah. So, this is from ETH Zurich. They're a research lab. There's a very fun little video of it playing badminton with somebody. What's interesting about that to me is that you you know, you often see these robots that are we have to feature one a long time ago on the show where it's a tennis launching machine, right? And it will launch a ball back and forth, but an instance where now it's getting to be pretty good at tracking where you are. So you can imagine a world where this is for practice for sports stuff but not just for that like think of all the other use cases for something like this whether it's like you know a training model or it's something even if you you know wanted to have I'm trying to think of some some other use case besides training like I literally can't think of another use case badminton or bust that's all this thing about crazy how about pickleball no it could never be good the robots will never be good enough to play pickleball you can clip that because it's never going to come back to haunt me. Out of my kitchen AI, it was trained in a simulated environment, which is interesting, right? And when you watch the fluidity, the speed with with which this thing is tracking the object, adjusting its little robot dog position, whipping the thing, it has the clearly the knowledge and understanding of where the racket head is, and it's flacking it back to the player. That is impressive. And this, we say it every week, this is as bad as it's going to get. So, yeah, there are there are implications here. I don't know exactly what they are yet, but there are implications.
Speaking of implications, it's time to talk about the things you talked about and showed off on the internet this week. It's AI. See what you did there. Sometimes you're scrolling without a care. Then sudden shout. Okay. this implications. We need to talk about the implications of RunwayML V3 because this may be I know we've been through a lot of things. This may be my all-time favorite video. I discovered this video on the Bard subreddit, which is a hilarious thing to me, but it has since blown up. Let's play a little bit of this video. A revolutionary athletic program is challenging everything we thought we knew about cats. Here's Brook Landry. You might think a synchronized swimming team for cats sounds insane, but a brave team of trainers is trying to prove the skeptable local news uh coverage. It's like they did a great job. This guy, his name is notice_analytics, the original poster, and he essentially created a news story about cats synchronized swimming. And like again, these are the moments of joy you live for if you're following this. Yeah. you're following this world. There's a big there's some big surprises towards the end. I don't want to ruin it, but it's a very long fun video. You have to you have to go grab the show notes if you don't do the video of this. If you do, then you already know. The sequins outfits that the cats are wearing, the little swim goggles, and then when it cuts to the trainer, that's like it is a laugh-out-loud moment in a in in a way that not a lot of AI videos get me. This was masterful. I do want to hint there's a little bit of a scandal towards the end where the cats might be uh ingesting substances and it it covered a whole thing. Anyway, really big shout out to Notice Analytics for doing this. He was basically like kind of wanted to try something and this is the kind of creative fascinating stuff that can come out when somebody just is to like dump their brain into AI video.
Yeah. I I have seen you know this is anecdotal but on LinkedIn Gavin a lot of even folks that were even never AI people are coming around to the All right. So, I decided to see what this was all about. And look at this thing that I just made. And it's usually a little snippet of something or a spec commercial or whatever else. RunwayML V3 is inspiring a lot of people. Again, it's anecdotal, but in my little bubble, I'm seeing a lot of people dipping a toe in the V3 waters and getting inspired. Totally. And speaking of that, Kev, there's a really cool thing that somebody made, Vexalhub on their YouTube channel made Vexflick. And what this is is Sora and V3, but kind of integrated into a Netflix UI. And so that it really does look like there's all these kind of fake uh movies that are playing. And you know what's cool about it is you see one of the what is the main video is like Heavy Lies the Cream and it's like a story of like putting cream into a cup of coffee. There's a a fake reality show called Gan Pool. And it's cool because each of these is is seen in like a popup preview that you would see on Netflix. The Lex the Lex Friedman standup special. Yeah, exactly. All this stuff is stuff that you can do. But what's cool about seeing another creative cool person putting this into a format we know it elevates it to something else which I think is very cool. Um I really quickly want to shout out it's not a visual something but Jack Dorsey the uh original uh Twitter founder and if you will I think he's like the Rick I know Rick Rubin is now into vibe coding apparently but I feel like if there was ever a vibe coder it would be Jack. He said, quote, "I now spend two to three hours per day reading research papers and building something with Goose that I didn't think it capable of doing. I never see a line of code and never uh and never trapped in an IDE." So Goose is their like skunk works project, this agentic coding something. um what he's alluding to here, what Jack is saying is is the promise of this future that you and I have talked about so many times, which is like it's great that AI can code, but there's a huge barrier to entry anytime you have to get into an environment where coding actually happens in a meaningful way. Like not little web web apps, not like tiny little snippets, but like real actual uh structural things. And what Jack is saying is that he's spending hours a day writing code without writing code. He's just talking to Goose and it is building things and despite quote requiring some nudging every now and then, it works nearly every time. So again, hyping his own bag, so to speak. But there's a lot of people saying that is the future of video editing, of coding, of painting, of visual effects. It's not having to directly manipulate the tools. It's interfacing with the machine so that it can do the work for you.
That's right. And it's a great way to kind of end this. And the whole idea of that like for today's kind of thesis around the show is really like these are going to keep getting better as we've said all along. And this is like one of the biggest like you know on paper developers that has existed the last 10 to 15 years who is diving into these tools full-time. So that is it today everybody. We will see you all next time. Stick around, share us your stuff and uh hopefully we'll see you on the internet. I thought to stick around was you were teasing what's coming up next. Gavin, what what's coming up next? Our next meeting. I don't know who what what are we the what are we the leaders for?
You came for AI for humans, but stick around because I guess maybe it's one of the other AI video people that we love and know.
I was going to be like bowling, bowling with bowling for robots or something. It would be something like bad mitten with the boys. Yeah. Like, what is the dumb V3 thing? That's kind of bad with the boys is pretty good.
Or maybe it's a bunch of robots that drink in an Irish pub and they talk about their old lives as playing bad mitten players. Like, it's that sort of thing. Stick around. Bye. Bye everybody.