Midjourney Goes Medical, Noam Shazeer & Dean Ball Join OpenAI | Diet TBPN

Midjourney. Why Midjourney pivoted to medical machines. It's not a pivot, it's an expansion. It's a second act, a third act for David Holls. Fantastic entrepreneur, extremely inspiring entrepreneur. Maybe we should give some background on David Holls.

Wait, should we talk about the show today first?

Yeah, please.

Because this is the first time that Derek Thompson and Jake Paul have been on the same podcast. And I wanna just take a moment to just appreciate that because a lot of people said it would never happen. They said it was impossible, that they were just—

We're going all over the place today.

Two different worlds, but we're bringing them together. Derek Thompson's kicking off our guest lineup today at 11:30.

Yeah.

You don't need an introduction there. We have Rene Haas from Arm, CEO, also on the board of SoftBank. Very exciting. And then we'll be closing it out with David Senra and Jake Paul.

Beautiful.

At the end of the show.

In live in the TV pen.

I'll turn it on.

Anyway, yesterday Midjourney announced a new division of the company, an expansion, not a pivot, called Midjourney Medical. We'll watch the video, we'll go through the Midjourney scanner. Tyler was actually at the launch event, has some news to break it down. But first, let's kick it off with the history of David Holes because it's fascinating. Grew up in Florida. Son of a dentist. He's not the dog walker. He's on team dentist over there. Son of a dentist.

We have a son of a dentist on our team.

We have a son of a dentist on our team. And played video games growing up.

Dentist cam, dentist son cam.

Dentist son cam.

Look at those teeth. Oh, wow.

Wow. Nepo teeth.

Nepo teeth.

He's got nepo teeth.

Go back, go back, cut back. Look at that. Big smile, big smile.

There we go. He's got the nepo teeth. That's crazy. That's crazy. That's crazy.

That's crazy.

Wow.

Your father really saved his back for you.

Born on third base with teeth-wise. It's crazy over there. Played video games as a kid, obviously, but starts hacking them, learns a little bit about computers, super intelligent, PhD in math track. I don't think he graduated or finished the PhD, but I mean, at that level of engineering— He turned down. He turned down the PhD to go work at NASA, I think. So he's at NASA, eventually branches into entrepreneurship, starts Leap Motion. Pretty traditional venture-backed startup, raised money. I think Founders Fund was in at least one of the rounds, but it was a grind by all accounts. And so the first product brought hand tracking to VR. We should pull up a video of the original Leap Motion because there's some cool demos and you can see how it worked. And it's not actually that crazy to think about what we're gonna talk about today with the Midjourney medical scanner if you know the history of Leap Motion and some of the—

Yeah.

Some of the technology that was employed there. Yes, this is a different use case, different product, very much bigger. But in terms of sensing, detecting, mapping, using algorithms, yeah.

Yeah, and so when David was talking about the backstory, he was kind of going through some of the technicalities of how the Leap Motion worked. And it was incredible because he was like— I mean, this is early 20-teens, and he's using these deep neural nets. He said it was a mixture of experts model. It was incredibly far ahead of the curve just on deep learning. And it's just on this kind kind of like weird, like side project. It's like doing hand motion.

Yeah, everyone thought VR was gonna take off. There was so much energy around Oculus and the dev community, but it was a tough, it was a very tough business. A lot of hardware challenges, much less mature supply chain, everything's expensive. And there's all these complex market dynamics with other big companies. Meta bought Oculus, Apple was investing heavily. Eventually Apple actually tried to buy David's first company Leap Motion, but the deal fell through at the 11th hour. I think it actually happened twice. There was a reporting that like Apple had printed welcome packets to all of the Leap Motion employees being like, "You're getting onboarded today." And then the deal fell through, which is crazy. You don't see that normally.

David turned it down.

