0:00
You're watching TVBN.
Search this video
0:04
>> Sam Alman and Matthew McConnA defined the weekend in my world on my timeline.
0:10
Uh, two former Joe Rogan uh, guests.
0:14
Interestingly, only Sam Alman has done Theo Vaughn and only Sam Alman has done Tucker Carlson.
0:20
Uh, Matthew McConna has gotten snubbed by those two.
0:22
Um, but hopefully we'll get we'll get McConna on Theo eventually, but until then, uh, Sam is the loan of the pair to do >> MC on the It feels like a match made in heaven.
0:39
>> Um, Sam obviously did well on the show and uh, but they both were talking about AI and um, caused, you know, a stir, predictions, debate, and I thought it would be interesting to weave through it.
0:52
So, uh, Sam Olman kicked off with a post here. He says, "Time is money. Save both.
0:57
Easy to use corporate cards, bill payment, and a whole lot more all in place." He said, "Go to ramp. com." No, he didn't.
1:02
That's fake news, but we love ramp. com.
1:03
And so, head over to ramp. com.
1:08
>> No, what Sam actually said was, "Over the next few weeks, we are launching some new compute inensive offerings." There we go.
1:15
>> Because of the associated cost, some features will initially be only be available to pro subscribers. That's me. That's me.
1:22
Uh, and some new products will have additional fees. Oo, interesting. Extra monetization.
1:26
Our intention remains to drive the cost of intelligence down as aggressively as we can and make our services widely available.
1:33
And we are confident we will get there over time, but we also want to learn what's possible when we throw a lot of compute at today's model costs at an interesting at interesting new ideas.
1:44
So, there were lots of debate over what he's going to launch.
1:47
Meanwhile, Matthew McConnA was over on Joe Rogan saying that he wants a private LLM fed only with his books, notes, journals, and aspirations so he can ask it questions and get answers based solely on that information without any outside influence.
2:02
Can we play the Matthew McConnA clip?
2:04
Because he had a very interesting phrase.
2:07
I thought it was very funny.
2:08
He doesn't say he doesn't say fine tune on this.
2:14
>> He wants to log >> he wants to log it in. We'll hear it from him. Let's go to Matthew.
2:16
I'm >> interested though in a private LLM >> where I can >> upload, hey, here's three books I've written.
2:27
Here's my other favorite books.
2:27
Here's my favorite articles I've been cutting and pasting over the 10 years and log all that in.
2:34
>> Log all that in >> and here's all my journals, whatever the people and log all that so I can ask questions based on that, >> right?
2:41
>> And basically learn more about >> myself, right?
2:42
I stand on the political spectrum, right? blah blah blah.
2:44
I'd like to >> No, that's that's what I'm would like to do, which is sort of a glorified word document, >> but it still would hold a lot more information and just Oh, can you do asking >> this is this is an important consumer.
3:01
>> This is the next market of consumers with the information I'd like to load it with. >> Right. >> Yeah.
3:07
Maybe even like I'm saying in this in the words of belief in in the in the man I'm working to be, the man.
3:13
>> I feel like the current state of LLMs are meeting his needs and that's really really important.
3:18
If you're open AI, you want to become the biggest company in the world.
3:21
You want everyone to use your products.
3:23
>> It's slowly learning about me through conversations then going, "Oh, I think this is what you like based on a conversation."
3:28
No, I want the answers based on what I've uploaded it with only, not from the outside world.
3:31
And he probably uses he probably uses Instagram and has has and Instagram is personalized does give him a a unique view on things he likes and has become very good at at servicing exactly what he wants.
3:46
We got Nvidia investing $und00 million in open AI.
3:50
The money flows right back to Nvidia in the form of chip purchases.
3:55
Basically over a couple years we're still in LOI territory. >> We love this.
3:59
We've been advocating for more handwashing generally for basically an entire year.
4:04
Uh we love I mean one hand washes the other.
4:07
People say it like it's a negative thing but how are you going to wash both hands if not with the other hand like true >> what how do you you can't wash a hand with one hand.
