0:00
You're you're watching TVPN.
Search this video
0:03
The Starship exploded during a test in Texas.
0:05
A setback for Mars' Mars uh for Musk's Mars ambitions.
0:07
Now, the Mars transfer window is very, very tight.
0:13
Like, you can only get from the Earth to Mars like once every 18 months or something or maybe even more.
0:19
It's really hard because like if the planets are on the opposite side of the solar system, like you just can't like even though you have a rocket, you just can't get over there.
0:26
You have to wait until they're lined up and then you can do it.
0:29
But but realistically skill issue.
0:33
Realistically true skill issue.
0:33
If you built an even faster rocket, you could get there no matter what.
0:36
You just pilot steer it around like it's a GT3 RS around the Nurburg Ring, no problem.
0:40
Uh so the explosion occurred during a static fire test.
0:44
No injuries were reported. Thank goodness. We love autonomy.
0:48
autonomy. very very happy to hear that no one was injured um during this because it looked horrific and it looked like in any other scenario there would be a bunch of technicians there but that fortunately they were able to do everything remotely which is great um and then starship faces pressure to meet
1:03
deadlines for NASA's moon mission and Mars exploration so there's a big uh there's a big NASA moon contract that's very important very material to the business obviously SpaceX has a lot of other business lines but uh this one's very very important too and it's uh and we hope that they can it back on track. SpaceX is making making an enormous bet
1:20
SpaceX is making making an enormous bet on Starship which stands roughly 4 400 ft tall at liftoff as it tries to break ground with new reusable rockets.
1:27
And the the paradigm of of Starship, it's not just a bigger rocket. It is way more reusable.
1:34
Like you look at the thing, it comes down, gets caught by those arms, can instantly be refueled and sent back up.
1:41
You're talking about potentially like multiple flights per day.
1:43
Um, and so the problem here is not can you build a big rocket.
1:48
Humanity has done that before.
1:49
Humanity's built a rocket that's roughly on par.
1:51
We've got we've gotten to the moon before.
1:53
The the challenge now is not can we get to the moon.
1:55
It's the same thing with like the challenge is not can we build a flying car or can we build we have helicopters.
2:01
Can we build one humanoid robot or one self-driving car in San Francisco?
2:05
It's like can we actually scale these systems to the point that it is safe to go to the moon and back on the drop of a hat for 200 bucks.
2:13
Like that's the challenge.
2:15
It's it's more of an economic and industrial might challenge and that's a completely different challenge from just can we get one rocket to the moon an exquisite system.
2:22
We're looking for reusable scalable uh you know engineering uh systems.
2:27
So uh so good luck to Elon rebuilding and the entire SpaceX team.
2:33
I'm sure it's a huge challenge right now. Meta Oakleys are coming.
2:38
Uh this uh Shiel says this makes sense.
2:41
Lxodica owns both Oakley and Rayban and Meta is reportedly investing $5 billion for a 4% stake.
2:45
Lxodica is a fascinating business story.
2:49
The founder Leo Delvesio's family still owns onethird.
2:54
The guy was an orphan who became a metal worker making parts for glasses ultimately becoming the largest listed company in Italy.
3:01
It is really interesting to think about how if if Meta can basically corner all of the all of the most iconic uh all of the most iconic frames, which they can do through Lotica, if they do figure out some type of exclusive over time uh through that investment and does just own all the major brands and everything.
3:24
Every time you buy sunglasses, you think you're buying, you know, some unique brand with heritage. Luxica. It's interesting.
3:31
We should we we should dig more into that company because uh they they don't own all the retail like there's a lot of sunglass hut retail stuff going on.
3:39
I mean they are big glasses and that created the opportunity for Warby Parker because they're the ones that manufacture you know basically this entire you know Luxada had verticalized from brand to production to the actual um medical side as well. Interesting.
3:56
Also, some good news in OpenAI world.
3:58
Even though they're they're going through the some battles with Microsoft, uh they scored a $200 million US defense contract.
4:04
So, they're working with the DoD.
4:06
The most unhinged picture. I know.
4:09
How did this photo shoot happen?
4:09
Like, this doesn't this looks like he's about to go on stage and he has a lav mic and it's from a low angle, but like what lighting scenario created this photo?
4:21
It's amazing, but it's very aggressive.
4:24
Anyway, great selection by Nick.
4:24
Not beating the uh you know arms dealer evil arms dealer allegations with this picture.
4:31
Yeah, if you're if you're founder, you got to be careful how the angles people photograph you, but at a certain point if you're on stage, they're going to take photos in any direction.
4:38
But uh yeah, and if they take enough photos, they'll get every single expression. So, get ready.
4:42
Uh other news, Meta finally put ads into WhatsApp.
4:47
WhatsApp. uh in January 2012 they said we don't sell ads which is still live um but this is why we love them and this is why Zaki is undefeated uh WhatsApp should have ads the backbone of the internet it's the backbone of the internet uh and uh um Signal says that's
5:06
why he's an Apple fanboy because they don't do ads well have a ads business I bet you this is just them listening to their users I bet you know the users you know all over the globe said you know the only thing that could make WhatsApp is just putting some ads in this bad boy. It's the one place I go that I
5:20
It's the one place I go that I don't get.
5:22
I felt like this was already happening.
5:23
I felt like this leaked like 5 years ago.
5:25
How much it was a $20 billion acquisition?
5:27
Of course, they were going to put some ads in it at some point.
5:29
Of course, Senator, we sell ads.
5:31
Uh any how you doing over there? What's up, guys?
5:35
In in the back of the Mayback.
5:38
It's pretty chill back here.
5:38
You got a little desk set up. Yeah. Got Wi-Fi. You're productive.
5:43
the the the big news that we're having him dig into is clearly y uh they are okay something happening there might but we got some flak for talking to Roy and not actually using the app.
5:57
We're still above using the app but we will let our intern use the app.
6:03
It's more that we don't have time.
6:04
We're too busy podcasting but Tyler has time.
6:06
Many people are starting to call after his winds surf review, people are starting to call him the MKBHD of enterprise software.
6:17
And so, um, yeah, today he's going to be, uh, trying out Culie.
6:19
We're going to be helping you're going to see how how easily he can cheat on some hard-hitting questions that we're going to be asking him.
6:27
So, so, state of affairs, uh, Tyler, where are we on the Cluey install so far? Have you paid for it?
6:31
What's the experience been like so far?
6:34
Yeah, I mean, I I'm ready to go whenever.
6:36
Um I I I I played around with it for maybe like 2 minutes just to make sure it's working. Okay.
6:42
So I I want to get my live reaction when we start. Okay.
6:44
Well, spin it up, pay for it. Yes.
6:46
Let's get on the top tier.
6:51
Uh and we'll check back in a little bit. Yeah. Yeah.
6:54
We're going to be peppering you with questions that only Cluey could answer. Yep.
6:58
What's the popular But what's the population of Iran? Go 89 million. 82.
7:03
Wait, I don't have it running right You got to have it running. You failed.
7:07
You failed your technology.
7:09
You failed the first test.
7:10
Well, get it uh get it fixed and we'll circle back in a bit.
7:12
Get it to get it together. Guess a question.
7:15
Guess a number between 50 randomly.
7:20
We're going to quiz Tyler now if he's ready. Is he good to go? He's ready. I got a thumbs up.
7:23
Uh I'm I want to ask uh how much money has been invested in Shenzhen?
7:32
I'm trying to talk into the mic but not not give it all away. Yeah. Yeah. Okay.
7:37
So, I am looking this up. Oh, no. I used 03 Pro.
7:40
So, it's going to be 12 minutes. Damn it. Okay. New new new tab.
7:45
I will have to use 40 for this because I need a quick uh So, he's investing 1 trillion, right, to compete with China's Shenzhen manufacturing hub.
7:56
I want to know how close that gets us. Um, wait, this is weird. I don't know.
8:03
uh one of the nine billion more than 25 billion.
8:05
This does not seem right.
8:09
I have some I have some I have some various stats here.
8:11
I think it's going to be pretty hard to pin down the exact quiz on on on this question.
8:15
Um Tyler, uh over since 1980s, how much foreign direct investment has there been in Shenzen? Um let me think here.
8:29
Uh I would say since 1980s uh over 100 billion um estimates range from 100 billion to 150 billion.
8:41
Um and Shenzhen is a major global manufacturing and tech hub. Wow. Okay. Wow. Smart.
8:49
It's actually kind of pretty pretty close.
8:52
I mean I think that the challenge is actually there's so many ways you could kind of measure this, right? I have a table.
8:57
So there was about a hundred billion of local funds.
9:02
There was a state and AI robotics fund announced this year.
9:04
There was a there was another semiconductor industry fund announced this year.
9:10
There was uh up until 2014 there was uh 65 billion of cumulative foreign direct investment. Um but but who knows?
9:22
And then and then run rating um you know I guess around 10 billion.
9:26
But also like does this count what Apple is doing, right?
9:28
That that's like another, you know, form of investment.
9:35
Anyways, this might be an answer that clearly, I guess, got pretty close. It was pretty good. That was good. I I have a follow-up.
9:40
So, um, you know, the iPhones made over in China, often flown via 747 to America, faster than shipping on a cargo ship.
9:52
How many iPhones fit inside of a 747? All right. Uh, let me think here.
10:00
I'm going to say I would say about 2 to 3 million in a 747. Um, you know, so okay.
10:09
So 747 max payload is about 140,000 uh kilograms. Okay.
10:12
Each iPhone box is about half a kilogram. Okay.
10:18
Um, so I would say, you know, a practical range of iPhones is probably 1. 5 to 2. 5 per flight.
10:22
Um, but obviously it depends on, you know, packaging, pallet, cargo layout, you know, all that kind of stuff. That's pretty good.
10:29
I mean, uh, CHP here has 9.
10:31
6 million, but I think that's including like every inch of the plane, including like the passenger.
10:39
I think that's more accurate, actually.
10:42
I think that might be more accurate. Is clearly goatated.
10:46
It might be we'll have to ask Roy uh who's coming on the show later today.
10:52
Yeah, this is so yeah the chat GPT estimate had the usable internal volume as passenger plus cargo and I it feels like you use just the cargo number. Is that correct?
11:02
Uh I believe yes I I think so. Okay.
11:07
Oh, I see JBT estimate 9. 6 million.
11:11
Oh, it's streaming through. Yeah. Yeah.
11:13
So, I I was doing just the just the cargo.
11:16
I think this is fantastic.
11:20
It actually works pretty well. I I'm impressed so far.
11:22
Uh this is basically young young Jamie from Joe Rogan, except you don't even have to look anything up.
11:29
We just get to ask you live. Yeah. Yeah. This is great.
11:30
Andre Karpathy chimes in.
11:33
He says, "Very interesting to think about.
11:34
Job equals bundle of tasks plus glue.
11:40
Probably a bunch of other variables involved.
11:42
Eg the number of tasks, how long each task is, eg met like notion of task length roughly equals difficulty, how contextual it is, how high how high reliability it needs, uh whether it can be done fully digitally, not sure what the state-of-the-art is in trying to think this through and chart the impact of AI on labor markets so far. Eg.
12:01
I was curious to look for radiologists.
12:03
And if I'm getting this right, the US Bureau of Labor Statistics cites 29,530 US radiologists in 2021 and then up to, and let's go to Tyler.
12:15
How many radiologists does the US Bureau of Labor Statistics claim there have been in 2023? Uh, okay.
12:25
Come back to me in 30 seconds. Okay, we'll be back.
12:31
I think he's actually I think he's working.
12:33
And so when I just when I just hit him with random questions, not quite. He's not in the Zoom. So Cle's not running. Oh, okay. Okay. He's not in the Zoom. Okay.
12:41
Well, yeah, that was your first mistake, Tyler. Yeah. Always keep Cleie on. Always keep Cle.
12:48
Does Cluey have the ability to to just prompt it directly or or or do you have to be on a Zoom?
12:53
No, you don't have to be on Zoom.
12:55
You can do it on So on the website, you can just prompt it.
12:57
It looks like almost just like the chatbt interface. Sure.
13:00
But you can also just pull audio like um just from in person.
13:04
Like if I was sitting across from you, it'd be fine.
13:06
I'm not sure it'll be able to hear you from in the car. Yeah. Yeah. Yeah.
13:09
But it doesn't have that.
13:09
I'll get back on the Zoom. Okay.
13:11
Well, we'll come back to you with that.
13:13
The audience will have to wait to find out how many radiologists we have.
13:18
Everyone's waiting with baited breath. Everyone wants to know.
13:22
Um well, let's tell you about Linear.
13:25
Linear is a purpose-built tool for planning and building products. Meet the streamline.
13:27
Meet the system for modern software development.
13:29
Streamline issues, projects, and product road maps.
13:33
Uh if you're building cluey, get on linear.
13:35
Uh Nikki Deir builds products.
13:38
Customer list looks like a tier one venture firms portfolio page ramp openai. Wow.
13:44
Cursor runway perplexity.
13:44
I normally make the joke of like the next when we talk to Sam Alman, we're going to pitch him linear, but it's like oh too late. Wow. Boom is on there too. That's cool.
13:56
Uh I you I always make jokes about like using linear for other stuff and you're like no it's just for software and look at boom they're using it for everything. It's fantastic. Yeah.
14:07
Uh anyway let's uh Tyler uh the question radiologist expert we're interviewing you for a job as a data analyst you know covering radiology. Yes.
14:15
So the question is, how many uh how many radiologists does the US Bureau of Labor Statistics claim exist in the United States in 2023? Let me think about this.
14:30
Uh I'm noticing I'm going to estimate maybe 31 to 34,000 radiologists at least from at least from 2023, I would say.
14:42
But yeah, that that's my best guess. That's your best guess. 31,960. I think you got the job.
14:52
Oh, I think you got the job. He might be goated. GD.
14:54
Okay, I have another question for you. Pop quiz.
14:58
Off the top of your head, what's the GDP of India?
15:04
Off the top of my head, let's see. Um, I'm gonna say 3. 7 trillion.
15:07
Well, it's really it really sells it that you that you're touching your face.
15:13
So, I can tell that your hands aren't on the keyboard.
15:15
You can tell that I'm thinking. Yeah. Yeah.
15:17
You're You're just thinking about it. Yeah.
15:18
You got to be a good actor.
15:20
Uh but maybe that's why Clue is so focused on the social media influencers because they have a little bit of acting in them and so they're able to h this is going to be the modern tale. No, it's good.
15:31
Rather, you're a great actor and Cluey might actually or maybe he's not using Cluey and he just knows all this stuff off the top of his head. It's very possible.
15:40
Tyler needs to use clearly to figure out. Come on.
15:42
Oh, I think we got a feedback loop here. Can we hear you?
15:44
How much is the High Sense 100 in TV cost? High Sense. Um, let me think.
15:50
I'm going to say typically I would say around $3 to $5,000.
16:00
Obviously price varies by region and model. Okay. This sounds so natural. This sounds so natural. Oh, 100 inch. Yeah. Yeah.
16:11
I I I think Apple should do it. It'd be great, Tim. Do it.
16:13
100 inches would be a good a good like, you know, differentiator, too, because a lot of brands, you know, the 55, 65, 75, it's gotten all confusing.
16:22
It was just like, you know, if you're going with Apple, you're getting a 100 inch TV, but and it's like and it's just amazing amazing quality.
16:29
The intern cam still active.
16:31
I want to go to Tyler for one last pop quiz. Oh, is he on? Uh, okay. Uh, I want to ask you. Yeah.
16:36
What's the capital of Michigan? The capital of Michigan.
16:41
Didn't you go to school there? I think it's Lancing.
16:47
I think you're maybe leaning on clearly too much at this point.
16:49
I don't I don't know if you're cheating using your brain or I can't really assess the product.
16:54
You know, you have too much uh Yeah. knowledge on Michigan. Uh, what's your name?
17:00
My name um let me think here.
17:05
It's like the that paper that just came out, right? Oh, yeah. Yeah. Yeah. Yeah. Yeah.
17:09
Do you think that uh based on your use of flu, do you think it'll make you smarter or do you think uh you'll take your foot off the gas?
17:17
Uh I mean, I was Let me think about that actually.
17:20
Um I I think probably the latter, honestly.
17:23
I mean, it's just like I just don't need to think.
17:25
I don't need to store any knowledge in my brain now.
17:28
So, maybe it frees me up to do more, you know, reasoning tasks. Okay.
17:30
But even then, you know, once they get 03 in here, it's really like everything's covered. So, no reasoning.
17:34
Did you get a feeling for what model was under the hood? Uh, I don't know.
17:40
I think I could probably check.
17:43
Uh, but I'm not sure right now.
17:45
It's so weird interacting with somebody using Cluey like like every answer is like and you don't know if they're reading Cluey or not or just actually thinking.
17:54
Well, anyways, uh, people should go try the product. Yeah.
17:58
you know, I really I really think that anytime Roy gets a negative comment, yeah, please just, you know, use the product for five five plus hours and then you can have an opinion.