He turned it down. There are some crazy reporting about like getting in and basically being like, "Ah, this is too corporate." Or like, "This is not the vibe. We're not gonna be successful here." It was really rough for the company. They had to do layoffs, but they kept building and they sold devices to hackers. Yeah, look at this, the Leap Motion. This is, yeah, play Fruit Ninja. Now a lot of this stuff can be done with cameras and hand tracking. And the hand tracking on the modern VR device is pretty good just from the headset. Hand tracking in Apple Vision Pro, my favorite device, is fantastic. But at the time you needed a third-party device. So you had, can you scroll the video down to show the actual device a little bit?

Yeah.

Yeah, this little, like, it looks like a stick of gum. Very small, very neat, very cool. But ultimately, Very much a dev kit, never got to, you know, mass consumer adoption like an Oura Ring or a Whoop Band or anything like that. It was not must-have for, you know, the average gamer, the average consumer, the average technologist. After about a decade, David was able to take a step back and start hacking on the next thing, AI image generation. You know, Midjourney, obviously, huge viral success, broke a lot of the traditional rules. So no venture funding, no front end, no website, no app. Like really, they, I mean, now they have a front end and stuff, but years went by where it was just was just a Discord. Everything was in a Discord. And they scaled that Discord to millions of members, so huge that the company Discord, I think it was the biggest, the Midjourney Discord was the biggest Discord for a while. They had to scale their systems at that company. But you're doing all this like off-balance-sheet R&D basically. And on day one, you just get a fantastic mobile app for every device.

Social experience.

Social experience, that was the second thing.

So yeah. Yeah, because there was so much, this was the air, like prompt, engineering era, right? Where—

Exactly.

—people had their favorite prompts. Yep. It really helped kind of cultivate—

It was multiplayer on day one.

Yeah.

And so David has the quote, I think it's from him saying like, if you show someone a blank box and you tell them, hey, this is a magical AI machine that can generate you a picture of anything, people will just type in dog and they'll get a picture of a dog and they would get the same picture if they just went to Google Images. That's not what AI is actually interesting at. It's the astronaut riding a horse on the moon, the thing that that picture doesn't exist already. So it's you flying an F-16 in your hometown and like personalizing and doing something that would typically take a very large CGI budget or some sort of, you know, Photoshop master to actually whip up or collage together. You could do that with just one prompt. And so the multiplayer nature of Midjourney, when I jump into the Midjourney Discord, the first thing I see is your crazy prompt, your, 12 levels deep prompt injecting, thinking of different keywords, doing different SREFs for style references, doing all these different tweaks. And then I can just see, oh, he used this model and he specified these dimensions and he's getting really good results. So I'll take that, but I don't wanna make it about, I wanna make my images about a dog. You're a mosquito guy, so you're gonna make 'em about mosquitoes, right? And you can take my prompt and just sort of remix it.

And so that's really— Some of the mosquito propaganda prompts. I had back in those days were crazy. Yeah. Actually, I never ended up sharing them. They were simply too good.

Yeah, well, you lost the war and the mosquito's going away whether you like it or not. Google is focused entirely on killing mosquitoes apparently. By generating 4 candidate images for each prompt, Midjourney got a bunch of really valuable data on what a correct image looked like for a specific prompt. This was another sort of Midjourney innovation. I don't know if they were the first ones to do this, but it was really key to the Midjourney flow. It would gen you, instead of generating you an HD image, it would generate you 4 low-res images and you'd be like, ah, that has the right vibe, that has the right style, that has the right layout. And then it would get that feedback and be able to iterate at the same time. Stable Diffusion had some really good results, but it ran locally, it was open source. And so they weren't having the data flywheel like Midjourney was. Yeah. And so there were a lot of really positive things there. Now every company has a data flywheel and can, can iterate on that, but it was an important innovation. The company still that we know of hasn't taken venture funding. So I think David Holes owns most, if not all of the company. And after the big Meta deal to Vower Vibes, there's clearly enough cash flow to fund other projects. David's actually been talking about this for years, this idea of a medical device.

Yeah, the Meta deal, I don't think any specifics were ever released, but I think it's safe to assume it was in the hundreds of millions Hundreds of billions, trillions potentially. Potentially, potentially.