4:16
You have to use the other hand and that's the nature of the partnership between Nvidia and OpenAI.
4:22
So lots of people saying you know uh this time oh is this time different?
4:25
Uh we're doing a little bull and bear here.
4:29
I think we might disagree on this take.
4:31
I wrote I wrote a take in our newsletter. Head over to tbpn. substack. com.
4:35
You can just go to tbnpn just tbpn. com and go sign up.
4:37
Um we've been growing that having fun writing up a little uh a little breakdown >> morning love letter to technology to technology and business.
4:46
Just just focusing on the top story of the day. >> It's hard.
4:49
Nothing is harder for you on earth than being bearish on technology. >> You can't catch me.
4:55
I I will never be bearish. >> Never.
4:58
I will always be bullish.
4:58
Um, no, obviously there are risks to this.
5:01
Uh, it is a little bit of an ora boros. It is a one hand wash.
5:07
>> There's not that many places you can get 100 billion from if you need 100 billion.
5:10
>> That was exactly how I kicked off my intro was someone was going to pay for it. We've been seeing news.
5:15
The one thing that all the AI labs agree on is that scale is all you need.
5:20
Scaling laws will hold a lesson.
5:23
Pretty much all the foundation labs are like let's do bigger training runs, let's do bigger reinforcement learning runs.
5:31
>> Elon focusing on gigawatt data center >> Colossus 2 massive data center.
5:34
Um and and Google you know Google is Demis is being a little bit more cautious with his language but they're investing tens of billions of dollars in capex.
5:45
Like they're definitely building huge huge clusters uh producing a ton of TPUs.
5:48
And so I wanted to revisit a post that we really enjoyed back in before you go in there. >> Ramp.
5:56
com, easy to use corporate cards, bill payment, accounting, and a whole lot more all in one place.
6:01
Sorry, what were you about to say?
6:05
>> What's the second most important thing to say? >> Ram, baby, let's go.
6:09
>> Uh, Buco Capital highlighted something this morning that I thought was relevant.
6:13
Uh, he said Gavin Baker was was uh covering this.
6:16
He said, Larry Pageige has said internally at Google many times, I'm willing to go bankrupt rather than lose this race.
6:24
Everybody is focused on ROI, but the people making the decisions are not basically saying that uh scale, it's all about scale.
6:31
Uh you just they're going full send, right?
6:37
They're not thinking they're they're optimizing for winning, not necessarily, you know, uh in a spreadsheet optimizing purely for uh immediate ROI. Yeah. Yeah.
6:48
I mean, if you're in the position to generate a lot of cash flow, it is kind of the per the game theoretic optimal thing to overinvest in a new huge technology.
6:57
You have to uh because if you don't invest and the technology wave happens, you look really stupid. You you lose your job.
7:03
If the technology wave happens and you invest, then you're good.
7:07
And if the technology wave happens, uh if the technology wave doesn't happen, but you didn't invest, you don't really get as much credit for that because then you're just kind of stable where you are.
7:16
Uh and so it's this it's this uncapped up opportunity with a lot of downside if you don't invest.
7:22
So people are pushing it. What you laughing at?
7:24
>> Canar says John is peaking at the peak of inflated expectations.
7:29
>> I'm at the peak of inflated.
7:29
I I live at >> You love a roller coaster.
7:32
You love roller coasters. >> I love the peak. Uh oh.
7:34
Uh, I have a question for Tyler.
7:39
Do you know what a phone book is? >> Yes. >> What is a phone book?
7:44
>> It's like a big book that has a lot of phone numbers. Okay. Name every phone book. >> Did you Wait, no. No, no.
7:47
Did you know that there's someone on our team, I don't want to dox him, who did not know what a phone book was.
7:51
We asked him to go out and get some phone books and he just didn't know. >> No way. >> This is real. >> I'm not surprised.
7:56
I know who you're talking about, I think.
7:58
And I'm not that surprised. Let's call him out. Let's call him out. Dylan. Dylan.