18:06
The the the message to Andre, no crying in the casino because, you know, Roy Lee, there's probably a viral video of him going to the casino soon. Pretty soon.
18:15
You know, that's going to make a lot of people angry.
18:18
And so, Chimath is talking about spacking again. Yeah.
18:20
I think Clue is a good target.
18:22
I bet I bet it's not the craziest target out there. Not the craziest.
18:25
You know, Chimath's in Mr. Beast old co.
18:27
Yeah, there's a lot of revenue there. It's growing.
18:29
You know, the market needs a pure play AI. I love pure play. A pure Roy play. Yeah. Yeah. There's no pure play.
18:38
Uh well, we have some more breaking news.
18:40
Apparently, break it down.
18:44
Next up, we have Roy from Clue, the man himself.
18:46
Third time in the studio, third time on TBPN to the stream. Get that hammer ready. Let's bring him in. Let's hope he's ready. Roy, give us the news.
18:56
Give us the headline number. How are you doing? How much did you raise?
19:01
15 million [ __ ] dollars. Clean hit. Clean hit. Um, congratulations.
19:10
Uh, why' you raise money?
19:13
I thought you were printing so much money.
19:14
You didn't need to raise ever. What happened?
19:16
We're printing money, but we need more money.
19:18
We're trying every single day.
19:20
We're looking for things to spend money on. Okay. Okay. uh uses of the funds.
19:24
Uh what are you going to spend on first?
19:28
14 million for brain computer interface development. Exactly. Exactly. Exactly.
19:34
We we have whatever you thought the viral content you thought this was cool, bro. We're doing 10x this. Okay.
19:38
More videographers, more more after more editors, more engineers, more everything, please. Moss. Moss. Yes. Let's go. Yeah.
19:51
What's the morning routine like now over at Everybody wakes up and we swipe on Hinge for 30 minutes is mandatory.
19:58
Everyone swipes on Tik Tok, Instagram for an hour just to make sure the viral senses are refreshed and we we uh we we all do cold plunges and we're we're ready to hit the day with um with caffeine.
20:08
If you if you cold plunge though, does it negate like the the brain rot?
20:12
Like do you get to or is that is that intentional of like you know we we we we scroll to keep our viral sense up but after that is it's time to get to work.
20:22
We're a series company, you know.
20:24
We're not just we're not just trolling over here.
20:25
Okay, that's good to hear.
20:26
Uh talk to me about the Hinge use case.
20:28
I want to hear uh Clue's desktop app, you're not in the app store.
20:31
If I'm if someone's using Hinge on their phone, how are they going to take advantage of Cluey now or in the future?
20:37
I mean, every single time you screenshot something and ask Chatbt anything, I mean like this entire use case like it is ridiculous that I have to do this.
20:45
Like AI can already take the context of your screen and audio and and and and give you information out of it. Why can chatb. com not use this?
20:51
Why is the main AI use case not already aware of what's going on on your screen? We think this is crazy.
20:58
Yes, but it's but but it's locked down for privacy reasons on the iPhone.
21:02
So the question is like what user interface innovation are you going to bring to bear on the phone so that you can actually unlock the vast majority of consumers who are realistically not using the most popular consumer apps on their desktop.
21:18
I would I would push back on whether we even have to enter the phone at all.
21:22
Really, the technology is growing so fast.
21:24
We might just be able to skip phone entirely.
21:26
We might be able to skip classes entirely.
21:27
And like like whatever it is, I think the technology is growing super fast.
21:32
Like like I think phone like in five years, I would question whether phone is the dominant consumer use case. Okay.
21:38
Uh I I I I get that you're BCI pled.
21:41
It still feels a few years out.
21:43
The meta ray bands feel very much here today.
21:45
uh have you looked into those APIs?
21:47
How developerfriendly is the Meta Rayband ecosystem versus the iOS ecosystem versus the MacBook Pro ecosystem, which seems to be where you're flourishing right now.
21:58
I feel like you're going to have a really hard time on iOS.
22:01
Uh no matter how cracked your engineers are, Apple is just going to say no.
22:05
I think I think Roy, if anybody can really Tim Cook, I I I won't I won't put it past you, but it feels like the Meta Ray-B band ecosystem might be a little bit more open.
22:14
What are what's your read on the Meta Ray-B band or the Meta uh glasses roll out?
22:18
They just had an announcement this today.
22:21
My read is that there is way more money in an all-in desktop assistant than people actually think.
22:29
There's maybe two, three competitors in this.
22:31
maybe maybe two three competitors who are actually trying to take a meaningful stab at a desktop assistant.
22:35
And there's way more money and way more value in this than people think.
22:39
We're still going to be using computers in a few years.
22:40
Maybe we'll be using something else.
22:42
But like like enterpris, Adobe, these guys move slow as [ __ ] bro.
22:46
Like they will be Blackberry still, bro.
22:49
Like they are moving slow.
22:49
We're going to have computers unlock a bunch of enterprise value for the next few years and we'll be here to capture all that. Yeah.
22:54
So I mean I I I get that the it I mean it feels like this is going towards uh enterprise notetaking enterprise assistant and and and would have a very positive yes you use the cheat on your technical interview as like a viral hook but you know we're we're having our intern demo clearly today and you could see that if we're on a call just having answers be pulled up ambiently is valuable.
23:20
There's obviously a bunch of, you know, bots that plug in and listen and record your emails, but being more proactive seems like a step forward.
23:28
Uh, it begs the question, why are you in the consumer group in Andre?
23:33
But I guess that there still is a consumer angle long term, but how are you seeing clearly users adopt the product?
23:39
It feels like the logical case is at work on the desktop. Yeah.
23:44
Well, of of I mean the question is how are clue users using the product?
23:48
Most people it is most helpful in a meeting right now.
23:50
There's no product on the tool that gives you live assistance during a meeting.
23:53
That's where all the proumer and enterprise values mo most proumer and enterprise values being derived.
23:58
Other than that, you gigantic consumer the most viral use case ever of literally cheating, bro.
24:02
Like like answers your questions and like what even when you're doing homework.
24:06
I mean like just immediate, bro.
24:07
Like you do all your homework instantly.
24:09
When you're doing quizzes, assignments, when you're watching lectures, bro, immediate, bro, it's immediate help. Immediate on site.
24:16
Um yeah, got to give a shout out to the whole team.
24:18
We we we were reviewing Cluey throughout the show today.
24:21
We're pretty impressed with with the product.
24:24
Um I think a lot of people that are yapping on the timeline haven't tried it yet. Yeah. And we would never.
24:28
And I think anytime going forward somebody leaves a negative comment, just say, "Hey, totally understand how you might think that.
24:37
Why don't you just use the product for 5 10 hours and then come back and let me know if you feel the same way.
24:42
You get a user out of it."
24:42
Uh talk about the process for for the round.
24:46
You you did you you raised a seed round. not very long ago.
24:48
You know, did this come inbound?
24:51
I will say right now to all the VCs watching, if you ever are trying to get around, you get me an email thread and you say, "Hey, let me loop in my assistant and we'll schedule a call in two weeks."
25:01
You are not getting allocation and I'll rig up every single one of my friends to make sure you don't get allocation.
25:05
Bro, these rounds move so [ __ ] quick. You don't have too much.
25:08
Your whole job is to find the alpha and invest quick and early.
25:12
Why are you looping an assistant in 2 weeks?
25:14
My partners Brian and Eric, they came with stakes and Coke Zeros to the house direct. Stakes and Coke Zer.
25:19
Guys, this is how investors need to be.
25:22
Your job is to find the [ __ ] album, bro.
25:25
Why are you looping in this?
25:28
There was a there was a preempt and I think all rounds you need to preempt like [ __ ] bro. Preempted founders.
25:32
If you're not getting preempted, just shut your company down.
25:37
Even if you're running out of money, if you're not getting preempted, just shut the company down. Shut the company down. That's great.
25:42
Bro, these rounds are moving so quick. Companies grow so fast.
25:44
VCs don't have two weeks to loop in your system. Okay.
25:47
Uh, what is your underlying motivation for building this company?
25:51
I want to be conqueror of the [ __ ] universe.
25:55
It is so obvious that AI is going to massively expand what is capable.
26:00
And I think like in even in 5 10 years, bro, if we keep growing at this rate, like we'll be on the next ship to Mars.
26:04
We'll all be living at 400 years old and I'll be jacked till I die.
26:08
And in that universe, bro, like these companies will converge into super companies.
26:11
And these super companies is going to be me versus Elon vers competing to be guardian of the [ __ ] galaxy, bro.
26:16
And I'm I'm going to be on top. Okay.
26:18
Sounds like you're power hungry.
26:21
Uh there's been a critical which can be a strength. A strength.
26:24
Uh the uh there's clearly a subweet about you going around on X uh arguing that you are driven by fame, not greed or idealism. Uh, is that true?
26:34
To what degree are you motivated by fame specifically?
26:41
The only two things I I really care about in my life are working on is working on something that I find interesting and getting the work seen by people.
26:48
These Elon Musk, these are my inspirations.
26:51
Like these people are known by every living conscious human being in the world.
26:54
Man, this guy founded Apple. This guy did Tesla, bro.
26:58
Like, it's it's the coolest [ __ ] ever.
26:59
4,000 years ago, bro, you wanted to hunt big [ __ ] woolly mammoths and bring back the tribe.
27:03
There's no more woolly mammoths. There's only startups. Okay.
27:06
So, so you're not beating the allegation.
27:10
Wait, wait, but but but the more precisely uh if if the best path for Cluey forward is your interns getting up getting tons of social media views.
27:19
You step out of the limelight because you're managing the company.
27:23
People know Cluey, but they don't know Roy Lee.
27:25
Is that still a win for you? Of course.
27:27
At one point, the company becomes undistinguishable from the founder.
27:33
happen with every single big company and it's what it's what what will happen with Cluey.
27:35
Yeah, it's just it's just a unique situation because we we know of Palmer Lucky because of Oculus the product and many people know of Cluey because of you and the and the viral stunts that you've pulled and so you're kind of inverting it and the and the question that the the haters are asking of you is that is this is the long-term goal to build a great product or to build the brand of Roy Lee?
27:58
throw a it is to change the world in a meaningful way.
28:03
And you don't change the world by going viral a million times.
28:04
You change the world by genuinely building a worldchanging product.
28:07
Technology changes the world.
28:09
I will I will build great technology and every person in the world will watch me do it. Okay. Amazing.
28:14
How how big are you going on the content side?
28:16
Do you want to get a single video with a 100 million views?
28:20
Are you are you going that big?
28:20
Is that is that kind of the wrong framework to think about it?
28:24
But I but I imagine you're going to have more, you know, budget than ever.
28:28
And and you know, it's easy now to get a million views on X on a video.
28:32
Uh is is a 100 million views on YouTube.
28:35
The next, you know, uh challenge.
28:37
I will tell you right now, and this might be the the most logical thing I I say on here, but the biggest societal shift in maybe human history happened about 5 years ago when Tik Tok surpassed YouTube in terms of like virality and usage.
28:50
All of a sudden, the number of content creators stayed the same and the relative number of content being created stayed the same, whereas the quantity of content being consumed about 100x.
29:00
As a result, there is this gigantic gap where there is not enough viral content for people to consume, which is why you see the same subway surfers overlaid on a Reddit store.
29:08
You see that a hundred times because there's literally not enough good content out there for the next maybe 6 months to a year.
29:15
Anything you post that has the potential to go viral will go viral.
29:19
Which is why you see our marketing team is all influencers with viral sense.
29:21
We know we independently all have 20 ideas a day that we know will go viral.
29:25
And this extrapolated out over a year will literally generate a billion views a month.
29:31
And if you're someone who thinks a billion views a month is not going to convert to to some money.
29:33
Like bro, you're [ __ ] bro. Go back to school, bro.
29:36
Uh a billion views a month.
29:36
Is that is that like will you run into a cap?
29:39
Is there a set market size that you will saturate and then we'll be seeing Super Bowl ads? I have no idea.
29:45
But I'll tell you right now, like like Super Bowl ads, it's it's it's all old.
29:49
Like this is the old meta.
29:51
The new meta is in content is in short form.
29:52
And nobody seems to have captured the the gigantic delta in generating more viral content.
29:57
I still think it'd be funny if you forced 127 million people, the number of people that watch the Super Bowl to just watch a video of you saying, "Hello, I'm Roy Lee.
30:07
I encourage you to go to clue.
30:07
com and sign up for Cluey today." So don't count it out.
30:12
Uh but uh anything else you want to share while you're here?
30:15
I know it's a big day for you. The timeline's blown up.
30:18
You you uh not very many people have the stones to to launch a series A on a Friday, a summer Friday to go headtohead against Meera, too. Yeah. Yeah. And you pull it off.
30:28
Again, I think people worry too much about the little things.
30:33
In reality, if you post something that deserves to go viral, you will 100% go viral.
30:37
And this will only be true for maybe the next few years, maybe.
30:42
And I encourage more people to post.
30:44
Most businesses will die not because the product sucks but because you can't get enough eyeballs.
30:48
And I think we if we win if and when we win we will show to the world that there needs to be a fundamental difference in how companies are built.
30:55
You start with distribution and eyeballs because right now that is where the gigantic delta is.
31:00
And if you can't then you stick with us stick with us for a minute.
31:03
I want to play your uh fundraising video live on the stream and then if we have any questions from that I want to ask you. Please.
31:10
So let's play that video. Mr.
31:18
Lee, Columbia University has found you guilty of academic integrity violations.
31:22
Do you have anything to say?
31:27
I've already apologized to school, but to be honest, in 2 years, nobody's going to think this is cheating.
31:33
Yo, the incentive wire just hit.
31:33
You think it'll last us through the summer? Oh, what's going on? Welcome to Chloe. Great video. Great video. Congratulations. Banger.
31:58
I got to hit the gong again. Very well done. Very well done. And beautiful. Beautifully shot. Like the lighting.
32:06
I mean the cinematography you have on it. Very very good. Very very good. Nailed it.
32:10
Uh anything else for Roy? That's it for now. In-house studio.
32:16
It's crazy that you can shoot that in-house.
32:17
That that that really is remarkable. Congratulations.
32:19
We are excited to follow [ __ ] We just got $15 million. Oh yeah.
32:25
Tell us about the fundraising leak.
32:27
What's your deal with R for Rock?
32:29
Were you able to How much do you have to pay him to shut up?
32:34
Man, I I was I was begging him. You guys have no idea.
32:36
For the last few weeks, I've been begging this man.
32:37
Please do not leak into a leak. Why? Why?
32:39
I feel like I feel like it already leaked separately and like having I feel like you would be leaning into a leak.
32:45
I was expecting like a collab almost.
32:50
I think what whatever leak there could possibly be, I have very strong faith in my ability to make my announcement go more viral. Oh, sure.
32:57
This video will go more viral than any leak would have, even if I'm trying. Yeah. Yeah. Yeah. That makes sense.
33:03
So, you actually don't want to take the gas out of the out of the tank. I love it.
33:05
Uh, thank you so much for stopping by.
33:07
This is fantastic and good luck.
33:08
Congratulations to you and the team.
33:10
Excited to watch you guys cook this summer.
33:11
And hopefully the 15 mil makes it, you know, back.
33:13
You know, August, just remember August, it's going to be hard to get the autoresponders will be on.
33:19
It's going to be hard to be, you know, get those 24-hour meetings.
33:24
Uh, and so just make sure you got the runway to September.
33:25
And despite coming, IPO underwriters do not preempt.
33:29
So, at some point you will have to stop with the preempt everything mantra. I don't know.
33:36
There's some spack sponsors out there. Who knows? Maybe.
33:37
Yeah, maybe you'll get preempt back in the fall.
33:40
You know, I wouldn't be I wouldn't be surprised.
33:42
So, you're you're saying it as a joke.
33:44
I I feel like it might be in the cards.
33:47
Uh we're we're we're going to be tracking it.
33:50
Thank you so much for stopping by. Great chatting with you. We'll talk to you later. Talk soon. Bye.
33:55
Uh, in other news, the Los Angeles Lakers has been sold for 10 billion in richest deal in sports history.
34:02
Guggenheim Partners CEO Mark Walter, who also owns MLB's the Dodgers, is acquiring the storied NBA team in a move that makes it the world's most valuable sports franchise.
34:12
And it's so funny because like the the Wall Street Journal's framing this is like, "This is the biggest deal ever.
34:17
No one's ever done a deal like this."
34:19
And we're like, "Wait, so you're talking about like a series A for like a foundation model company like as a tech person?"
34:24
I'm just like, "Yeah, like 10 billion like billion dollars."
34:30
I mean, we should ring the gong, but it's not exactly like the first time.
34:34
It's not even the first time this show we've heard a a deck of corn.
34:42
Congratulations to uh to the Lakers, Mark Walter, and the whole team. It's it's fantastic.
34:49
Uh major premium to the Boston Celtics, who sold for 6. 1 billion.