Who knows?

But enough to be able to take some really, really, really big swings and be able to have a balance sheet like a heavily funded private company.

Yeah, so it's really cool. I was hoping for a VR headset, honestly. I really wanted a VR headset.

I mean, this is just one of the hardware projects, right? He says like, I think there were like 8 little, you know, images.

Teasers.

Yeah, teaser things. So there's a bunch more hardware.

Because after the Apple Vision Pro launched, I think he put out some sort of review being like, ah, we're gonna have to do it ourselves, which I love. Anyway, let's play the Midjourney launch video for Midjourney Medical and see what's going on here in this 2-minute video. The goat is still on screen. The goat is in our stream, not in the Midjourney. Did the aesthetics of this video carry through to the actual event experience? I only saw a few photos, but Did the event have a nice vibe to it?

Yeah, it was very cool. They had similar, like, warm lighting everywhere. They had, like, kind of fake versions of the real machine. It's, like, way too big, obviously, to move around. But they had these kind of big tubes that, like, resembled it. Then they had a smaller version you could put your hand in. I did it in my hand.

Interesting. Are you healthy? Is your hand healthy? We need you coding at all times, prompting. I think it—

It's hard because that was like a small demo version, so the resolution was lower. But someone else said that they figured out that they had some wrist thing because of boxing.

Whoa, carpal tunnel.

Yeah, they found some issue.

You can just use an if statement for that.

In real time?

Yeah. If San Franciscan, then yes, carpal tunnel. Very cool vibes. I've always said this about Midjourney that I think when you, like, people debate, is AI art, art or whatever? I always think of Midjourney as a David Holes art experience. And I feel like you as the prompter are merely the museum goer in the museum of David. And he is the one who is creating, and the Midjourney team is the curator. They're the artists. And you are experiencing the art through the prompts, but the decisions and the taste comes from David. And so our Midjourney images tasteful. I think they have taste. I think they have David's taste. And then there's the opportunity for a user to put a little bit more on top. But by default, I think the reason Midjourney has been successful is because it's so opinionated and it's not trying to just be a stock photo generator.

Yeah. Even this video was fully filmed by David.

Really?

Like, he was like the one holding the camera.

That's cool. Yeah. Ah, love this guy. It's amazing.

It makes me a little bit emotional. It's so cool.

Yeah.

It makes me feel like we're finally living in the future. Emotions detected. No one making this kind of device has ever been as cool as David Holt.

It's true.

Straight up. Let's move on. I'm willing to state that as a fact. It's just true. Like, this is not the first medical device product that we've talked about on the show.

Yes, but the medical device industry does not have aura.

Yeah.

They have no motion.

He's bringing motion and aura—

To the medical device—

Field.

Category.

Yes.

He's inventing medical devices.

I think a lot of people are gonna wanna get these scans because it is a beautiful representation, futuristic representation of the human form. It's personalized.

Also just never black pill on startups. Like he could be so successful and he doesn't have any venture money and he's never gonna sell this business. And so you have like 12 biomedical device companies that are like, we have to, we cannot get steamrolled by, David Holes, we gotta work with this company, we gotta buy from that YC company. Like the competitive dynamics of the market are always more complicated than just a single winner comes and steamrolls all the startups and all the incumbents. That's not how things ever play out. Life is more complicated, the economy is more complicated, so never black belt. Let's play the deep technical dive, the technical deep dive. [MUSIC] Did he make this music too, I wonder?

Yeah, and he also made this in, I think, Blender, he said.

Whoa.

Like all the animations are all made by him.

By hand though?

He's the one person.

No AI?

I guess he's using Blender. For this, it's a great use case for Blender because you're cloning, you know, personalities.

I don't know if it was actually Blender. It was some 3D— I mean—

CGI stack.

Incredible music selection.