8:00
He's our head of logistics.
8:04
>> Our head of logistics. >> You know what? He's He's 19.
8:05
No idea what >> he had to Google. What is a phone book? >> Googling.
8:12
No, he probably asked as asked an LLM.
8:15
>> Yeah, >> he had to go over to Gemini.
8:15
You had to go over to Chad GVT. What is a phone book? >> Hopefully.
8:20
>> Give me a deep research report.
8:22
>> I just thought I thought it was hilarious and I wanted to I want to see that.
8:24
Anyway, uh Reream, one live stream, 30 plus destinations, multiream, reach your audience.
8:28
And I want to play a clip that should have been on Reream. It was on CNBC.
8:32
Hopefully they're using Reream.
8:34
If they are, they should move over.
8:35
Uh, I want to play the clip from CNBC of uh, Sam Alman, Greg Brockman, Jensen Wong talking about this deal, their plan. Let's play the clip.
8:44
Uh, I am here at Nvidia headquarters in Santa Clara uh, with the CEO of the world's most valuable company and the CEO and president of the world's most valuable uh, private company, Jensen Vaughn of Nvidia, Sam Alman, and Greg Brockman of Open Air.
9:02
So, let's dive right in to the news.
9:05
Jensen, Nvidia is making a hundred billion dollar investment in open AI and working together to build out I think you're saying 10 gawatt of capacity uh over several years.
9:15
The investment is going to come with the gigawatts one at a time.
9:20
You're telling me you guys as quickly as you guys can get for a second.
9:24
Imagine imagine a VC is like, well, we're going to give you 100 million, but it's it's going to be in trenches with the gigawatt.
9:30
Oh, oh, would that be so crazy?
9:33
>> Do you know the original Sequoia YouTube memo?
9:36
>> It's a trunch investment. It was trunch. It was trunch.
9:38
>> Greatunch investment of all time. >> It is. It is.
9:40
Uh and and and structured rounds are coming back.
9:43
Uh tranched investments have made a little bit of a of a resurgence.
9:47
Uh we call them structured rounds now.
9:49
Uh this this happened during the uh the crossover era.
9:54
Tiger, CO2, those firms were sort of famous for doing deals that would give you a very high headline valuation, but then would allow the firm to kind of average into that valuation based on the time value of money.
10:06
And so uh tunching is uh maybe underrated.
10:08
You got to know how to tanch anyway.
10:11
Let's keep >> just don't do it at don't do it at preede.
10:14
>> Don't do it at preede.
10:14
And also be very careful.
10:16
Make sure there's demand.
10:16
But we'll find out what >> this is the biggest AI infrastructure project in history.
10:21
This is the largest computing project in history. >> Hit that.
10:25
>> Well, the reason for that is because computing demand is going through the roof for open AI.
10:30
>> You know, Chad GPT is a the the single most revolutionary AI project in history. >> Sure.
10:37
>> It's um being used everywhere, every industry, every country.
10:39
Um every person practically that I know uses chat GPT.
10:44
The computing demand is going through the roof.
10:46
And so this partnership is about building an AI infrastructure that enables AI to go from the labs into the world.
10:56
This is about the AI industrial revolution arriving. It's a very big deal.
11:02
>> Uh hundred billion dollars, a lot of money.
11:04
Sam, Greg, you guys are used to dealing with a lot of money and big projects.
11:09
So Sam, I think it was just 8 months and a day ago the initial start date announcement talking about the overall move that OpenAI is making in building out this capacity. Where does this fit?
11:21
>> So as as Jensen said, building this infrastructure is critical to everything we want to do.
11:25
Without without doing this, we cannot deliver the services people want.
11:28
We can't keep making better models.
11:30
And now that we really see what's on the near-term horizon of how good the models are getting, the new use cases that are being enabled, what people want to do, this is like the fuel that we need to drive improvement um to drive uh drive better models to drive revenue everything.
11:44
So this is this is helping us get to a world along with our partners at Stargate, Microsoft, Oracle where we can build out increasing amounts of infrastructure to deliver on what the world is demanding out of these services.