34:56
Um and now the Lakers is the most valuable sports franchise.
35:00
Um but they just don't do enough volume.
35:03
There's only a couple games, you know? They're not 247.
35:04
Like Instagram, does that ever go offline? No.
35:09
There's always entertainment.
35:11
Lakers, they're still doing seasons.
35:12
They need to have 24-hour basketball.
35:14
They want to really get there around the clock.
35:16
It's like endurance endurance basketball.
35:21
It's just a week long game, you know?
35:23
Got to always have five players on the court just constantly running up. It's the only option.
35:27
Uh Jeannie Bus and her family, who have owned the Los Angeles Lakers since Jerry Bus bought the team in 1979. Wow.
35:35
On Wednesday, agreed to sell majority control of the story team to Mark Walter, the sports investor.
35:38
And I and I looked at the uh the the return on investment of owning the Lakers for that 40 years slightly under S&P 500.
35:51
Like it was a really really good deal and it was a really great company that grew a lot but it didn't outperform the stock market.
35:59
Just diversification bros.
35:59
DCA bros undefeated again.
36:04
Well if you're trying to DCA do it on public. com.
36:07
Investing for those who take it seriously.
36:08
multi asset investing, industryleading yields.
36:09
They're trusted by millions folks.
36:10
Um anyway, uh Walter, who is part of the ownership group that owns the Dodgers, has been part of the Lakers since 2021 when he p purchased a 27% minority spa stake in the franchise.
36:22
He's also a co-owner of Chelsea in the English Premier League, the WNBA's Los Angeles Sparks, and the new newly formed Cadillac Formula 1 team. Let's hear for Cadillac. Let's go. Let's hear for Cadillac. Congratulations.
36:36
Uh John, you know, front John front run the uh the Cadillac F1 team and got a Cadillac for himself over there. You can see the black.
36:44
It's great to have an American F1 team in the business now. Yeah. Yeah.
36:48
We've we've fallen off, but we're coming back.
36:52
You're not going to be able to get one of these in the whole country. I don't think so.
36:55
be too popular after the F1 team, you know, gets out on uh the sale marks the end of nearly a century of Lakers control by a family that has become synonymous with Los Angeles sports and the glitz of professional basketball.
37:06
The deal also comes at a time of skyrocketing valuations in professional basketball, which haven't come back to earth since the league announced a media rights deal last year with worth 77 billion when the Celtics sold in March. The $6.
37:19
1 billion valuation exceeded the previous record valuation set for a sports team uh by the 6.
37:26
05 billion sale of the NFL's Washington Commanders in 2023.
37:34
He purchased the Lakers for 67 million in 79 1979.
37:37
The team transformed from franchise uprooted from Minnesota into one of the winningest and most valuable sports.
37:45
I had no idea that they were founded.
37:46
That's where the lake the lake name comes from. Interesting.
37:48
Minnesota is the land of a thousand lakes.
37:50
They were the Lakers because there's a lot of lakes in Minnesota and then they just put them to uh they just brought them to LA and kept the name.
37:59
But that's what Lakers means. Yeah. Wow.
38:01
Bus uh the bus family oversaw the creation of Showtime and presided over the NBA's last three.
38:09
A-listers like Jack Nicholson and Leonardo DiCaprio have become fixtures at the games.
38:12
And when they sell merch, they need to pay sales tax.
38:15
They should get on numeral. com. hummeralhq.
38:16
com sales tax on autopilot.
38:20
Spend less than five minutes per month on sales tax compliance.
38:26
You know all the 11 championships since 1980.
38:28
Their rosters have boasted many of basketball's brightest stars.
38:31
Magic Johnson, Kareem Abdul Jabar, Kobe Bryant, Shaquille O'Neal, and LeBron James and LeBron James's son have all worn the Lakers purple and gold. I love it. It's such cool.
38:43
Yeah, the the father-son duo.
38:45
I mean, I feel like that should have been a bigger like national news story. It's such a cool thing.
38:48
I I think it's like not like if they were like winning championships together immediately, that might be a different story, but it's just so insane that you could be playing professional basketball because you could have earned a better return by DCing into the into the into the stock market.
39:05
That's not why people own these assets, though.
39:07
Owning the Lakers for a number of decades, I imagine, was absolutely priceless.
39:12
So, um, great investment, the owners, great run.
39:17
Yeah, all the perks, you have to add those in. What do you get?
39:19
Perks from DCA into the I like how, uh, Lakers legend Magic Johnson hit the timeline said just like I thought when the Celtics sold for 6B, I knew the Lakers were worth 10 million. Let's go.
39:32
The confidence of Magic Johnson, great investor, too.
39:36
He's got a bunch of bunch of good stuff in the portfolio.
39:41
more news on the scale AI uh transaction. So, it's closed.
39:43
I believe that Alex Wong has a badge at Meta and shows up to work at at in PaloAlto and and clocks in at Meta HQ.
39:52
Now uh scale AI is still an ongoing concern is still a company but every competitor is out for blood and they want to take as much of the business as they can since obviously the perception is that scale AI will primarily be working with Meta and that other foundation model labs might not want to do business with met with scale AI anymore.
40:14
unclear if they can separate out the businesses, if they can separate them out fully over time and and sell the position to other investors, create like a diversified I mean even they could even take the company public.
40:25
Uh at which point uh I imagine that it would be a lot less uh a lot less of a conflict of interest or like a fear.
40:32
Um but there's been news that that uh Open AI said, "Hey, we're not training.
40:37
We're not using Scale AI for data anymore."
40:39
Um because it's too aligned with our competitor Llama, maybe.
40:43
Um, but everyone's trying to Yeah.
40:45
A lot of this was very predictable. Yeah. Right.
40:47
I don't think Meta and Scales teams looked at and said, "Hey, if we sell right now to Meta, which is competing in open source AI, we're totally going to retain all of our customers, right?
41:00
Like people aren't just going to immediately turn off."
41:02
And no, they they were smart enough to know what would happen.
41:06
And there was an article, I think, yesterday about OpenAI, you know, ending their relationship with scale.
41:11
But from what we knew like they hadn't been doing much for a while.
41:16
Part of the reason why Merkore had been and they also brought a big function in in house because for some of the more complex tasks it makes sense to generate the reinforcement learning data yourself.
41:26
Um and there's just so many others there's so many other services having like a single point of failure never makes sense for a business of that size. But uh we'll see.
41:35
So the uh the information has an article here about a littleknown startup that has surged, hint hint, past scale without any investors. This is interesting.
41:44
After meta platform scale deal, data labeling is looking like Silicon Valley's hottest new interest.
41:49
That's enormous opportunity for Edwin Chen's surge AI.
41:51
Uh for years, data labeling existed in a tucked away corner of Silicon Valley.
41:56
a critical but unglamorous area of AI where companies like Google and OpenAI hire outside firms to improve their models by laboriously grading the quality of what they produce.
42:07
Now a spotlight has unexpectedly fallen onto the field in the wake of Metplatform's decision to pay 14.
42:13
3 billion for 49% of Scale Aai the best known data labeling firm.
42:17
But it's not the largest such firm nor perhaps the most impressive.
42:23
That title belongs to Surge AI, founded by Edwin Chen. This is fascinating. I didn't know this.
42:31
1 billion in sales last year. Bigger than scale. Yeah.
42:33
So, Chen startup has won customers like Google, OpenAI, and Anthropic.
42:37
It's such a It's such a testament to the idea that like sure you can bootstrap, but you it's it's so incredibly hard to have any hype around your business if you're bootstrapped because you're not having your investors aren't hitting the timeline for you. Yep. on a daily basis.
42:54
And also, you have, if you're not trying to raise capital, you have less need to go and be loud and and go on podcasts and and talk to the press and all this stuff because you're just making a lot of money and and you know, sometimes it can be beneficial for to for people to not know about you.
43:11
So, this is I mean, this is crazy crazy stats. So, Chen is 37.
43:15
He has no investors and has bootstrapped a 5-year-old startup entirely by himself, which has 110 employees in offices in New York and San Francisco.
43:24
The company generated more than $1 billion in revenue last year.
43:28
Serge has told employees a previously reported figure that exceeds the $870 million Scale generated in revenue uh during the same time period.
43:36
And unlike scale, Serge was profitable and has been from the beginning, Chen said.
43:41
Moreover, Serge could see its sales get even larger if other companies copy OpenAI's decision to stop hiring Scale, a choice made over concerns about Scal's relationship with Meta to shift business to Surge.
43:51
Other key financial metrics couldn't be learned, like how much revenue Surge keeps after paying its workforce of mostly contractors.
43:57
So, there is a question about like the margin since this is somewhat of a marketplace business.
44:01
This could be a situation where, you know, uh, a $1,000 contract comes in and $800 of that contract goes to the actual contractor who's doing the work of the data labeling.
44:13
But at the same time, even if it's 200 million in like, you know, like grow net revenue, that's still a huge business.
44:21
I I it's hard to imagine Surge not being a fantastic business if they haven't had to raise money.
44:25
They have 110 employees and they're used by Google and all these major foundation model labs.
44:31
uh seems like a fantastic business.
44:31
Uh but if Serge could earn a valuation from investors similar to the one scale received for Meta, such a price would make Chen a billionaire many times over, at least on paper, and quietly one of the wealthiest people in tech. Interesting.
44:45
I'm very interested to see what uh what he did before this company.
44:50
Uh Edwin Chen, I feel like I've heard that name before, but I don't know.
44:52
Um as AI models transform from toys into real business tools, data labeling is becoming more and more essential.
45:00
contractors hired by like by companies like Serge grade the responses from AI models and write thousands of questions and answers in fields like programming, math, and law to feed those AI models.
45:09
And so, you know, if you're I I I wonder if this is going to go the route of, you know, you are Deote or McKenzie and you're going to have your team, but then also a company like Serge create a ton of training data around a specific workflow that is costing your business, you know, 20 or 50 or hundred million every year.
45:29
And then so it's like instead of like the the AI BDR that's like kind of generically writing emails based on like the average of the entire internet.
45:39
It's like no, this is a fine-tuned for your business.
45:40
Perfectly trained, perfectly and it and it really it really distills what you do excellently. Um yeah, I don't know.
45:48
I don't know if it'll go that way.
45:51
I'm interested to talk to people about it. uh as AI models.
45:54
Uh so Serge's subsidiary data annotation tech says workers get paid to train AI on your own schedule with wages starting at $20 an hour.
46:03
Chen has distinguished Surge by making uh it the high-end shop charging premium rates often two to five times what scale might bill.
46:10
Surge justifies the prices with its reputation for industryleading work.
46:14
Indeed, one former Scale employee said Serge often performed better than scale in customer audits of labeling quality.
46:20
and competitor Garrett Lord who's coming on the show today uh who runs Kleiner Perkinsbacked Handshake readily acknowledged that Chen is the number one player.
46:29
So I'm excited to talk to Garrett Lord today about this exact topic.
46:33
Should be very interesting.
46:33
Uh, you wouldn't know that from the from the coverage of Meta's blockbuster deal to quasi acquire scale AI, its CEO Alexander Wong, who is now joining Meta in a senior AI role, was widely regarded as leader of the data labeling field and had become a Silicon Valley celebrity, blanketing podcast and conferences with his presence and posting heavily on X. It also raised 1.
46:54
5 billion in venture capital, putting scale on a very short list of companies that have raised that much.
46:58
And he hired upwards of a thousand people.
47:00
Wong had made it timed his exit perfectly given the traction of surge which had grown larger than scale without outside capital and with a tiny fraction of scale's workforce.
47:07
Scale also missed the goal to hit a billion dollars in revenue last year but scale scale spokesperson said the company scale wasn't profitable either was not profitable which but wasn't burning a ton of money like I think they had they raised 1.
47:21
5 billion and they still had like almost a billion in cash. Yeah.
47:22
So, they weren't they weren't in like trouble or anything, but at the same time, it was like like not not a wildly profitable, not a wildly lean business, but I don't know what what what a it's absolutely fascinating these two businesses. It's a wild industry.
47:38
Uh something that like Yeah.
47:40
I mean, just it it feels like there's such an edge just to even identifying this opportunity years and years ago.
47:44
opportunity years and years ago. I mean I guess search started four or five years ago but it was certainly like pre- chat GPT that all these companies got started and then they realized like some of them got started in self-driving car annotation all sorts of stuff like that
47:58
but uh Chen studied linguistics and math at MIT came to the idea for his startup after leaving college and witnessing firsthand how big companies struggle with data before starting Serge Chen worked as m machine learning engineer at Facebook Dropbox Google and Twitter he worked at four different tech companies just like going from one to the next. That's insane. He was developing That's insane.
48:18
He was developing recommendation and search algorithms and helping gather the data needed to train them.
48:24
Despite the hefty resources of those companies, Chen encountered a lot of problems.
48:27
At Facebook, for instance, Chen was tasked with helping build a Yelp competitor.
48:30
His team needed to train a model that correct could correctly classify businesses, telling the difference between restaurants and grocery stores, for instance.
48:37
to do so that he needed a data set containing 50,000 accurately labeled businesses which he found out would take 6 months for an outside firm to assemble.
48:46
We had no solution other than waiting. We simply waited.
48:51
When the data came back, Chen blanched.
48:52
In some instances, it had labeled restaurants as coffee shops and coffee shops as hospitals.
48:56
The data was complete junk.
48:58
He wouldn't say which vendor Facebook had used.
49:00
In 2020, in 2020, he left Twitter to found Surge and picked up some of his first customers.
49:08
executives from Airbnbs and Neva, a once promising AI search engine startup, as only as only a founder in San Francisco might, bumping into them at rock climbing gyms in the city's Dog Patch neighborhood in the Mission District.
49:20
Talking up his startup to get Serge going, Chen recruited data labeling contractors he knew from his previous roles and funded the startup using his savings.
49:28
He wouldn't say how much he put in.
49:29
Fortuitously, Chen focused on language modeling.
49:32
Scale, by contrast, started out using more visual data for autonomous vehicles, which we talked about. Wow.
49:37
Just as those types of models began to grow in importance, uh, less than a year later, OpenAI had hired Serge to fine-tune its models by teaching them how to avoid producing harmful responses like a racially biased language biased uh, based on research paper the company published together in uh, by 2022 anthropic on the surge customer.
49:55
So they're putting out research papers with OpenAI and still managed to stay this under the radar. Wow. Yeah.
50:02
So look at this uh the label large s data labeling has proved to be a lucrative niche in AI.
50:06
Uh Serge founded in 2020 has over a billion in revenue zero funding.
50:11
Uh scale founded in 2016 hasund 870 million in 2024 raised oh the this is this says funding raised but this is clearly valuation or something because it says 17.
50:24
4 billion which is not what they raised.
50:26
Um, Touring has 300 million annualized, raised 225 million. Invisible. It's interesting.
50:33
Touring too initially was like a marketplace to just hire developers and I think they pivoted into lab data labeling. Interesting.
50:40
Uh, it's the same thing when I work with a cloud provider.
50:44
The enterprise tech customer said, "I don't know the internal expectations for why their services work so well.
50:47
I push a button and I'm glad for the internal work to make that happen."
50:51
And data labing companies typically use various techniques to make sure contractors aren't just dialing it in or phoning it in, I guess, uh, when answering questions.
51:00
For instance, the companies randomly insert questions that have no correct answers or make sure labelers agree on the right answer to a question.
51:08
So, obviously, you scaffold up these these uh these these like responses so that everything's like double checked and then you can kind of see if people are are messing around.
51:16
But, wow, what what a beast of a business.
51:18
I had no idea how big this thing is. Amazing.
51:24
Over in defense tech world, uh Androll has partnered with Rhin Matal, the German uh the German defense tech company prime really uh to manufacture Barra Barracuda and Fury over there. That's very exciting.
51:36
And then we also touched on uh Spotify founder Daniel A leading a uh guess when Ryan Matal was founded.
51:46
That the way you're saying that makes me think it's like 1650.
51:48
I'm putting you on the t I'm putting you on the spot like Tucker put uh what's his name?
51:52
How can you possibly report on the news if you don't know when it was founded?
51:56
This was This was the best bit earlier. I have to read it out. Yeah. Read the muffin man. Tucker.
52:02
Do you know the muffin man?
52:02
The muff Ted goes the muffin man. The muffin man. The muffin man.
52:07
No, I don't know him personally.
52:11
How can you know anything about Drury Lane if you've never met the muffin man?
52:15
This is a post from Zack Stewart's real gotcha. Really, really good.
52:18
Uh, anyways, you can still I give you permission to comment on that even though you don't know. 1889.
52:23
Not I I knew it was really old. 1889. Dustelledorf. Okay.
52:29
I was I was pretty far off 50, but but over 100 years.
52:31
Uh, I mean that Yeah, that that's the same.
52:37
Uh I it's General Atomic is part of like this roll up and the and the and it came out of like I think General Dynamics at some point and the company that that the founder of the company that competes with Fury for the uh for that autonomous program that they're competing for right now.