Yeah. Oh yeah, there's words we should be reading. There's no voiceover. But we can kind of give you the summary here. The, the overviews, we'll start with something called the transducer. We're going through, transducers act as speakers and microphones. Imagine a choir of 9,000 singers and an audience of 9,000 people all listening carefully. The transducers are controllable at a rate of 100 million times per second. Together they fire in a pattern that sends out structured waves. Oh. Those are the structured waves you're seeing. The waves travel out at a speed of 1,481 meters per second. We construct a ring of 40 of these systems, 70 centimeters in diameter. The ring consists of 358,000 ultrasonic sensors. The chips take turns sending out waves. As the waves have a chance to dissipate, we fire the next one. One by one, they fire at a rate of up to 1,000 times per second. You're vibing out. The waves travel across the tank in 480 microseconds, about 1/2,000th of a second. Hundreds of thousands of transducers listen. Each sensor resolves motions smaller than the width of an atom, not micrometers or nanometers, but picometers working together Hundreds of sensors can even push into the subatomic femtometer range, finer than the scale of atoms, a scale where we don't even have anything you've heard of. The system captures data at a leisurely 17 gigabytes a second. Gigabytes of sonic reverberations flow around your body, and from vibrations, They form images with this device. Each probe sees different angles of your body. We combined hundreds of waves and thousands of sub-images to get the final product. Over, over 40 gigabytes of data moves through the system to just see one slice of your body. We analyze the images, making out organs, structures, and tissues. Of course, all good uses of AI. There you go. Not going to read all that. Here we show a slice of what we can see with 25 different biological structures. Look at this. We do this again and again and again as you move through the ring. So are you moving through the ring or is the ring moving around you?

No, no, you're moving down. The ring stays.

Oh, interesting.

That's why you saw in the video.

Yeah.

The little platform you stand on.

Platform, interesting. Over a period of 60 seconds, the goal is to obtain several hundred slices of your body, reconstructing across 21 servers with 2 petaflops of compute power and up to 806 terabytes of raw data. The slices form a 3D map with the ability to resolve internal tissue details as small as half a millimeter. Wow.

Limina Leap says, still can't believe they missed the med journey opportunity here.

Less than a dozen of these systems operating at full speed can do a full body scan of every— their goal is a fleet of 50,000 of these scanners capable of a billion scans a month. It's basically everyone on Earth.

Tyler, what's the logic of putting these, integrating these into spas? They want to make standalone spas or they want to make a first one in San Francisco that's like a flagship and then—

Yep.

I can't imagine all 50,000 scanners would come with a full spa, but—

Yeah, I think he said the plan was you start with the big one. I think there's going to be like 10 or so in the first SF location. And then in some places there'll be like hundreds of these, some places there's like 1 or 2. But I think that the spa is like mostly because, at least he rationalized it because like, oh, you know, going to the clinic or whatever, it's like not a good experience.

Yeah.

This should be like very quick.

Yeah.

And it should feel much more like going to, you know, sauna or steam room or whatever than going to the doctor. Because you can imagine you don't really need a doctor present during it, right? It's autonomous.

They should create a bunch of needles in the ring that come in and just connect with you, deliver whatever peptides, hormones, PEDs that it deems necessary.

Potentially.

Well, he also talked about this system is pretty overpowered, he said. So you could theoretically, instead of just basically reading what's going on in your body, you can write stuff too, right? So you can do non-invasive surgeries, whatever, by beaming light in different ways.

Yeah, yeah, we have some folks who have done that where it's like it pulverizes the cancer, the tumor, with ultrasonic waves.

Yeah, think about EUV lithography or something. If you position the light in very certain ways, you can actually make a change too, not just read.

You're saying potentially we could pivot this into more chip fabrication capacity.

That would be the goal.

Ultimately. Yeah, yeah. Yeah, sorry, we can make way more money just making more chips.

Yeah, one of my—

We're gonna go to the data center.

Friends over on X, @antibology, great account, says, "The obvious next step once you have a full-body ultrasound scanner is to use the ultrasound to do useful things to the body. Delete tissues, make cells divide, reprogram cells, read plus write. If you're an engineer interested in solving the right side of sci-fi medical devices, we're building this and hiring." Well, let's go through some, More of the reactions.