11:57
Uh there's like no partner but Nvidia that could do this at this kind of scale, at this kind of speed.
12:00
It's it's really like quite quite incredible.
12:04
Um but this will this will expand on the Stargate ambitions and let us push further and further.
12:08
We have found every step along the way that we did not quite set our sights big enough given the market demand.
12:13
So this will help us push towards that next level.
12:15
The compute constraints that the whole industry has been in our company in particular have been terrible.
12:20
Um we're so limited right now and the services we can offer.
12:24
There's so much more demand than what we can do.
12:25
And as we look forward another year or two years, if you have, you know, let's say it takes 10 gawatts of compute or 5 gigawatts of compute, you could choose one of two things.
12:33
You could choose to cure cancer by doing a bunch of having AI do a bunch of research or you could choose to offer free education to everybody on Earth.
12:42
>> This wants to make that choice.
12:42
Sam is saying increasingly make me choose between curing cancer and free education for the world.
12:49
You don't want to make me choose between those two things.
12:52
>> Both you want both of them. Don't make me choose. Don't make me.
12:54
It would be tough to get >> simplification.
12:56
It's a simplification, but I think that it's a it's reasonably justified. I don't know. >> It's good.
13:02
It's definitely good framing. >> Yeah.
13:04
I mean, it puts it in very concrete terms as opposed to just being like, look, like the value of ChachiPT, we can just pause the video.
13:11
Um the the the the value of CHP will be a lot of people, you know, searching for things, getting like 1% little gains all over the place, and that will add up to growth overall.
13:22
um that's that's a lot more amorphous than something as concrete as like the they will cure cancer and educate people.
13:29
It's like so much more tractable and you have to put it in that uh in those terms if you're going to be >> Mr.
13:35
Moral in the chat says let's give it up for a lot of money.
13:38
>> Let's give it up for a lot of money. You love to see it.
13:40
Um, anyway, I was reminded about that famous image we pulled up of Masayoshi Son back in February was meeting with Sam Alman in Japan and he had a crystal ball.
13:51
He dropped the crystal ball >> but he picked it back up. It didn't break.
13:56
>> He had a crystal ball and he was using the crystal ball as a metaphor for his ability to see the future, see that AI is coming, AGI is coming and that it was worth investing uh very aggressively.
14:08
And at the time it was it was it was met with uh the idea of like oh like Chad like OpenAI has to tap Masa and Soft Bank like maybe that's top signal because we work was so fresh in everyone's mind everyone was thinking about oh the last time Masa came out and was doing a big tour in Silicon Valley with >> Yeah.
14:28
There's a whole generation of of people that just know him for that investment. >> Exactly.
14:32
And they don't know him for >> Yahoo.
14:34
Yahoo, Alibaba, and then also ARM.
14:37
And so he's actually made hundred billion twice.
14:40
>> They didn't know him investing in Nvidia early and selling out of the entire position. >> That was rough.
14:43
I mean, that's the nature of MASA.
14:45
They like he he's he takes huge swings.
14:47
Sometimes it pays off, sometimes it doesn't.
14:49
Uh but overall, he's been able to stay in the game and continues to write really big checks.
14:54
And so, um I think uh this time is different.
14:57
I do think AI is a different story than we work.
14:59
I think Open AI is is way different than we work.
15:04
We work was a financial innovation on top of a technology, housing or office space or building that's literally thousands of years old.
15:11
Uh it was a reconfiguration of of the asset from we went from you know you you build the you build the office with sticks and you build the building with sticks and stones to you know you build it and then someone owns it and you mortgage it and then there's financialization there to you rent it on a monthly basis or an annual basis to you rent it on an hourly basis.
15:32
So that type of financial innovation can provide some value.
15:33
Regus is a good business.
15:36
I think it's still around.
15:37
But the basics basics of this deal are for every $35 billion of GPUs that OpenAI buys.
15:46
>> Nvidia will invest 10 billion.