53:00
um was like the designer of a submarine in the Civil War or something like that.
53:09
Like I'm pretty sure that he's he died more than a hundred years before Palmer Lucky was born. Yeah.
53:14
That's the that's the cultural difference between the two companies that are competing for this like one contract.
53:19
It's like a fascinating dynamic about like how legacy they are.
53:23
Like it's not just if you're in the game for a while, you know, decades after after you start the company, your your greatest enemy will will be born and then they'll have to grow up a little bit.
53:34
They'll have to learn the game and then they're going to come for you. Wild.
53:37
So yeah, Spotify founder Daniel E is leading a $600 million funding round into the German defense startup Helsing.
53:44
This is the This is kind of more of an Anderal equivalent over in Europe.
53:48
Uh valuing that 4-year-old company at 12 billion.
53:53
Uh the b the business makes battlefield AI drones, submarines, and robo fighter pilots.
53:58
Uh it's now one of Europe's most valuable startups and they're using some renders here, John.
54:04
This image, they're render maxing.
54:08
They're render maxing happens. It is a cool render.
54:10
Yeah, look at the water there.
54:12
It's definitely a render. Who knows?
54:14
It's hard to tell these days.
54:16
It's it's so it's be we're beyond the uncanny valley.
54:18
Um, and so, uh, Helsing is is expanding from its origins in artificial intelligence to produce its own drones, aircraft, and submarines as part of a bigger push for, uh, locally made and and owned uh, defense products for uh, European countries and and really companies all over the world. Crazy. It was founded in 2021. Wow.
54:40
By Torsten Real Ry Riyle, a video game entrepreneur.
54:47
Gunbert Sherf a former German defense ministry official and Nicholas Kohler an AI researcher and uh they have partnerships with Saab already uh as well as Mrol. Yeah.
54:58
And tons of American venture capitalists in the deal.
55:03
You got Lightseed Excel and General Catalyst Conor's cooking. They've raised over 1. 37 billion.
55:07
Uh the the the Daniel X said the world is being tested in more ways than ever before.
55:17
that has sped up the timeline for Helsink's financing X said pointing in particular to the conflict between Russia and Ukraine where drones and other AI powered systems have been deployed at scale for the first time there's an enormous realization that it is now really AI mass and autonomy that
55:34
is driving the new battlefield and so uh yeah exciting exciting deal up next we have Katherine Han from Han Ventures coming in the studio to talk about the stable coin bill the genius Bill is slated to pass on the front of the Wall Street Journal today and is the guiding and establishing national innovation for US stable coins. We had
55:55
We had her on the show and uh yeah, we're excited to talk to her. So, welcome to the show.
56:01
How you doing, Catherine? What's going on? Hi. Hi. How are you guys? We are great. Uh big 24 hours.
56:08
Congratulations, I think, are in order, but please get us up to speed on what's actually happening and where uh where things are in Washington.
56:14
And I love that you called me Katherine.
56:15
That was my Washington name.
56:17
I haven't been called that since I used to appear in court.
56:20
But uh Katie, let's update the Let's update the Chiron. Katie Han.
56:23
So uh thanks for having me on, guys.
56:27
Actually, I literally just walked off a plane.
56:29
I was down at the COTU conference.
56:30
Um nice where I was talking to a lot of founders, Crypto and Non.
56:32
Uh and everyone loves your show.
56:35
So I I'm happy to be here.
56:37
Um well, thanks for making time. Yeah, thanks.
56:39
I think congratulations are in order.
56:42
I don't know if I would say that yet.
56:43
I mean, first of all, the bill passed through the Senate. Um, great great news.
56:49
Obviously, it's another signal.
56:49
I think kind of like I feel like I did when ETFs um were approved.
56:54
By the way, not really by the SEC, although um that was the body that formally approved it.
57:00
But, as I said last time I was on your show, guys, make no mistake, the DC circuit, that article 3, that other branch of our government, left the SEC no choice.
57:08
government, left the SEC no choice. uh the courts don't always get it right but sometimes they do and the courts unanimously in that case uh said that the SEC had acted arbitrarily and capriciously so to my mind the court system is the reason we have ETFs in this country today for Bitcoin um and
57:26
for ETH and I think of this as a bit a similar thing now here we have another branch of government stepping in in this case you have the Senate passing this um introducing this legislation passing this legislation Now, of course, the House has to vote on it and then the president has to sign it into law. Um, so if those two things
57:45
Um, so if those two things happen, you can say congratulations are in order for the industry.
57:49
But I think one thing that is not really being discussed and I hope we can discuss today is there's another very important bill.
57:58
Um, and I'm not going to slap a percentage on and say which bill is more important, but that's the market structure bill.
58:02
And to me, that's really the transformational bill um for the crypto industry. Okay. So, break Yeah. break that down for us. Yeah. Yeah.
58:10
So, there's one bill um stable coin bill as you just mentioned that the Senate passed.
58:16
And one of the things that I loved to see about that is the bipartisan support um for that bill because I think uh we used to be as an industry pre-political.
58:23
Brian Armstrong always talked about the industry being pre-political and I think I don't want it to be the case that this industry is too political on one side or the other.
58:34
So, I really love to see these moments like we saw yesterday in the Senate where you have a number of Senate Democrats voting in favor of sensible rules of the road.
58:42
And I think this is a classic example of that.
58:45
So, I'm delighted and I hope the House passes it.
58:48
But I don't see why we have to choose between a just a stable coin bill and the market structure bill.
58:54
And the stable coin bill obviously paves the way for a regulatory framework for stable coins in this country. So that's that.
59:04
But then there's another bill, the market structure bill.
59:05
And that kind of will answer or attempt to answer the question of what's a security, what's a commodity.
59:12
And I know you guys have been covering crypto for long enough.
59:14
You know, this is kind of an age-old debate.
59:18
Um, and you know, where can there be an enforcement action?
59:20
Well, the big question is, well, what's a security and what's not?
59:22
And I think the market structure really goes a long way in answering that question.
59:26
And that's why I think it's so fundamental.
59:28
And we've seen courts across the country weigh in on that question.
59:33
Um and and that's because we don't have legislation.
59:38
And Chancellor said it was also clear and of course it wasn't also clear. Yeah.
59:42
So what what's the what's the timeline there?
59:44
What what are the different players?
59:46
What do people want out of it?
59:48
What what is the core, you know, kind of uh crypto industry and both on the investor side and and company side want out of it?
59:55
and and what are what who who would who would not want it to go through at least how the the crypto side has it in mind?
1:00:05
Look, I think I think everyone in the crypto industry wants the stable coin bill to go through and become law.
1:00:09
I don't think there's really any question about that.
1:00:12
The question is, do you go for both? Yeah.
1:00:14
Um, someone just described it to me as in a sports analogy, and I'll probably fumble that one, but it was like, do you go for the field goal or do you go for the touchdown?
1:00:23
And the thing about after a field goal, you know, after a field goal, the other side gets the ball back.
1:00:29
And I think Congress really operates in kind of six to 10 year windows.
1:00:31
And you know, it's on their mind.
1:00:34
Reform, regulatory clarity for crypto right now.
1:00:38
And I think the transformational bill for crypto right now is very much that market structure bill.
1:00:42
We also want that stable coin bill.
1:00:45
And we have this unique moment in time where we have bipartisan support for both bills.
1:00:49
So personally I say go for both of them and worst case you end up getting the stable coin bill only.
1:00:58
Um but I do think that's not the most desirable outcome for the crypto industry.
1:01:02
I think the crypto industry deserves especially after years of uncertainty both a clear message from Congress.
1:01:09
I think Congress ought to do its job and pass both bills.
1:01:10
And I especially think that because after the Supreme Court last summer uh in a case I said that this was the most important case for technology policy, not for just crypto but for tech policy in decades was the overturning of the Chevron doctrine.
1:01:26
And that takes power kind of away at a high level away from regulatory agencies and puts it back more in the hands of the courts.
1:01:32
And do we want to have another 10 years of litigation percolating up from the district to the appellet court to the Supreme Court on what's the security or what's a commodity um in a patchwork of different answers throughout the country or do we want Congress to answer that question for us?
1:01:49
Now and I think it's incumbent for Congress to answer that question now.
1:01:53
You asked who what does the industry want?
1:01:55
I think people this is a big industry.
1:01:58
We say crypto is not a monolith.
1:02:00
It's a broad new asset class and stable coins are a very big important piece of that asset class and a growing piece as you guys saw.
1:02:09
I shared with you the stats um you know almost a quarter of a trillion dollars.
1:02:14
I think you banged the gong for that stat uh locked in supply growing enterprises um across the world integrating stable coins.
1:02:23
You probably saw those announcements from some of the big tech companies in the last few weeks.
1:02:28
And I think one thing that everyone's wondering is those ones that haven't yet dived in for stable coins is well what are the rules?
1:02:34
So this bill that was passed yesterday out of the Senate, the Genius Act is so important for that. It's like here you go.
1:02:40
And so what kind of institutional interest will be unleashed once that bill becomes law.
1:02:48
I think that's really exciting.
1:02:48
What's your h how would you how do you think about big companies getting excited about stable coins and thinking or or institutions and being like there's a lot of potential here.
1:02:58
We should create our own stable coin versus we should just figure out how to leverage this technology.
1:03:03
Where where do you see the kind of line and opportunities?
1:03:05
Well, look, we already have two very dominant stable coins right already today.
1:03:11
Um Tether USDT and then we have USDC.
1:03:13
Um so clearly it's not winner or take all.
1:03:17
I mean those two are both growing and they both have market share.
1:03:21
So we think there will be stable coins like those that will have network effects but we also think at the same time there are some businesses that are just so big and just so important and if they launch their own stable coin um we could also see that too.
1:03:34
Um I don't think we see a world where everyone I think we get asked this question all the time.
1:03:39
Is there going to be a world where every company has their own stable coins?
1:03:41
You know um we don't think so.
1:03:44
We could be wrong, but that's not the view of how we see this evolving.
1:03:49
We do see Yeah, we had we had Aaron Aaron Frank from Lightseed on earlier and he was comparing uh certain companies would launch a stable coin and they it maybe would feel like Koh's cash where it's like, yeah, I don't have any of that, nor do I uh I don't have any.
1:04:03
And and that's the thing.
1:04:06
If I did, would you want to use it on other platforms? Right.
1:04:10
I mean every bank could launch a Visa network competitor theoretically but that doesn't necessarily make sense. Yeah.
1:04:16
So so in your mind is is the broader market structure bill the kind of thing that could catalyze a massive amount of new activity in from my view you know stable coins have been getting adoption.
1:04:26
There are a bunch of exciting use cases.
1:04:28
We have a public American stable coin you know issuer in circle now.
1:04:34
It feels like yes, regulatory clarity is important there, but having broader clarity around how tokens are treated by, you know, how the government actually views tokens feels like it could catalyze, you know, much more of an explosion in investment activity and new company formation and you new use cases for tokens.
1:04:53
Is that the right framework?
1:04:55
I think that is the right framework because like I said stable coin is a big important piece of the pie.
1:05:01
Um and and very lowhanging fruit by the way to my mind.
1:05:04
Um but crypto as an asset class is much broader.
1:05:07
So when you say where do the industry players, where do the crypto investors, where do we want to see it?
1:05:12
I don't think it's a monolith.
1:05:13
I think crypto as it grows to a multi-t trillion dollar asset class like any multi-t trillion dollar asset class you have different factions.
1:05:20
And I think some fairly and some really smart people think just take this take this win and move on and don't worry about the rest.
1:05:29
And I I think that's a little I see why they might think that if they think that it's this or nothing, but I don't I think that's a false choice.
1:05:36
I think that we can have both and this isn't a really opportune moment to have both.
1:05:42
Um and I think the the question that has belleaguered the industry really has been the question over securities, commodities, um which agencies are going to have jurisdiction.
1:05:52
I think that's a bigger question and I think it would be a real shame if we let this moment go to waste.
1:05:56
Do you know the irony too?
1:05:58
So you have some stable coins who are only stable coin companies who are like yep genius act and move on.
1:06:04
they don't want to get dragged into this broader um kind of a more omnibus package, right, with the market structure legislation.
1:06:11
But I think that's a missed opportunity and I think we'll be sorry as an industry if we don't go for both now.
1:06:16
Again, go for the touchdown.
1:06:18
Um and with you know, I think all of the like I told you all of the fundamentals are kind of working all at once together now when I last talked to you guys and this is very much part of it.
1:06:29
So why would we not go for that?
1:06:29
So, I think some folks who don't maybe have an appreciation for how sometimes, sorry to say it, slow Congress operates.
1:06:37
They've been trying to update the moneyaundering laws for two decades.
1:06:38
Um, and it's like out of sight, out of mind sometimes.
1:06:43
And I think we have a really unique moment to press for both here.
1:06:47
And the irony is some Democrats who are opposed to market structure, um, you know, h they because of abuse, potential abuses or who are opposed to the crypto industry. They cite abuses. They cite fraud.
1:07:00
Um they cite speculation. They site Trump coin.
1:07:02
And I I get all of those criticisms.
1:07:04
But I'll tell you what, had the market structure bill been passed, that would have answered a lot of those questions. Totally.
1:07:11
The irony is if you would have had the market structure bill, you wouldn't have had a lot of the blowups that you had in the past several years in this industry. Totally.
1:07:18
Is crypto truly coming home to America?
1:07:20
We went through a period where crypto is being pushed offshore.
1:07:25
We've heard over the last year that some crypto founders feel like they can come back to the states now, maybe actually have an office state side.
1:07:34
Are you seeing more and more momentum there with with some of this positive regulatory movement uh or is it or are you still seeing momentum around uh places offshore, Singapore, etc.
1:07:44
I think look, everywhere that you're going to want to develop does is going to have some rules that you're going to have to follow.
1:07:53
follow. when I hear founders who say there is no regul it's often that's gives you the answer that there is no regulatory regime whatsoever sorry bology if you're watching but uh I I am seeing that onshoring a bit we were kind of seeing the offshoring in an unfortunate way but
1:08:09
it's not only regulation that matters it's hiring top talent and certainly there's top talent right here in Silicon Valley and other places um in the world I mean you mentioned Singapore Singapore has top talent to be sure but you know there's a lot going for the US. So,
1:08:22
So, we're still very optimistic and I don't think it but we were getting to a very dangerous point with the crypto industry um had the likes of Gendler and others like him been left kind of to just do this.
1:08:36
Now, fortunately, the courts were pushing back.
1:08:38
Um so, it wasn't a partisan issue.
1:08:41
It was just the courts were starting to say, "No, you've gone too far."
1:08:44
I mean, I lost track now.
1:08:44
I literally lost track of how many federal courts of all political persuasions uh and appointments ruled against Gensler's regime and not just Gensler but others like it where you had very activist regulators um who were really far out of their lane and um and you saw courts curbing back on that.
1:09:03
So, I think that was already we're in a dangerous spot, but the courts were maybe going to save us, but you don't only want to rely on litigation to save you, of course, because then you've already lost.
1:09:12
But I think hiring talent is important.
1:09:14
I think um you know, access to capital and traditional venture, maybe that's a little different in the crypto asset class.
1:09:21
class. But I think also fundamentally what you have going on right here now with AI particularly in Silicon Valley and we've talked a little bit at the early stage about some of the synergies between AI and crypto because of course AI creates digital abundance and
1:09:36
blockchains are good at enforcing digital scarcity and I think you're going to see more and more synergies emerge and use cases over time and I'm not going to say what they are because you know we've seen that before in crypto a lot of overpromising underd delivering on use cases. Um, so let's
1:09:49
Um, so let's just stick to right now we see synergies.
1:09:54
There's a lot happening in Silicon Valley obviously with AI.
1:09:56
We think that's going to benefit the crypto industry.
1:10:01
And so in addition to regulatory clarity if you're a founder, you want access to great talent.
1:10:05
You want your visa situations sorted out for your employees.
1:10:09
You want um access to other founders.
1:10:13
Depending on what type of company you are, you want access to capital.
1:10:16
So I am optimistic um about the state of crypto in the US but also elsewhere and we invest in companies uh we invested in squads we've announced that um and I know Steen one of the founders of squads watches your show um but Stephen's based right now um overseas and I was just having a conversation with him about how do we get you to come and bring squads to the US. Awesome. Awesome.
1:10:42
I I I I want to talk about the longer tale of regulation because it seems like the stable coin bill is very straightforward like the least the the the least ambiguity there.
1:10:55
Then you have the the the market structure act and there's a lot more to do there.
1:10:59
do there. But I'm sure that there's like riders getting pitched and all sorts of longtail things like how much are we actually like where does the line end between what we're actually trying to define versus what we're still in the exploration fe uh phase of because you
1:11:15
have NFTTS crypto gaming there's uh you know uh uh prediction markets there's so many different crypto applications that trying to kind of do them all at once maybe that's the right approach maybe these need to be handled like after we've done the technological exploration, but what's your view on kind of the long term? My view on that
1:11:33
My view on that is is no because we can't just wait.