Midjourney launching an ultrasound scanner is such a clear example of freedom enjoyed by bootstrap companies that VC-backed companies would never have. Only time will tell if it's the right bet, but such bold bets require ownership and a kind of devil-may-care attitude. That's somewhat true, but the counter to all that is like the Elon Musk projects and a variety of founders.

And SPACs.

And SPACs, but also, you know, like Sam Altman has done this too where it's like, okay, there's a new idea, get a new team, fund it, and it's on the stock market like there are a variety of entrepreneurs that are playing the VC game, marshaling capital, and then have the permission to work on crazy stuff, like a mass driver on the moon, et cetera. But in general, I agree with this point, that a lot of founders get locked in a VC loop of you gotta raise money every 18 months, you gotta hit the core KPI, you don't have the time or cash flow 'cause you're win, win, win, fight, fight, fight. And it's very, very interesting and such a narrative violation that Midjourney has become such a machine financially in, you know, people would say it's a commodity market, people would say it's a highly competitive market. You're going up against Google with Nanobanana, you're going up against OpenAI. And yet Midjourney has carved out a fantastic community, a fantastic user base, fantastic business model. And it's really, it's a testament to something. It's just awesome. I don't know what else to say. What did Noam Shazir see? What did he, what did he see? Because Noam Shazeer, the legendary AI researcher, formerly of DeepMind, has joined OpenAI. Very exciting news for Noam and OpenAI. He says, "I'm excited to share that I'll be joining OpenAI and look forward to working with the exceptional team there. It was a difficult decision to move on. I'm incredibly proud of the amazing team at Google and everything we've built together. It has been an honor and a pleasure to work with all of you." Everyone is very, very excited about this. And Yaxin shows— Wild move. —some interesting details here. Here. He says, "Noam Shazeer, 6th war." I don't know if that's true.

Just getting right to the important stuff.

But I like that we are apocryphally building up our heroes.

Yeah, number one on the Metas list.

Number one on the Metas list.

Do you remember the Metas list, which we released summer of last year during the height of the talent wars?

Yeah. Co-author of Transformer, T5, Switch Transformer papers, one of the pioneers of sparse MoE models. He's leaving his VP of Engineering Gemini co-lead role at Google DeepMind to join OpenAI. Lisan Al-Ghaib says, "This is likely the most significant AI talent move of the year. It makes you wonder what's going on at Google." Only matched by the next day, Dean Ball joins OpenAI. Tyler, you've been at Ball Nowhere for a long time. We've had him on the show multiple times. I know Ball. We love Dean and he's been in the complete opposite camp. He's not an AI researcher, he's policy, a lot of experience there. Had a really even keel, I think, on a lot of the analyses of should something be—

I think the main thing is he really cares about getting this right as a country.

Totally.

Right? And he's been critical of almost every company in the space, but only again, because he cares. What does Jim Cramer think about these moves?

Yeah.

He posted at 3:00 AM this morning, Noam Chazir, top AI thinker, goes to AI from Google. Big win for AI.

And drop the O. Drop the Open. Just AI.

It's cleaner.

It's cleaner. There's closed source models anyway. Just call it the AI company, I guess. In other news, Riley Walls did it. He did the unthinkable. He bought a street in San Francisco and auctioned it off and Notion bought it. And now the Notion Way is officially born. I'm so happy. I'm so happy that Notion won this. It could have been some crypto thing. It could have gone so odd. It's like high risk. Whenever you turn something loose like this on the internet, it can go a bunch of different ways. Notion Way is not something that I think would be annoyed. I personally wouldn't be annoyed if a street corner in my neighborhood was Notion Way. It is an ad, but it's a— It's a nice ad.

Well, you certainly wouldn't be annoyed. You'd be thrilled.

But it's just a term that feels like it could just be a street name, not something that just feels totally like an ad, like with a .com in there or something. So fantastic.

Yeah, I remember when— Perfect execution from Riley Walls. I remember talking with him about this idea for the first time. Absolutely loved it. We told him like—

They painted the street too.