15:48
>> So they're effectively just like paying for for GPUs with like 30%ish equity.
15:55
>> It's like higher than their margin. Yeah. >> Yeah.
15:57
>> Their margin is like 50 60%. >> Yeah.
15:59
From from Nvidia's >> point of view, I think that I mean it makes sense for both parties, right?
16:04
Openi is also going to be able to >> basically take those investments at presumably higher and higher valuations if they can continue to >> grow.
16:13
Uh so this is not just like a hundred on 500 billion.
16:15
This is not like a a traditional financing by any means.
16:20
>> So the question that like I I I think people are you know throwing up the red flag because of the circularity of the deal.
16:27
you can start waving the red flag around if you want.
16:29
Um because more commonly what you would see is a company like Nvidia that's throwing off a ton of cash would send the money to the investors in the form of a dividend.
16:39
And so the money would flow from the money would flow from Nvidia to their biggest shareholders.
16:48
Vanguard owns 9% of Nvidia.
16:50
Black Rockck owns 6% of Nvidia.
16:52
Fidelity owns 4% of Nvidia. State Street owns 4%.
16:55
And so the money could flow from Nvidia.
16:58
Nvidia has like 60 billion in cash and the money could flow from them to their investors and then then their investors could choose to invest in OpenAI, give them the cash and then OpenAI can buy Nvidia GPUs.
17:07
Um but they're doing it more directly and so people are more worried because it calls back the stories of the dotcom boom.
17:15
Um everyone in the dot boom if they dig into it they they usually point to uh the Lucent WinStar deal.
17:23
This ended in WinstStar's bankruptcy.
17:25
Lucent extended $2 billion in vendor financing to WinStar to build >> Tyler.
17:30
Do you know what a bankruptcy is?
17:32
Not many of these have happened in uh in in the world of tech in the last few years.
17:38
>> You were around for SVB. >> Yeah. Yeah, I remember that. >> Yeah. >> Okay, that's good. Okay.
17:41
Um >> the new generation's all right.
17:44
>> But uh so so WinStar was saying, "We're going to build out a wireless network."
17:48
Lucen said, "Hey, we have a bunch of technology that will allow you to do that.
17:51
You have to buy stuff from us.
17:52
You'll build the network.
17:52
you'll do it on our hardware and we will extend $2 billion in vendor financing.
17:57
We'll give you $2 billion.
17:58
You buy our stuff, right? And it didn't work.
18:01
Uh this was announced in October of 1998 and uh and Lucent went uh uh Winstar went bankrupt and Lucent had to pay a $244 million fine for uh the do mess that ensued.
18:11
But that's not the only time this sort of circular deal has happened and it hasn't always ended in tears.
18:20
So there's a different example from 2012 when ASML, which makes the lithography machines that go sit at TSMC and actually make the chips that Nvidia designs, ASML, the Dutch company, they did what's called a customer co-investment program uh where Intel, TSMC, and Samsung pitched in $6.
18:40
8 8 billion across R&D uh funding and equity purchases in the company in order to help pull uh in order to help ASML. This is 2012. >> 2012.
18:54
>> You don't remember ASML going bankrupt soon after because they didn't. They're still dominant. They've done very well.
18:59
And so what happened was that uh the Intel, TSMC, and Samsung said, "Hey ASML, we want to do the next version of lithography.
19:08
We want to make more advanced chips." What will that take?
19:12
Well, we have to build two new technologies that are going to be extremely expensive.
19:15
One is called extreme ultraviolet lithography, EUV.
19:20
This is what has wound up driving the current generation of of uh of semiconductors.
19:25
It's an incredibly complex uh machine that has a a mirror that's ultra flat and all these crazy lenses and if a single dust particle gets in there, it destroys everything.
19:35
Like, it is the most advanced, most precise machine that humans have ever created.
19:39
And it was an immense amount of work and it required a lot of money.
19:41
And and then they also wanted to scale up to I think a 450 mm wafer that they would etch all the chips out then you cut the chips out and then you get everything.