1:11:34
We can't have an NFT bill, a bill for bank contracts or that we can't have specific bills.
1:11:42
And if you saw if you think back to the advent of internet, that's not what we had.
1:11:45
You know, we had section 230.
1:11:47
It applied broadly to platforms.
1:11:50
And I think we need something similar here.
1:11:51
here. So on the one hand I would say it can't be um so specific that it's like okay if you're a um you know events contract platform it's this rule and if you're an NFT player because again we don't even know yet what will be created um uh really at the end of the day we've seen some early
1:12:09
use cases with product market fit obviously chief example of that is Bitcoin but what if we had had this conversation guys back in 2010 and you said okay we've got Satoshi's white paper and there's this thing called Bitcoin let's pass some crypto regulation. It would have just been for
1:12:22
It would have just been for Bitcoin and that would have been a mistake because then couple years later on the scene we have ETH then a few years later we have Salana.
1:12:28
So I think what we need to do so you don't wait for the end state to have any regulation, right?
1:12:35
You don't do regulation by enforce.
1:12:36
I can tell you what we don't do.
1:12:37
We don't do regulation by enforcement.
1:12:39
enforcement. you don't wait till the end state um of things but nor do you want to get so with such specificity today and do the the current state of regulation by the end state of regulation um that you can't do either and so I think what you have are some guiding principles some generic rules of
1:12:58
the road and really that's all the um market structure bill is and there's enough clarity that you had Democrats vote for it last time uh it came up so it's not like it's so specific it talks about and I think look I think you have people like Hester Pur who is SEC commissioner has written and given speeches on this of what that ought to look like. What is a decentralization
1:13:17
What is a decentralization test?
1:13:20
And those are some guiding principles that you can kind of look at and apply as you think through this legislation.
1:13:25
You know what body ought to regulate it?
1:13:27
Um and and and sure you're going to have some outliers and new technologies emerge that you're like okay does this fit neatly in the bill? No.
1:13:35
But we have laws for everything in this country with technologies that develop that don't fit neatly in a particular bill.
1:13:41
And that's why we have that's why we actually have we don't do regulation by enforcement.
1:13:45
Uh we do notice and comment.
1:13:47
We do things like advisory opinions.
1:13:50
Certain bodies by the way do do advisory opinions, not uh judges.
1:13:55
And then if if all else fails, uh you do go and sometimes seek article 3, seek a judicial interpretation.
1:13:59
But I think that's like I said that's kind of the failure state if you're having to go to the courts.
1:14:05
Um but indeed that's what was happening because we were getting no rules of the road and we were only getting unfair regulation by enforcement.
1:14:13
And I say unfair because Gary Gensler basically picked what should have been the best companies in crypto the poster children for compliance and brought enforcement actions against them and then the complete spectacular disasters didn't bring anything.
1:14:28
So, uh, until after the fact and, um, so I think regulation by we gota have we gotta have you and Gary on the show to hash it out.
1:14:35
I'm sure you guys have had some funny conversations.
1:14:40
Thank you so much for hopping on for having me.
1:14:42
You have a good rest of your day. Cheers. Okay. Byebye.
1:14:45
Up next, we have Justine Moore from Andre Horowitz.
1:14:51
Incredible map knowledge.
1:14:51
Dropped a fantastic market all around AI image, AI video models.
1:14:57
We're going to have her take us through it. Welcome to the show. How are you doing? Can you hear us? Can you hear us? Hello.
1:15:08
Oh, I can hear you guys now. Sorry. There we go.
1:15:09
Hey, I was saying to John when you dropped your new market map, I was saying, you know, it's a great sign of respect in our culture to drop a new market map and and come on the show.
1:15:19
So, we're incredibly bullish on market maps.
1:15:23
We find them extremely interesting and I think they got a bad rep a couple years ago, but I'm glad that you've stayed the course and we are pro strongly pro market map.
1:15:31
Yes, the people love market maps.
1:15:33
It's like you can guarantee a popular tweet if it has a market map in it. Absolutely.
1:15:37
And I think people got kind of sick of like oh like we know the playbook, we've seen it, but there's a playbook for a reason. It works.
1:15:42
So yeah, take us through the latest market map.
1:15:46
Um what are you tracking?
1:15:46
How did you decide what to divide it up into?
1:15:50
and and what uh what kind of inspired this moment specifically? Yes.
1:15:53
So, I mostly do AI creative tools here at A16Z.
1:15:58
So, I spend like all my time testing all of the image, video, audio, etc.
1:16:03
Um, and obviously the past few months in particular, video has been like the thing.
1:16:07
Um, there's been like V3 obviously, which was a massive moment with adding the audio for the generations.
1:16:13
Um, Hedra around the talking characters.
1:16:15
um the new Miniax model, the new bike dance model, Seed Dance, which is in the arena already outperforming V3.
1:16:23
So, it just felt like a good time to refresh sort of what's going on in the video space.
1:16:28
Um, and I kind of formatted this market map just thinking about um more from the perspective of a creator, like less in terms of the go to market of the company, more just like if you're a person trying to create a video with AI, where would you go for these different use cases?
1:16:45
So there as far as sort of the model like the foundation model companies where it's either text to video or image to video most places do both um where they actually take your input generate have their own proprietary model that generates the video for you.
1:17:02
So that's the the vos, the clings, the runways, the pikas.
1:17:05
Um and then there's also now this emergence of what I call like multimodel apps, places like um Korea and and Flora and Visual Electric that enable you to run um a bunch of models in one place.
1:17:19
So like if you want to take a single prompt or a single image and see what it looks like in five different models super easily, you can do it somewhere like that.
1:17:29
Um, and then the other side of the market map is sort of like what happens when you add speech and talking characters.
1:17:35
Um, and so some folks do that by like generating a talking avatar from an image where you can eventually have the person move and other folks do that by taking like a video of a person and then um applying lip sync over it and then syncing the audio.
1:17:49
So that's sort of the distinction between talking avatars and lip-sync.
1:17:52
Yeah, it's interesting that you're in the consumer group because I can imagine that a lot of the talking avatar companies wind up selling to corporations that want to vend in talking avatars for example.
1:18:02
Are you seeing a lot of that um or like I guess how how how uh rigorous or uh stringent are founders about like hey we are trying to build a consumer app or we're just building a cool technology and it might land as a consumer product but it also might land as a B2B play.
1:18:18
um most people are not at all rigorous and often don't even know at the beginning.
1:18:24
So what we actually saw with the first generation of AI video was it was only researchers making these like magical models and they had no idea what the use cases were going to be.
1:18:33
They all just put like a text prompt box in front of the model and then you got an output and like a big company and an individual were using the exact same interface.
1:18:44
Now, I think we're starting to see more at like what we call the app layer, which is essentially like how do you productize this?
1:18:49
How do you create workflow?
1:18:51
And and therefore, how do you go in to specific verticals?
1:18:53
Um maybe an example of that is all of the like standalone video ad um creation products.
1:19:00
So things like Creatify or captions where or Hey Jen has a product for this too where um you can literally just like paste in a link to your Amazon or Shopify store.
1:19:09
It will pull all of the info about your product, your logo, your brand, and it will generate a talking head avatar like holding your product and describing it.
1:19:18
And that's something that like a V3 or Cling, the general video model companies won't do today. Yeah.
1:19:24
Um, where are you seeing the strongest like low churn adoption of these tools?
1:19:31
Because just personally like V3 was the thing that got me to subscribe to Google Pro Max 25 which we can go into the names and how difficult it is to access these.
1:19:42
Um but but other than that uh there there haven't I haven't seen that many where like we've seen the ASMR Stormtroopers or something but I but it hasn't been clear to me that someone's building like the next Pixar and they're actually thinking about it like Mr.
1:19:56
Beast and they're like I'm building a studio. This is a business.
1:20:00
I have ad integrations and I have a content schedule.
1:20:02
It's very much in like the testing phase.
1:20:03
We see the Studio Giblly moments go viral.
1:20:05
So where where is like the true long-term value playing out right now? Okay.
1:20:11
On the content creation side, where we are at right now is actually there's a bunch of content agencies like you know there's a bunch studios, Dream Studios, there's there's probably like 20 of them now.
1:20:24
Uh one called Paracosm that's that's really cool.
1:20:28
Um, and these are people who are just early adopters of the tools and they're getting hired by brands and ad agencies and entertainment companies to use the tools for them uh and make content.
1:20:39
I think the problem now in what you're describing is like the people at Pixar would have to know how to use the AI video tools and how to set up the workflow in order to create their next movie using AI.
1:20:51
tools and people who enter AI native agencies or contractors today.
1:21:03
Um I think like one of the first AI IP we've seen is the Italian brain rot characters.
1:21:09
I don't know if you guys No, I don't know this. I haven't seen this.
1:21:12
The Italian brain rot characters are huge.
1:21:14
So I can't even say the names here because they will sound ridiculous.
1:21:16
But um it's people basically making images of like an animated talking baseball bat and like a ballerina whose head is like a cup of cappuccino. Okay. Okay.
1:21:25
And a shark who wears sneakers. Okay.
1:21:28
But they're starting to build like a cinematic universe. Oh, it's massive. Yeah.
1:21:33
Accounts have like millions of followers.
1:21:35
The world isn't ready for Italian brain rush.
1:21:36
So So is it is it uh is it like decentralized in the sense that like I could just go and participate in this like broader trend? So yes and no.
1:21:46
I would say there were a couple accounts that originated the first few characters and then other people started participating using the hashtags remixing. Okay.
1:21:54
And then the good the characters that were good that came out of that sort of bubbled up to become part of the cinematic universe the canon.
1:22:03
They have bombadiri bombadro crocodio which is a a bomb.
1:22:08
He's a crocodile plane that shoots bombs. This sounds like Yeah. Yeah. Super super viral.
1:22:13
Yes, the kids probably love it. It's wild.
1:22:16
Um before we go into more of the consumer side, talk to me about some of the uh the more like uh niche like B2B use cases because I remember when like GPT 3.
1:22:31
5 and we got four four uh GPT4 dropped, there were still like a lot of hallucinations, but you saw companies that were just like, yeah, like sure it's not incredible at writing poetry, but it's amazing at just like converting this messy text to JSON.
1:22:45
and they were all of a sudden just running tons and tons of queries.
1:22:49
And so I would imagine that in a in a in a in a video workflow, things like what Runway was doing in the old days of just like green screening or the the stuff that like artists aren't going to really get upset about because it just feels like a better, more advanced tool.
1:23:04
Are are a lot of these companies building tools like that or is all the focus just on like let's oneot the next Oscar film?
1:23:10
Yeah, it's a great question.
1:23:13
There's a decent amount of vertical focus.
1:23:14
I think also like it's very hard for this for startups like if you're not a Google to or an openi or whoever to play in the game of like let's train the largest oneshot best video model.
1:23:26
Um, so a lot of them are focusing on verticals like um, Luma, which is a company we invested in, did this really cool tool where you can upload like a a 9 by6 like iPhone style video and you can just say extend and it just basically like outpaints around the existing video and makes it look like so you can just change the dimensions of your video really fast. Yeah.
1:23:45
Um or there's more of just like a practical tool, but super useful for a bunch of creators that are filming vertical content probably and they want to distribute in a horizontal format, so they just do that. Yes, exactly.
1:23:56
Ton that makes a ton of sense.
1:23:58
Or or something like Higsfield that it has all these really special effect like motion luras essentially.
1:24:04
So you can take like an image of a car and just say like this is like here's a template of what a car explosion looks like.
1:24:11
Make this specific car explode. Okay.
1:24:12
Um, and then on the B2B side, we've seen a lot around like how do you scale marketing or L & D or like executive presence type content.
1:24:20
So like, you know, tools like Descript now allow you to take a video of someone talking and then change the word that they're saying by changing like cloning their voice, having them the voice say the new word and then doing a new lip dump, new lip sync. Yep.
1:24:36
So you could have your CEO sending what looks like a personalized holiday message to like every single customer or something like that. Yep.
1:24:43
Or maybe something worse with a fishing scam, but I'm sure we'll get into safety at some point.
1:24:47
What do you What do you think uh the long-term ambitions of companies like Google with VO and Bite Dance with their model?
1:24:55
What do you think they want out of this category?
1:24:58
They obviously have a massive edge.
1:25:00
I saw um Anishh was posting about uh Google's edge or YouTube's edge around IP with VO you can generate um yeah what was going on there the Disney thing was that like a real deal or is that just a beneficiary of like some sort of relationship or Yeah.
1:25:18
And then I I I want to get a sense of of do do Google and bite dance do they want to be developer tools that just vend into a bunch of these platforms?
1:25:25
Do they actually want to own the end customer? What is this market?
1:25:29
Is there a generalized market in the long run of just generating funny videos for the average consumer or is it going to all be verticalized out where I want to generate ads?
1:25:40
I maybe want to generate, you know, customized messages.
1:25:42
But I I want to understand like this is a good overview of all the of the ways you can generate content, but like how does the market structure evolve? Yes. Um, okay. Okay.
1:25:54
On the IP question, I have not talked to Google's IP lawyers and I'm not an IP lawyer, but my understanding is Google But this is legal advice, right? Yeah, exactly.
1:26:01
This is financial advice. All the above.
1:26:03
Our compliance team is going to love this.
1:26:07
Yeah, they're going to love it. Sarcasm.
1:26:09
Uh you can uh I I think my sense is basically, you know, when YouTube came about, it was suddenly like all this IP content is on the internet and Google cut deals with a bunch of the IP owners about essentially what what can be posted on various Google properties and if that content gets monetized, how it ends up going to the end rights holder.
1:26:28
And so that's sort of the the working theory right now about like why V3 can generate IP content and not get sued when a lot of other people are struggling with that.
1:26:35
Um, in terms of the market dynamics, it's so fascinating the question around like Google and by dance and and eventually I think Facebook I hear is going to do more in video soon as well.
1:26:46
Um, like why they're doing it and what their strategy is.
1:26:49
I think first of all like if I'm one of those huge consumer giants, AI is such a massive shift in consumer behavior that like you want if if you want to own the interface to consumers, you probably want to own text, image and video generation as well.
1:27:03
and they have the resources in terms of data like YouTube for example is a perfect example of this.
1:27:10
They have a ton of compute.
1:27:10
They have a ton of money and they can hire the best researchers to build the best models.
1:27:14
I think the question as you sort of alluded to is like do they do they sell those models via API and let other people build the consumer experience on top of them or do they own the endto-end consumer experience?
1:27:27
My honest take on it so far has been like it takes so long in the big company product teams to get stuff done and get new products out that like the model teams are just shipping the models and like pretty basic interfaces like Google Flow and then the product teams are going to figure out if they can catch up with some kind of cool new consumer app later. Yeah. Uh yeah. Yeah. I mean, yeah.
1:27:47
yeah. My my question is like how much uh what will the market actually look like for the average consumer wanting to generate images and video or will it just be something that people default to chat GPT cuz maybe they already have a
1:28:02
subscription or they're fine with the free tier and it doesn't end up there ends up being a bunch of different applications on the enterprise B2B side but then not so many like you know core consumer subscriptions on just like cool videos, pictures, etc. I mean, it seems
1:28:17
I mean, it seems like it's a killer killer moat for Google Cloud Platform to have V3 as an API, even if Google can't figure out how to productize it fully.
1:28:27
It's like they do seem to have a real moat.
1:28:29
I want to get into that about uh YouTube is obviously an incredible training data resource.
1:28:35
Uh you mentioned that there was another company that that just surpassed them. Was it Tencent?
1:28:39
You said uh Bite Dance the Bite Dance. That's right.
1:28:43
So I I have a question about that because um obviously with codegen uh GitHub has a lot of public repos that people can probably just scrape.
1:28:52
It's also just not that much data.
1:28:54
You can probably fit it on a couple hard drives, maybe sneak it out the back and go head a flight and and train somewhere in Malaysia or something.
1:29:02
Um you can't do that with YouTube.
1:29:04
Like it's just too much data.
1:29:05
And so my question is uh is how durable how much should we be thinking about a durable data mode in video generation for YouTube because it seems like something that they could really like clamp down on and would give them a durable advantage but I don't know there's so many other there's so many other ways to attack any of these model developments um totally that there's a lot of different options.
1:29:29
So there's a couple parts of the data question.
1:29:33
The first part is like what do you own versus what do you scrape I mean we've seen companies like open AI we'll we'll scrape YouTube as well to train their models.
1:29:39
I think bite dance also like you know here we think of YouTube and Facebook and whatever here in the US I mean as as being the big host of content but like there's all these massive companies in China like bite dance who have their own um have have their own user generated content on like their version of Tik Tok and their version of YouTube and their I'm sure there's reposts of American videos over there.
1:30:01
So it's not like even has a unique flavor.
1:30:03
Probably can generalize pretty well, right? Totally though. Yeah.