We told him that we would backstop it.

Yeah.

Basically like if no one bid on it, —then we would have to upgrade.

Yeah, we were like, we're good for like $1,000 or something like that.

No, I think it was like—

Maybe $10K or whatever the cost was.

I remember it was like $20K, $20K something.

But we were like, well, we'll backstop it. We'll be like the buyer of last resort if you need that.

And the winning bid was $40K. Wasn't necessary at all.

Very, very interesting.

They get the blessing of the mayor.

Yeah, yeah.

Daniel Lurie.

That's very exciting, very exciting.

Aiden Gomez accidentally became important at work.

I love Aiden Gomez. Can we get him on the show?

Winning his life.

We haven't had enough Death Grips fans on the show. We gotta have Aiden Gomez on.

Quoting really funny things to post from.

Also on the Transformer paper, by the way, Aiden Gomez, they call him Aiden Goatmez for a reason. He's coded. No, I love him. He's fantastic. And he did a podcast episode with 20 BC wearing a Death Grips t-shirt, and that was incredible. Anyway, a bunch of other funny things happening at the G7. Apparently Donald Trump had to ask Sam Altman how to adjust his chair up and down, how to use the chair. And that's gotta hurt because you're Sam Altman, you've spent billions, tens of billions of dollars, years of your life, a decade of your life building a machine that can answer this exact question.

Almost any question.

You take a picture of the chair, it tells you how exactly to operate it.

So to do that right in his face.

Right in his face.

Just completely reject the premise.

Yeah, yeah. And you know, you know, on the tip of the tongue, he's like, "Actually, like this is a great opportunity to use ChatGPT. You can use images, upload a photo of that chair. It can go and agentically pull the actual diagrams, the manual from that chair." It's chair super intelligence. It's chair super intelligence.

Mr. President.

And Mr. President is just not interested. He's just like, "I'm going with a human on this one." And that's why diffusion takes so long. That's why we're in the slow takeoff anyway.

You know, I don't think anyone who does good things in this world or big things in this world doesn't have haters. I mean, it's the, it's the day one story. And I honestly, you know, since day one, uh, when I first went viral, the video I was talking about, like I— Instantly people in my school started hating. So like, I don't even, I don't even remember life without haters.

Yeah. Your, your mom, you know, took that video when you were born. I'm sure some of the nurses were like, yeah, this kid's—

They're like, yo, fuck this guy. Yeah, yeah. No, you have to have it. And you know, good news travels fast, bad news travels faster. And the haters actually will talk about you more and say bad things. But at the end of the day, people don't really remember what was said. They just remember your name and your face. And so you could do with that what you want. And, you know, it's really, they're adding to the algorithm at the end of the day. So it's really just math. And if you just have fans that are saying things, you know, let's say that's 10,000 people, but add 10,000 haters in there and now 20,000 people are talking about you and it just adds to clicks, views, talk, trending.

Mm-hmm.

So that's the way, I've always looked at it and the biggest and best people in the world all are also the most hated.

Ferrari Luce has a lot of haters.

Will you be getting one? Say it again.

The Ferrari Luce. Oh, new electric. Oh, bro. Ferrari.

Are you in the market?

You gotta be careful. Cause if you want any of those halo cars, you know, this is going to be permanent. I want to say, I know you might want an F80 at some point. You gotta be really careful. I just want you to be really careful with this.

This is a political answer. No comment.

Okay.

No, that shit's ass. No comment, no comment, no comment.

Oh well. That was a lot of fun. Well, thank you so much for coming on, David.

Yeah, congrats on the new fund. Congrats to you. Have fun down in El Segundo. Hopefully give our best to everyone. It was great. The South Bay, I'm sure you guys are gonna be spending a lot of time there post-SpaceX IPO. It's just gonna be more and more action.

Flash bang out, leave us 5 stars on Apple Podcasts and Spotify, sign up for our newsletter at tbpn.com, and thank you for tuning in today. Goodbye. Goodbye.