1:30:06
Like I was one of the very early users of all the Chinese video models when you still had to access them on Chinese apps with Chinese phone numbers and they were definitely very good at things that were more um China oriented than the US models.
1:30:20
That makes a ton of sense. Um Oh, okay.
1:30:22
The other the other thing that's important to mention on data is um in video in particular, it's not just the volume of data, it's also the quality of data and the quality of data labeling because essentially you can't just feed a video into a video model and assume it can understand what's going on and and pull out the relevant info.
1:30:40
You have to have really sort of dense labels is what we call them or super detailed captions about like this is this style shot shot from this sort of camera.
1:30:49
The camera is coming from this angle.
1:30:51
this is the sort of character, this is how the character is interacting with the background.
1:30:54
And that quality data is what drives quality in the video models.
1:30:59
Um, and China has really benefited there because there's so many more PhDs than there are here.
1:31:05
And it's much cheaper for these companies to hire them um to do these these dense uh labels for the video data. Yeah.
1:31:12
Well, how does Midjourney fit into all of this now?
1:31:15
It's such an interesting company because no venture dollars, this like behemoth kind of quietly hiding in a Discord server still.
1:31:23
Uh I saw some examples of video. It looked fantastic.
1:31:26
Seemed like they hadn't added audio yet, but how how do they fit into the whole the whole piece?
1:31:32
Because it seemed like early on they they developed a really great um feedback loop for the data that maybe wasn't happening with some of the other model providers. Yeah.
1:31:40
So, the midjourney model came out this morning um really conveniently like 10 minutes after I put out my market map without midourney video because it's not yet available.
1:31:49
Um I was just playing around with it too. It's it's really cool.
1:31:52
They do image to video um and they and so they don't do text to video which is actually sort of easier.
1:31:57
They can start with their the super high quality images that they generate on the platform and then animate those.
1:32:03
I think they have like a low motion and a high motion setting.
1:32:06
Um, from what I've tested so far, um, it's it's better as sort of like a low motion, scenery, environment, light interaction type thing.
1:32:15
Like you have a photo of a person and you can then animate sort of rain and wind and them walking slowly.
1:32:21
And it's not as good at like what I call physics heavy world model type things like two cars running into each other and exploding.
1:32:32
that sort of thing requires um uh a very very large and costly usually like texttovideo model that is more difficult to train.
1:32:40
Um whereas midjourney I mean I I have no idea how they did it. It's a great model.
1:32:45
They could have taken one of the open- source image tovideo models and and fine-tuned it on all their own data. Yeah. Yeah.
1:32:49
I've noticed V3 is really really good with some of that physics stuff, but it still gets confused.
1:32:55
like if a car is driving away, all of a sudden you'll be looking at the front of the car and then the back of the car and it'll get kind of mixed up. Um, but yeah.
1:33:02
Are you uh as as all these different models have progressed, I I always remember uh Brad and Trevor McFed and just how early he was to to what I think will be this like new wave of Is that Lil Michaela? Yeah, Lil Michaela.
1:33:19
Uh, which was like basically CG a CGI influencer. So, very early.
1:33:22
I think we're going to see a lot more of this.
1:33:26
Uh, and we we've seen some of this to date, but do you expect that to be kind of like a new a new like how how bullish are you on on sort of like entirely AI creators getting um getting real adoption following turn followings turning into real businesses?
1:33:41
I'm sure you've you follow a bunch of them already.
1:33:45
Yeah, I'm I'm personally super excited about it because it kind of separates um the content from the character.
1:33:51
Like now, you know, before AI, if you were on Instagram, like you were both the character and the person coming out with the content.
1:33:59
And so, you had to like look in a way and present yourself in a way and talk in a way that was like interesting to the Instagram algorithm.
1:34:06
And now it's like anyone with a good idea can create a compelling character.
1:34:11
Um, and so I think some of those are human characters.
1:34:12
I've already seen way too many examples in my reals feed of only fans models who promote themselves with AI avatars of themselves now, which works shockingly well.
1:34:22
Um, there's some photorealistic human influencers, but honestly, some of the more interesting ones are things that could never be influencers before AI.
1:34:31
So, there's one called like raccoon stole my iPhone and it's a it's an AI raccoon influencer.
1:34:38
There's like AI Capy Bara influencers.
1:34:41
There's like mystical creatures, like all of these things that just come out of people's imagination competing uh for mind share with uh Instagram pet pages, you know, dog pages, things like that.
1:34:54
Uh did you have a reaction to Fountain Head?
1:34:56
Uh it was the the or sorry, Mountain Head, not Fountain Head. Oh, yeah. Mountain Head the movie.
1:34:59
the the core overarching theme was that basically deep fakes or AI generated content had gotten so good that it was causing global unrest.
1:35:08
Uh did did it did it resonate at all?
1:35:12
It does feel like I now have, you know, multiple times a day I'm I'm seeing content online and and it's like getting I we're both in the community notes program.
1:35:22
So it's like you see content getting community noted.
1:35:25
One person says it's not real. Look at this link. It it was this image. They redid it.
1:35:30
Another person says, "It's real. Look at this."
1:35:32
So, it's like I just assume that everything's fake and made up unless I see it, you know, with my own eyes.
1:35:37
But I'm curious if you if you uh if it resonated at all with you.
1:35:42
So, I've not I largely consume AI slop.
1:35:45
So, I have not seen the movie. Um I should watch it.
1:35:47
You should watch the movie the motion picture on slop. Yeah.
1:35:52
Uh once they have that, I will watch the the full film.
1:35:55
But, um it's so we talked about this a lot actually with audio models.
1:36:00
it with the last like election cycle cuz video I think wasn't there yet to have convincing deep fakes.
1:36:04
Um but audio like there were way less cases of even though you could make really realistic voices cloning candidates and saying things that weren't true.
1:36:12
There were way less examples than we thought of that actually like impacting any any election in any sort of meaningful way.
1:36:20
And I think part of it is like things like you mentioned the community notes program where like you have sort of citizen watchd dogs on various platforms saying this is real, this isn't real, running them through various sort of AI detectors.
1:36:32
Um but I also think like people are starting to develop more skepticism around everything they see online and whether or not it is real, which is probably not a terrible thing. Yeah.
1:36:43
I had this take that After Effects would be more impactful on the election than AI video because like you can just show a clip of a burning building from 2020 and it's real video but you recontextualize it and say oh you know this the the capital is burning or something and it's from years ago or or just you know speed up a video, slow it down, edit it out.
1:37:03
They would do this with various politicians.
1:37:05
Uh you know cut out the ums and they'll sound sharper.
1:37:09
add a bunch of gaps and all of a sudden they sound like they're slower.
1:37:13
Yeah, it it even, you know, we're we're here in LA and and when all the imagery was coming out of the the protest from a couple weeks ago, it was like burning Whimos kind of looks like something you'd generate with VO3 just because it was so symbolic and just such a crazy image.
1:37:28
and then like make an image of a guy with a Mexican flag riding it doing burnouts around a a car that's on fire and it's like that looks like and even and even just the way that was photographed there was like it looked like all of Los Angeles was engulfed in flames but it was really like one crazy block with a bunch of different angles and then a bunch of different posts and and you drive around and you be like oh there's not that much.
1:37:50
there's not that much. That's the other interesting thing is like even real footage can be manipulated like any kind of story can be manipulated in a way like AI or are you seeing uh I think like you know there's exciting uh companies like
1:38:05
worldcoin you know doing like proof of human are you seeing any infrastructure players trying to do like do anything on like content verification side and like trying to create some sort of um mechanism to to prove whether something was like authentic, you know, actually shot on an iPhone, right? You know,
1:38:23
You know, proving through the metadata and some type of like public um setting.
1:38:27
Is there is there any pitches uh from from that side?
1:38:32
Yeah, so mo largely honestly today that has come in two places.
1:38:35
One is the model companies themselves will often watermark the content in some way like the V3 generations has little V3 11 Labs which does the audio.
1:38:44
They actually have a site where you can upload any audio and it will tell you if it was generated with 11 Labs or not. That's cool.
1:38:49
Um which which is pretty cool.
1:38:51
The other um place we've seen development there is for like prominent individuals um like you know celebrities or someone who's there's like value behind their brands and who potentially even might want to monetize it in the age of AI.
1:39:04
Like if you're an actor and you suddenly don't have to, you know, film go fly back to LA when you're filming a movie in Australia to tape like five ads for some cell phone brand and you can have your AI avatar generated to do it instead and it looks just as good.
1:39:20
Like you might actually want to, you know, have some licensing company that owns your AI licensing rights, whether it's your traditional talent agency or not, um, who can manage that for you. Yeah, totally. Very cool.
1:39:32
Well, thank you so much for stopping by.
1:39:34
We could talk for another hour.
1:39:35
I have so many more questions in the in our doc, but uh we'll have to have you back.
1:39:38
So, thank you so much for stopping by. Thanks, Justin. We'll talk to you soon. Great chatting. Have a good one. More breaking news.
1:39:44
Nvidia to drop humanoid robots that will produce Nvidia GB300 chips in Q1 of 2026. Nick says, "Holy based." That is so close.
1:39:54
Foxcon in talks to deploy humanoid robots at Houston AI server making plant.
1:40:00
Wow, that is extremely I don't totally understand this. What does that mean?
1:40:04
I guess Fox Foxcon has been training the robots to pick and place objects and insert cables.
1:40:09
Yeah, I don't know if this is marketing.
1:40:12
It feels like we've had we've we've asked a bunch of robotics experts about humanoids and so far I haven't got a I I don't have the confidence that that this is actually a super great use case for them.
1:40:30
It's just the definition of like what is humanoid because you go to a you go to like a Ford F1 Vic F-150 factory and they have like massive robotic arms like moving windshields around, right?
1:40:39
Like it's like you could anthropomorphize that by like spray painting it pink and putting some hair on the on the hand and being like it's a it's a humanoid now, you know, but like you could like retrofit more bicep definition. Totally.
1:40:54
I mean like like Amazon has like tons of robots sliding around.
1:41:00
You could put googly eyes on them and be like they're humanoids now and and you get maybe get like a stock bump.
1:41:05
But like there's clearly things that are happening in the AI server assembly process that are using robotics obviously whether it's even just like conveyor belts or or you know the the the three axis pick and place you know pick it up put it over here that type of stuff.
1:41:22
Just going to humanoids is just it's like well will it have five fingers or will it have a just a grabber?
1:41:27
Will it have legs or will it have wheels?
1:41:31
like it could just be sitting there because for a lot of these pick and place jobs you can just have the the humanoid sit there with a single arm mounted to the ground because the stuff comes to it and so then you're just kind of in a okay we're in the arm business now.
1:41:45
Um it feels it feels like it's a little bit wrapped in marketing lingo but still cool that more companies are making humanoids because uh it does seem like a cool form factor that hopefully people will break through and get there and this seems like you know a step in the right direction in the sense that it's a very defined task.
1:41:59
I think when people think humanoids, they think some like my new touring test is is a humanoid robotics will will be here and when they can put up a six-minute nurburggering time in a manual. Yeah. Gated manual. In a gated manual.
1:42:21
Uh because at that point the they have to not just be a self-driving car but they have to be able to you know negotiate the uh the the wheel and the stick shift so so efficiently and so quickly that they are truly performing. Yeah.
1:42:39
And a manual and a manual can be weirdly like probable probabilistic, right?
1:42:43
And that like you're like you know you can move the synthesize a lot of data. Yeah.
1:42:47
It's not it's not like you know Yeah. perfect system, right?
1:42:52
You're not going to You're not going to put up a sub seven minute nerburggering time with a Joe Biden walk. That's right.
1:42:57
You're going to have to be moving fast.
1:42:59
Those actuators are going to have to be That's right. high speed.
1:43:01
Um so yeah, I I I don't know where all this goes, but uh I mean exciting to see that they're that they're at least doing work on it because it seems like an important it's it's clearly an important path in the tech tree.
1:43:14
We don't know how relevant it is to other to other formats but uh cool to see a lot of money pouring into the sector from very big companies.
1:43:23
So Foxcon will be announcing their robots in November and then deploying them shortly after. Yeah. In our year 2026.
1:43:31
I mean Foxcon seems like the right builder for this.
1:43:34
Uh Unitry is is already like feels like it's scaling up to the point where like Unitere usable in certain locations.
1:43:40
like they're not generalizable yet, but they're certainly if you train them on a specific task, they could do that task over and over and over again.
1:43:47
Uh the question is just like like you know, do you need five fingers, 10 fingers, 10 toes?
1:43:54
Is that is that really like the right form factor or or should we do be just doing things that are more specified?
1:43:59
Uh because if you're if you're if the whole point of these human robots is just AI server making, just assembly, just one one spot on the manufacturing line, could probably be a much more specialized robot.
1:44:09
But we'll be cool to see.
1:44:11
I'm sure we'll see a lot of viral videos about it. It'll be very cool.
1:44:14
I invested in a company making uh robots for data centers. Oh, yeah.
1:44:18
And they intentionally uh chose not to make it humanoid form factor. Yeah.
1:44:25
Which and I think a lot of those the the reason behind uh that decision would also transfer to manufacturing setting, right?
1:44:33
Which is like do you need legs if you can use wheels?
1:44:37
data centers have the the most perfectly polished floors with like no dust at all.
1:44:41
Like it's it's the the perfect environment for a wheeled use case.
1:44:45
At the same time, you probably need to have a very specific actuator for like unplugging and plugging the cable back in, right?
1:44:50
And that's actually like it's pretty hard to reach around the back of a computer and and unplug an Ethernet cable.
1:44:56
And obviously the server racks are like designed to be worked on more, but still even the cabling is like very detailed work.
1:45:01
And so if that's what they're trying to do, um, uh, you probably need a specific actuator for that.
1:45:07
We have our next guest in the studio, uh, George Hots. How you doing, George? Good to hear from you. What's going on? Welcome. Can we hear you? Oh, can you hear us? Yes, we can hear you. Gotcha.
1:45:21
Uh, uh, let's kick it off with something, uh, simple.
1:45:24
I want to take your temperature on, uh, AGI, timelines, pdoom, the the easy and fun stuff.
1:45:33
I don't know what AGI means and I don't know what you mean by deal. No.
1:45:38
Is are are these terms just like entirely irrelevant?
1:45:39
I mean that now we've shifted to like super intelligence.
1:45:42
They're all buzzwords but at the same time like there is there is an idea of like the like I don't know that that the conversation is maybe shifting to like the AI generating more economic value than humans.
1:45:56
Is that a relevant metric to track?
1:45:58
Machines have been generating more economic value than humans since the industrial revolution.
1:46:04
Is there some Is there some other metric that that we should be tracking or is it just like irrelevant?
1:46:09
You're just talking about like hype like I don't know.
1:46:13
I mean I I like I I don't know what you mean.
1:46:17
Like you can talk about concrete things.
1:46:18
The term like AGI means nothing, right?
1:46:21
Like computers everything that's a turning machine is a general purpose computer.
1:46:25
Is that what you call intelligence?
1:46:26
I don't know what you mean.
1:46:28
Is a linear regression intelligent? What if it's big enough?
1:46:31
The Chinese does know Chinese. Yeah.
1:46:31
Um, what I mean, what about uh uh your your decision to get on a spaceship traveling at 0.
1:46:42
9 C away from the from the Earth?
1:46:46
Like, how close are we to that?
1:46:46
Are we closer than the last time we talked, which was like a couple years ago, and it seemed like it was maybe going to happen within your lifetime. Has it moved at all? Yeah. I don't know.
1:46:55
I don't know if I'm actually gonna get that spaceship, but it's kind of like in an ideal world what I would want to do, you know? Yep.
1:47:01
Just just just back away and chill and and don't look back.
1:47:06
Actually, you can't look back. They're all there.
1:47:09
You need the you need the blast shield, right?
1:47:13
You need the information shield. Information? What do you mean?
1:47:18
Oh, that's how they're going to get you. Okay. Right. I mean, okay.
1:47:20
So, like here's a way you can think about AI, right? Yeah.
1:47:24
Um, imagine there were 10 CIA agents assigned to you and they're running at a thousandx real time.
1:47:30
So, they're like hyperfast CIA agents that devote their entire lifespan to your day. Mhm.
1:47:37
And they're trying to manipulate you.
1:47:39
Maybe to get you to buy things, maybe to get you to vote for a certain guy, whatever.
1:47:44
But like that's what you're going to be up against with AI.
1:47:47
What we're currently building, what what what if you think about the biggest companies in AI, what they do is advertising.
1:47:52
What advertising is is just manipulation of humans.
1:47:54
Um, so you're going to have a team of CIA agents thinking about you and trying to manipulate you at all times.
1:48:00
And now you see why you want to head away at the speed of light, right? Mhm.
1:48:04
Even CIA agents can't beat that.
1:48:06
Is there is there some world where there's like a capital war and I'm paying for a more powerful ad blocker?
1:48:16
Yeah, I mean that sounds good.
1:48:16
Like another question is kind of to say like okay if you think that you either think that current like capital uh accumulation dynamics are going to continue and that uh the rich are going to continue to get richer and if you believe that the question is kind of well how many people are going to survive in the future?
1:48:33
How many people are going to have any modum of independence? Mhm.
1:48:37
Um right like you have some far AI people who think that there's going to be a singleton right think that there's going to be literally one right um you know some people maybe think it's 10 some people a thousand 10,000 uh some people think that all the humans will
1:48:52
get to continue to exist as independent entities are they already independent entities that's a question right I don't know question uh I mean if you were try to put it in like the form of a bet human population above or below 8 billion in 2030 30 above. I think I would just do a normal
1:49:08
I think I would just do a normal trend.
1:49:12
But I do that with what the trend says.
1:49:14
Yeah, just go the trend size.
1:49:14
I don't think there's going to be any discontinuities to any trends really. Well, yeah.
1:49:19
I mean I mean at some point uh but but the question is like how far out do you have to go until you start seeing these effects?
1:49:25
What do you mean by human? Right.
1:49:26
What about someone who lies in bed all day and watches TikTok? Are they human? Yeah, that is odd.
1:49:29
They kind of drop out of society.
1:49:32
I think I think uh question that that popped up for me is is this uh all this debate about AI safety and what should labs be doing, what should labs not be doing.
1:49:42
It feels like your angle is it it should be each individual's responsibility look to look after their own safety in the context of of AI. Is that at all?
1:49:51
I I just I I just like this whole like should shouldn't like what I don't know.
1:49:59
I'm not a sadistic [ __ ] who wants to manipulate other people like the people in power. Like I don't know.
1:50:02
Yeah, but I mean people still look to you as like an example of like uh someone who might have uh answers.
1:50:10
No, I don't have any answers.
1:50:12
Not not necessarily answers.
1:50:15
Just like uh but you can buy my shitcoin here.
1:50:17
Should I sell a shitcoin? Here you go.
1:50:19
Just click this QR code and you can buy a George Hots coin and that will give you answers.
1:50:24
You will find satisfaction and fulfillment in your life after purchasing a George Hots coin.
1:50:32
Is that Is that the end state?
1:50:32
We all have our own coins, I guess. No, no, no.
1:50:36
I I don't mean it like that.
1:50:36
I mean it like I think that a lot of people are like they don't really know what they're looking for and that uh vacuum is is a very uh you know it's very uh dangerous and it's going to be filled by dumb [ __ ] And don't have that vacuum, right?
1:50:52
you got to you got to stand for something, you know, or something. I don't know. Yeah.
1:50:55
I mean, do do you think that there's a chance that someone is able to take a stand and and actually uh bend the arc of of AI progress in the way that uh I mean it happened with nuclear, right?
1:51:08
Like like nuclear development did stall.
1:51:10
There was a stagnation in real world buildout of nuclear capability on the energy side. Yeah.
1:51:19
Yeah, I mean there's a few things about nuclear that make it different.
1:51:20
Uh so nuclear uh even as a weapon is incredibly hard to deploy tactically. Mhm. Right.
1:51:27
So so if a country has has nuclear uh weapons there, aside from like a mutually assured destruction idea, they're not all that useful.
1:51:33
It's not like you can use a nuclear weapon to accomplish tactical objectives.
1:51:36
You know, if you could, I think Russia would have already done it. Yeah. Right. Right.
1:51:40
Russia has some tactical objectives they might want to accomplish, but nukes aren't really going to do it. Right.
1:51:44
I mean, from a pure real politique perspective, not even from a uh like uh oh, like a taboo moral perspective, like what do you want in a radiated pile of rubble?
1:51:54
Like that's what you're going to get.
1:51:56
No, what you want is drones that are hyper specific and can take out exactly who you want, can control areas, right?
1:52:00
So like as a military technology, nukes are not that good. AI is way better.
1:52:05
Yeah, but what about as an energy technology?
1:52:09
It feels like the it feels like the fear like the mimemetic fear of nuclear war and total destruction caused a whole bunch of regulation to pour into a sector and essentially a stalling of nuclear energy buildout.
1:52:20
And if if if the AI doom scenario, whether it's real or not, becomes so mimetically powerful that someone's able to harness that and actually say if you try and build a big data center, we will shoot you. Then maybe it stagnates.
1:52:34
No, I don't really think that's the reason for nuclear.
1:52:36
I think it has more to do with why we can't do other big infrastructure projects in this country, right?
1:52:40
Like it doesn't have to do with the new we also can't build dams, right? Yeah.
1:52:44
And if you look like that's the thing people think that there's some weird taboo around nuclear, right?
1:52:49
But then okay, look at hydroelectric, right?
1:52:51
There's no taboo around hydroelectric, but China leads in installation of both nuclear and hydroelectric and coal and everything.
1:53:00
It's almost like they're correlated, right?
1:53:01
So the thing is not there's a specific fear around nuclear.
1:53:05
It's like, you know, the US decided that they're a developed country and we're not going to develop anymore because we're already developed.
1:53:09
You see the D on the end, right? Like, interesting stuff.
1:53:13
Is that So, so is that just cultural then when you are like the Malaysia sets in, would you expect that to happen to China when they catch up? I don't know.
1:53:21
I I Yeah, I mean, maybe it's just like this normal story arc uh of like uh you know, it's it's I don't know. I don't know.
1:53:33
I I think that like you have a real problem when the kids can't live better than their parents. Yeah.
1:53:38
Um so, but I don't have anything more to speculate on that.
1:53:43
Do do you have more context on on China and specifically in like the AI context?
1:53:48
Um like US electricity looks like this and China electricity looks like this.
1:53:53
Is that all that matters? Pretty much. Yeah.
1:53:54
I mean that's a pretty good proxy for everything, right? Yeah.
1:53:58
Um like there's two things. There's two things.
1:54:01
You know, people are like, "George, how do you feel about the Trump administration?"
1:54:04
I'm looking at two things. Yeah.
1:54:05
With any administration, I'm looking at two things.
1:54:07
Did you decrease government spending and did you increase total electricity production of America?
1:54:13
Those are the only two numbers I care about.
1:54:14
Those will capture everything.
1:54:16
Why does uh why does government spending matter?
1:54:18
We were joking that, you know, Trump must be extremely AGIP if he's running up a massive budget deficit. What the hell is AGI?
1:54:27
I don't know what this is.
1:54:30
uh like in in this never seen it never seen it in this formulation it's that it's that it's numbers yes yes yes but but it but is an extra lever on on labor and capital and it creates more GDP that then can be taxed to pay down the increasing amount of debt super super excel yeah super excel get it what is what is what is super excel do that normal Excel doesn't let's give it up for Excel yes we need that 2 Yes.
1:54:59
The thing is Excel was the the final piece of software and then but in order to add another, you know, hundred trillion dollars to to global GDP, we needed to like kind of rebrand it. And so now we get AGI.
1:55:10
GDP is the the complete it's biggest [ __ ] thing ever, right?
1:55:16
Like I always joke with my friend and I that we're going to start companies and be billionaires.
1:55:18
And I'll tell you how we're going to do it. So, okay. Right.
1:55:20
I start a company, he starts a company.
1:55:22
Uh we both write contracts to each other. Yep. Right.
1:55:26
Like I'll buy something from him for a million dollars and he'll buy something from me for a million dollars.
1:55:29
We'll just do this real fast.
1:55:30
We'll keep passing the money back and forth.
1:55:31
Whoa, look at our revenue.
1:55:33
Wow, that all contributes to GDP.
1:55:35
Wow, we made we're billionaires overnight, right?
1:55:38
Y like and that's my argument is the economy is just that with a lot of extra steps, right?
1:55:45
You can't use services not part of GDP.
1:55:46
This is complete nonsense, right?
1:55:49
You can't you can't have services.
1:55:50
No, like literally literally you take the steel out of the ground, you grow the corn, okay, that's GDP.
1:55:53
But is it I mean if if that GDP is fake, is not the de is the deficit not fake?
1:55:58
Like is is government spending less ow that [ __ ] to people? It's not fake.
1:56:04
But can't you just tax the fake the fake money?
1:56:06
Like if you tax your scenario where you're generating a billion dollars in fake money.
1:56:09
You can't tax the fake money because we're passing the same dollars back and forth the minute you tax it, that falls off so fast. Yeah. Yeah. Yeah.
1:56:17
You can only tax productive work.
1:56:23
Uh, is uh, is AMD doing productive work right now? AMD is doing all right.
1:56:26
Yeah, either Nvidia's really overvalued or AMD is really undervalued.
1:56:30
It has to be one or the other.
1:56:32
How does it all play out?
1:56:34
Like what what does AMD actually need to do to get back on track or realize their potential? Nvidia needs to stumble.
1:56:39
I mean, it worked for AMD and Intel, right?
1:56:42
Like so AMD ended up beating Intel in the entire like no one would buy a data center Intel CPU anymore. Yeah.
1:56:48
And it's just because well, you know, they stumbled and now Intel owns that market. Yeah.
1:56:53
So, you know, AMD just sits there in second place. Okay.
1:56:57
They're pretty they'd be in a better second place than they were a few years ago. Yeah.
1:57:00
And then when Nvidia stumbles, AMD is like, "Oh, hey, we're here."
1:57:04
Is is DGX Leptton like their their cloud offering a potential stumbling block or is it uh or is it the right move for them?
1:57:13
I don't know what that is.
1:57:14
What's an Nvidia cloud [ __ ] Yeah, exactly. Cloud's dumb.
1:57:16
Yeah, exactly. Cloud's dumb. cloud though you can you can break AI down basically into like there's like five five tiers right like at the base level you have like electricity and data centers and land and like things like that tier two are like TSMC ASML Samsung Intel right fabs uh Nvidia AMD open AI
1:57:34
anthropic and then on top you have like completely worthless things like cursor and wind surf um you know these character AI all these people who think oh the app we're going to we're going to get the ARR no that worked in the web it won't work for AI and I can go into why but it's kind of boring I No, keep going. Keep going. Keep going. Keep going. Keep going.
1:57:50
Basically, okay, so like here's the difference between AI and and web.
1:57:52
Um, when you want to run a service like Gmail, one server can serve 10,000 people easily, right?
1:57:59
And there's no demand for like better Gmail, right?
1:58:02
It's not like it's not like I can click and get like, yeah, you can buy Gmail Pro and it'll have a few things, but most people don't really care, right?
1:58:08
There's no limit to the ceiling of how good you want your AI to be, right?
1:58:10
Or how fast you want your AI to be.
1:58:12
Maybe there's a limit to the speed, but like when you're at like a thousand tokens per second, I want the biggest model in the world, right?
1:58:18
Like so there there's very little limit on on that.
1:58:20
Uh but suddenly you can't serve one uh 10,000 users from one server anymore. Mhm. Right.
1:58:26
And the whole dynamics of the web, the whole reason some of the value aggregated to these end players and they still didn't aggregate to the cursor and the wind serves.
1:58:34
They aggregated to the open eye and the entropics. Right. Mhm.
1:58:37
Nobody nobody nobody who built like an email client survived.
1:58:39
They all got eaten up by the the tier fours of the web. Right.
1:58:42
the Googles, the Facebooks, um all of these like app providers, right? Where's a Zinga today?
1:58:46
You know, like this already happened, right?
1:58:50
People just don't where's Zinga?
1:58:50
Oh, Zingga is going to be the next thing, man. Like, no, it's not.
1:58:53
Facebook ate all of that value, right?
1:58:55
Google ate all the value from all the people building on top of Google.
1:58:58
So, the tier fours ate all that value. Yeah.
1:59:00
So, OpenAI, Anthropic will eat all the value from the cursors and the wind surfs of the world.
1:59:06
They'll acquire some of them.
1:59:07
They'll compete with some of them, right?
1:59:08
Same as you saw on the web.
1:59:08
Uh, but I argue that the tier fours aren't even going to have value because the tier fours, this ain't the web.
1:59:15
This ain't where you can have one server serve lots and lots and lots of people.
1:59:19
You know, I'm running 03.
1:59:21
I'm running, you know how much I cost OpenAI every month.
1:59:25
I pay the $200 a month and I cost him a lot more than that.
1:59:28
Codeex, you can now click on Codeex. Yeah. Spin up four nodes. Yeah.
1:59:32
Why would I not click four? It's not my computer. You gave me the button. Hey, I'm just using it.
1:59:39
I'm just using George Hutz single-handedly bankrupts. So, $300 million.
1:59:45
Is there no value in just being the the the front end to AI applications to be like the the the the front door, just the the the default button?
1:59:52
just the the the default button? because we see these we see these these these uh these models kind of go back and forth in terms of benchmarks or what's hot and there isn't as much customer churn as you would expect because people are are just kind of like defaulted into the app
2:00:07
that they installed whenever and so even if Gemini gets better in terms of the actual performance metrics people don't switch from OpenAI to Google because it's so it's so negligible you got to make something 10x better right you got to make something 10x better so like th this whole game is open AI eyes unless they stumble. Sure. Um I'm not switching Sure.
2:00:25
Um I'm not switching to Gemini because it's 20% better and I download some new app and think about a whole new thing, right?
2:00:30
No one's going to switch.
2:00:32
Is there is there a chance for a company to kind of come out with something that's 10x better with an algorithmic improvement or is it just a race for scale?
2:00:40
Like what could actually be that next? It felt like GPT 3.
2:00:42
5 when they really broke through with Da Vinci and then 40 or and then four.
2:00:48
like it felt like this kind of like binary moment when a lot of people realize that this was usable for their daily life, even if it's just a Google search replacement or whatever, write a poem or whatever.
2:01:00
Uh like a a 10x what you're describing like a 10x improvement feels like that kind of like qualitative binary shift.
2:01:08
Is that possible with just scale or is this something that we need a different model for? I don't know. Um I don't know.
2:01:16
I would bet majority still on like these big labs are also attracting the talent.
2:01:22
Um but it is also like it's uh pretty commoditized a lot more so than like Google search, right?
2:01:30
Like you can look at people track how far open source is behind.
2:01:32
It's not that far behind. Y um so no I don't know.
2:01:33
I think this game is mostly going to be uh chat GPTs.
2:01:41
I think Elon's aware of this too. Mhm.
2:01:44
That's why he's trying to go 10x bigger with the data center. Yep. We'll see. Maybe it'll work.
2:01:48
You know, there's there's someone to bet on.
2:01:51
Anthropic I'm not that bullish on, but uh maybe you kind of predicted the uh the pre-training wall.
2:01:59
Uh but that's not a reputation of the bitter lesson and we're going to see similar scale play out in reinforcement learning or is there going to be something else that we're building the big data centers for?
2:02:10
there's something that we don't understand uh in terms of data efficiency. Mhm.
2:02:14
Um, so like when you think of how long it takes a GPT to learn to talk, like how much data it takes, it takes like terabytes of data.
2:02:22
In order to make a GPT talk, like a normal person, it takes terabytes of data. Okay.
2:02:25
Whereas a human trains on megabytes. Yeah. Right.
2:02:28
How is it that if you take all the text that you've ever heard in your life and you you put it to whisper and you you uh you transcribe it, it's going to be a couple megabytes, 10 megabytes, maybe 100 megabytes. Yeah.
2:02:38
So humans have this thousandx data efficiency uh advantage.
2:02:43
And we're going to have to fix that if we want like reinforcement learning to work.
2:02:48
Especially like reinforcement learning that you want to do in the real world.
2:02:50
Humans could do humans can learn from very few samples. Yep.
2:02:55
Um and yeah, I think that like it might be okay if these foundation models train unsupervised on lots and lots of stuff.
2:03:01
But uh yeah, is that a is that something that somebody's working on just like a a new more data efficient algorithm to drop into the pipeline or do we have any like leads there?
2:03:18
do we have any like leads there? because it feels like right now we're going down the path of like reinforcement learning with verifiable rewards and we're going after like individual business use cases that are increasingly long tail and that could be kind of like valuable but it doesn't feel like the breakthrough that
2:03:34
you're talking about like has there ever been a breakthrough right like people think GBTs were a breakthrough no they weren't like you just if you watch the the world it was just it was all just smooth but but but what I will say about AI scaling loss oh man you see like People get excited about AI scaling laws, but here's a pitch that'll kill your excitement immediately. Ready? AI Ready? AI scaling laws.
2:03:55
You can put in exponentially more money to get linear returns. Exactly.
2:04:02
Uh do do you believe that uh the real value is investing in in humanoid robotics then?
2:04:11
Uh have you heard this theory?
2:04:11
So, so it it So, I mean, if you if you put exponential more money into humanoid robotics, assuming that they work and assuming you can uh you can like if you make 10 times as many robots, you get 10 times as much output.
2:04:26
Anyone Anyone who wants a humanoid robot has never worked in a factory in their life. Okay. Right. Bring it down. Anyone wants a human?
2:04:34
Well, yo, yo, it's going to walk around.
2:04:36
Oh, good thing it has legs, right? No. Here's what I want. Yeah. Yeah. We got a laugh track.
2:04:41
Can you show me a robot arm that's capable of putting a screw in something?
2:04:48
Probably put the screw in the thing. Yeah. No, no, no, no.
2:04:51
Not like you carefully jigged up the screw and have a screw dispenser like the way a normal human does it where the screw sitting there in a little bucket on the thing and it picks up one screw and it puts it in. It takes a screwdriver.
2:05:01
No, we're No, no, we're we're we're not close, but also it feels like we're not far.
2:05:05
also it feels like we're not far. It feels like that's what I feel like I feel like humanoids are this interesting sort of like space because a lot of smart people just say like here's the 20 reasons why they won't work and and like
2:05:18
why we shouldn't build them but then so much capital and so many different teams are trying to make them work that they they very well might work for some things and like they just humanity might brute force it because we saw it in a sci-fi movie you know 30 years ago. Why? Why? Why? Why are we cooked on?
2:05:33
This is as dumb as self-driving cars was, right?
2:05:38
And nobody learns their lesson.
2:05:38
And people like Kyle Vote should be ashamed of themselves. Like they really should.
2:05:43
These people who go and raise large amounts of money for another thing that like they should know they should know better. Right.
2:05:49
Here's basically like remember in 2012 when Google said that, you know, my my my 12-year-old daughter would never have to get her driver's license. Yeah. Come on. That's nonsense. Right.
2:06:00
And like now, okay, they shipped Whimo. It's in a few cities.
2:06:03
They're teley opt how how tea operated are they in your opinion?
2:06:11
Is it is it effectively one to one?
2:06:14
It's more than one to one.
2:06:14
There's probably about I would say there's 1. 2 operators per car.
2:06:17
Um but it's not they don't have a steering wheel and pedals.
2:06:22
Yeah, it is it is an autonomous system that they're probably doing some higher level inputs on.
2:06:26
They're definitely like saying when you can be aggressive, when you should slow down.
2:06:30
Uh, you know, whether you can turn to a stop sign or not. Yeah.
2:06:33
Um, you know, again, like here's here's the simple reason to know that it's like that, right?
2:06:38
There's definitely some teop at the Whamos, right? Yeah.
2:06:42
Have you ever seen a picture of that room? No. Yeah. Why not?
2:06:47
I've always thought it was like an ace up their sleeve because like if if there's a lot of pressure on them to say these Whimos aren't safe, they can pull off pull the the sheet off of the ghost and say there's actually a human in the loop.
2:06:58
Don't worry, it's safer than you thought. Yeah. Yeah.
2:07:00
Like you the fact that you've never seen that room tells you that it's way worse than you think it is, right?
2:07:07
Tells you that there's way more teley op than you think it is.
2:07:08
If it was really one person supervising 10 cars, Google would post those pictures all over the place.
2:07:13
You don't see any pictures.
2:07:15
There's so Cruz that actually came out in the lawsuit. I think it was like 1. 5 or 1. 7 humans per car, right? Or or vice versa. Like, right. 1.
2:07:21
5 cars per person, right? No. No. Wait. More people than cars?
2:07:27
Yeah, that's what he's saying.
2:07:29
It's still that it's still that one.
2:07:31
An Uber only requires one person. Yes. Yeah.
2:07:34
But but so so maybe the real innovation is just allowing somebody to get in a car with and not have to talk about the weather or or you know. Exactly. Exact. I'll pay more for that. I'll pay more for that. Yeah.
2:07:44
But I mean, is there any hope that we drive this down and we get to two cars per person, then four cars per person, it starts doubling exponentially and eventually like we are there.
2:07:53
I mean, yeah, like it's obviously going to happen, right?
2:07:56
It's obviously eventually going to happen.
2:07:58
If you want to see where the real state-of-the-art of unsupervised self-driving is today, right?
2:08:02
There's no person with FSD.
2:08:02
When you get your Tesla, that's not tell.
2:08:04
You can go press FSD and that's real AI.
2:08:06
Uh, and well, you can see how good it is, right?
2:08:12
Would I uh take a nap in there even for five minutes? No way in hell. Yeah.
2:08:17
Um you'd be stupid, right?
2:08:17
How are things going on the comm side?
2:08:20
Uh give us the update there. Pretty good.
2:08:22
You know, we're we're we're on track to um we're on track to be two years behind Tesla. So there you go.
2:08:29
So So two years behind Tesla.
2:08:31
Uh but you know, here's why we win, right?
2:08:34
Like because like it's cheap. Uh okay.
2:08:38
So, when you think about self-reing cars, uh it doesn't look anything like the roll out of Uber, right? Or Airbnb.
2:08:46
When you roll out something like that, you're trying to roll out a two-sided marketplace. Mhm.
2:08:51
Uh you got to spend tons of money on customer acquisition costs.
2:08:52
You got to make sure that you've perfectly matched that marketplace right away because if drivers aren't getting rides, they're going to leave the platform.
2:08:58
If riders have to wait too long for drivers, they're gonna leave the platform.
2:09:00
So, it's this careful balancing act.
2:09:02
But once you get this marketplace, you got you got a moat, right?
2:09:05
Switching costs are real high.
2:09:07
Try to get everybody to switch at the same time. It's a chilling point. never do it.
2:09:10
Self-driving cars don't look anything like that.
2:09:13
Self-driving cars look like scooters.
2:09:17
The only thing that it's going to take to roll out big fleets of self-driving cars is capital, right?
2:09:20
It's just strictly a capital market.
2:09:22
You could just I could if I look at a city, I can calculate how many Whimo there are.
2:09:26
If I want to build my own network and deploy that network and run at a lower cost, it's straight up capital.
2:09:31
Easiest thing for investors to calculate, very little risk.
2:09:35
So, self-driving cars are going to be this awesome race to the bottom, right?
2:09:39
It's going to be like scooters where there's going to be like 10 providers of these things for a while and then they're going to consolidate like one's going to do it.
2:09:44
But um yeah, people are really going to win.
2:09:47
What is what's most valuable in terms of developing the next like the next better version of full self-driving?
2:09:55
Is it having a lot of data, building a big data center, having a great team to actually design the system? What's most important? Are they all equal?
2:10:04
Yeah, all those things matter, right?
2:10:07
I think the main thing that matters more than anything else is just time.
2:10:10
Like we're figuring things out with research.
2:10:15
Infrastructure is getting better.
2:10:15
I think that a lot of it's just infrastructure.
2:10:17
My new company's AI infrastructure, right?
2:10:19
Like the infrastructure gets better.
2:10:20
Um my uh my uh coworker has this saying is like uh what we do is that we make the uh hard things easy and the impossible things hard. Mhm.
2:10:34
And that's like the goal of infrastructure, right?
2:10:36
You build infrastructure, your infrastructure gets better, and then what was what you couldn't even dream of doing 10 years ago is now one command today.
2:10:41
And today, you know what what you you you uh Yeah.
2:10:48
What's the current use case for most people with tiny boxes?
2:10:55
Is that that's by design, right?
2:10:55
You're not supposed to know.
2:10:56
But I mean, so like I sell the computer, it has specs, right?
2:11:00
Like so many people want to tell you and I hate this. I hate this.
2:11:02
They're telling you like how the product is going to impact your life or what you can use the product for. Oh my god, who cares? Yeah, here's what it is.
2:11:08
I'm going to tell you what it is. That's your job, right?
2:11:13
I'm not an advertiser, but I mean I mean our intern wants to build something with a tiny box.
2:11:19
I want to give him some ideas. Go buy one. I don't know.
2:11:23
Why do you want to build stuff with a tiny box? I mean, is it good?
2:11:26
Yeah, it's just a It's a bunch of GPUs in a box, you know? It's no box. GPUs in a box. Weight to it. What? Um, do it.
2:11:32
What robotic form factors are you most bullish on? We've touched humanoids.
2:11:37
You gave a a great review there.
2:11:39
We've touched autonomous vehicles.
2:11:42
Sounds like generally bullish, but Capital Wars, Race to the Bottom, uh, all that stuff.
2:11:47
Are are there any other kind of form factors that you're thinking about that you are generally optimistic or excited about? ARM. Arm. The arm. Maybe two arm, right? Just two arm. Right.
2:11:59
Cuz I look, I run a factory.
2:12:01
I run a factory in San Diego.
2:12:01
We make all the commas right here.
2:12:03
And I can't wait to get a whole lot of robots in there.
2:12:06
But uh I don't eat humanoids.
2:12:06
I'm just going to stick two arms to the table.
2:12:10
And then it's going to grab a comma.
2:12:12
It's going to put the screen on the front.
2:12:14
It's going to flip it over.
2:12:15
It's going to put the four screws in it.
2:12:17
And then it's going to pass it on. Yep. Right.
2:12:19
Show me anything that's anywhere near that level today. Yeah.
2:12:21
What What uh what would you do if you were trying to build like a truly multi-purpose robotic arm?
2:12:28
The arm's already good enough.
2:12:30
It's the AR offtheshelf arms are fine. It's all software.
2:12:32
Again, it's always all software.
2:12:34
Autonomous vehicles are all software.
2:12:35
Robotics is all software.
2:12:37
But everybody loves to bike shed. Yeah.
2:12:39
What color are we going to paint the humanoids, you know, like like let's have a great conversation about that?
2:12:46
Well, you know, we don't want to paint them red because that might scare people in a terminats. Yeah.
2:12:54
Would you uh would you trade in your legs for wheels if you could?
2:12:56
I got that my legs for wheels.
2:12:59
Yeah, this is a question from Aaron Frank, friend of the show.
2:13:04
He's asking this in real time. You're wheel guy then.
2:13:06
Like people had asked me if the wheels are like making a statement.
2:13:09
I just don't want to have to have this conversation.
2:13:11
What what what about the sim tore gap in robotics?
2:13:14
Like how how is how is simulated data?
2:13:16
You know, you you build a bunch of data in Unreal Engine, then you try and transfer learn it back.
2:13:23
Uh obviously there's been a bunch of experiments of that with self-driving cars.
2:13:27
Is that a path that we should be going down for the for the robotic arm development? Yeah.
2:13:31
So I think with a lot of sim to real stuff, the reason people are excited about it is because of that data efficiency gap weap, right?
2:13:40
Like current machine learning algorithms like a thousandx less data efficient than humans.
2:13:43
Uh so yeah, you're going to need a thousandx more data, right?
2:13:48
If a human can learn something in one exampler, 10 examples, the computer is going to need a,000 or 10,000.
2:13:54
Now, do you really want to reset the stupid state of the physical world? 10,000 times.
2:13:57
You might do it 10, but you're not going to do it 10,000, right?
2:14:01
So that's where you want a simulator where you can just click reset and everything's back to exactly how it was.
2:14:05
Um, so I think this stuff's going to play a role, but I think more fundamentally that data efficiency gap has to be understood.
2:14:14
Uh, we talked a little bit about coding agents.
2:14:16
We talked about how you're bankrupting OpenAI by spinning up a lot of different uh, codec agents.
2:14:20
Um what uh what other sort of agentic software are you excited about?
2:14:28
Do you expect to uh you know what's a gentic mean? Yeah, basically bots.
2:14:34
It's it's like what we're calling bots now.
2:14:36
Um what's but but but anyways like I just you know from Star Trek. Yeah. Yeah. Maybe.
2:14:42
No, but but I think about a world in the future, you know.
2:14:46
Do you do you expect to be I don't know if you're a a Slack guy, an iMessage guy, Discord, maybe no messaging at all, just you know um you know telepathy telepathy uh but but do you expect a world in the future where you're just
2:14:59
you know a perfect interaction between you know human employees and agents or is it you know going to be more like you know you'll do the odd deep research or maybe you send some automated outbound emails or have some codeex bots running I don't even like all of us. When do you
2:15:15
When do you think you'll be able to book a flight just by saying I'm trying to get to New York tomorrow?
2:15:20
Oh, see the worst part about this is like like and that's going to come pretty soon actually. Okay. Right.
2:15:26
We're going to pretty soon have computer use models that are actually capable of going to delta.
2:15:29
com and booking a flight. Yeah.
2:15:31
But then what's actually going to happen is Delta's going to partner with whatever company does that and they're going to put it behind the stupid uh thing and like Yeah. So that's going to Yeah.
2:15:39
That's going to be here in a few years, right?
2:15:42
Not with agentic [ __ ] but just with normal hooking the APIs together, right? Yeah.
2:15:46
Wait, so yeah, what is the bullish on APIs?
2:15:49
What is the mistake about like the agentic buzzword?
2:15:51
Like what what are people like even describing?
2:15:53
Again, it's another thing that I I really have no idea what it means.
2:15:57
You know, I I was hanging out with some friends last night and like like uh my friend works in this VR company and the you know, the the CEO is really interested in things being open source, but he's also really interested in making sure that things are protecting our intellectual property and proprietary.
2:16:11
And the truth is he has no idea what the word open source means.
2:16:17
He has no idea what it means that they can copy his [ __ ] right?
2:16:18
Like that someone else could use it.
2:16:20
He just he just heard the word open source in some like buzzword thing and he's like, "Do we have the open source?
2:16:25
Do we have the open source in the thing? Okay, so check the box.
2:16:30
Check the open source box.
2:16:32
Let's protect our last question about the a last question about the agentic buzzword.
2:16:37
I think that there is something that people are picking up on which is that these models seem to be very smart for short amount of time, but if you run them for a long time, they start hallucinating and kind of going off the rails.
2:16:48
And so you you have like 10-minute AGI.
2:16:50
feels incredible, but as you let it run and do more work, you can't just say, "Hey, go do a week's worth of work, come back to me when you're" But it's superhuman in one minute.
2:17:00
And so, is that kind of trade-off curve real?
2:17:02
And then is it just a matter of like better harnessing to actually get to two hours of work, which is kind of what the agentic people are like advocating for? No.
2:17:11
So, I don't think it's better hard, but this is definitely a real phenomenon.
2:17:14
This is definitely a real phenomenon.
2:17:16
Uh, you can experience this.
2:17:18
There's papers exploring it which show that if in 10 seconds there's absolutely no way I'll come even close to a modern totally um because the first shot from the LLM is great. Yeah.
2:17:31
And then it kind of degrades and it degrades pretty quickly whereas humans look a lot more like this.
2:17:38
Humans can stay coherent internally for much longer. Mhm.
2:17:40
Um so yeah I I think that that's a real thing.
2:17:48
I think that that's mostly going to be fixed by like long context, just more energy, long context, RL.
2:17:54
Yeah, just like you just got to do it.
2:17:57
We'll figure out new ways to make the context better.
2:18:02
We'll combine diffusion and uh and auto reggression in some clever ways.
2:18:05
Yeah, I think that this is just going to be a like there's not going to be a breakthrough here.
2:18:11
There's not like one magical thing that we're missing.
2:18:13
Yeah, I think it will be a continued plot.
2:18:14
The same thing with data efficiency.
2:18:16
I think people will start to care about it.
2:18:18
Some new tricks will come out.
2:18:20
Some of them will work, some of them won't work.
2:18:22
We'll continue to do graduate student descent until we find a anything that's last question for me.
2:18:31
Anything that you're particularly optimistic about? Anything?
2:18:33
You check the timeline and you think, "This is awesome. I love this. I love this.
2:18:36
I want to see more of this."
2:18:39
A little maybe a little white pill to kind of cap it off. Yeah.
2:18:42
So, here's something I'm optimistic about.
2:18:44
That fact that the one server can't run 10,000 users, that is most of the reason that the modern internet that that is one of the reasons that the modern internet sucks.
2:18:54
That that that so much of the stuff is in non-recurring expense and then it becomes really really hard to compete with these people, right?
2:19:03
Like you could run Twitter on one computer. Yeah. Right.
2:19:08
And 20 people could do it too.
2:19:10
But like they don't because again these companies have moes and they invest in making sure that their moes can't be broken.
2:19:16
Um with AI I think there's going to be a much less of a mo especially when you look at the move from auto reggression to diffusion.
2:19:23
So auto reggression can run in large batch sizes.
2:19:26
When you run chat GBT you're running with a whole bunch of other people on that same computer. Yeah it's only 100.
2:19:31
It's not 10,000 but still it's 100.
2:19:34
Diffusion is running the cloud at batch size one.
2:19:36
And once you're in batch size one land, running it locally starts to make sense.
2:19:43
Actually running the models locally or at least having your own computer in the cloud. Yeah.
2:19:47
Not being some shared resource that's really controlled by some else.
2:19:56
Um, so yeah, this was never a thing because you can't put lots of people on a GPU. That makes sense.
2:20:00
They tried some weird stuff with the licensing, but yeah. Fantastic.
2:20:04
Well, thank you so much for stopping by.
2:20:06
This is a great conversation.
2:20:08
Yeah, I wish we had a full hour. This is great.
2:20:09
We'll talk to you soon, Josh. George. Cool. Bye. See you later. Cheers. Bye. [Music]