19:51
So um they wanted uh so the customers said hey we want the next version of your technology and you need to raise money so we we'll we'll fund it. We'd be happy to.
20:00
And so what happened was they wound up uh giving them some money, buying some equity, and it all worked out like it's it felt like the same odd uh ASML was trading at around $70 a share. It's now at 961.
20:17
>> So hopefully they help. >> Yeah.
20:18
We have our next guest, Katherine Bole, in the waiting room.
20:21
We'll bring her It's been too long, Katherine. How you doing? >> It's been too long. Welcome back.
20:26
Thank you so much for >> way too long.
20:29
Thanks so much for having me.
20:31
>> You have been extremely busy. Give us the update.
20:33
What's new in your world?
20:35
>> Gosh, what's new on on our world?
20:35
Well, the the big news is we we launched dynamic tech defense reform, which I know we're going to get into today.
20:41
Um, which is a huge initiative on behalf of all of defense tech, all of the defense tech community.
20:47
tech community. Um, but also it it's I just returned from Washington last week, had a host of conversations across the Pentagon, across Congress, and it feels like for the first time in probably 10 years since I've been doing this, that everyone is aligned, that we need to see major major changes happen to defense
21:04
procurement, and not only is that sort of things that people talk about, but it feels like it's finally going to happen in this year's NDAA, um, and that we're on the one yard line, um, sprinting to a finish that I think will will hopefully, um, end up great for for all of the the companies in the defense tech community. So, we can go further into what's
21:21
So, we can go further into what's changing, but I always like I I I feel like I'm bringing like actual good policy news like probably like >> that's like the one thing that everyone in Washington can agree on, Republican or Democrat, uh you know, House or Senate, Pentagon, everyone is aligned that we're not every lobbyist agrees.
21:39
>> Not everyone lo is remarkably complex.
21:39
>> Not everyone lo is remarkably complex. I remember learning about uh program of record and then SBIRS and then we had Will Herurd on the show last week and he was telling us that he had an SBIR but then also some some funds allocated for in a bill and it's just like there's
21:58
seven different pools of capital to pull for probably more than seven probably 200 uh and so some streamlining of that makes a lot of sense what does this actually mean for defense tech startups scaleups growth stage companies I've heard about the valley of death getting from the SBI to the program of record. Is anything are you optimistic about
22:14
Is anything are you optimistic about anything changing in that narrative or is it just easier to get to program record?
22:19
Like how should startups be thinking about like the goal that they're like working towards? >> Yes.
22:24
So I think there's there's a lot of changes happening but but the the things that we're most excited about are like very specific changes that are going to help all companies whether or not you are an early stage company with five employees wanting to scale working with the DoD or you are an anderal scale company that has you know that is no longer a startup with with thousands of employees um that still has these problems inside the Pentagon.
22:48
the the real issue with the Pentagon is that they still view tech in many cases as as small, right?
22:54
Like they view us as like small, we are not the primes.
22:56
Um, and what I think has changed in the last few years and particularly with this administration, with this Pentagon, and with this Congress, is they're realizing that we cannot achieve the aims of supporting our men and women in uniform.
23:09
We cannot win the next battle if we don't have emerging technology and if we don't have it produced at scale.
23:14
And so there's a lot of sort of like fine print that still exists in a lot of these contracts, whether it's a program of record, whether it's an OTAA or or whether it's, you know, different types of contracting vehicles, but by changing some specific language in this year's NDAA, it's going to it's going to open the the floodgates for competition in really good ways.
23:33
So, I'll give you a couple examples.
23:34
Um, and this has been something that, you know, our our friends at Andural and and our friends at Seronic and others have talked about for years, which is the the past performance requirements, um, which are are often written into a specific contract from the department of war.
23:47
So, if you are a company and you are brand new, um, and and in the eyes of the department of war that has been working with companies for, you know, decades, a company like Anderl is seen as a new company.
23:59
Um, there is preference given to a company that has been around for 30 or 40 years on contracts often because they have what's known as past performance.
24:08
Even if it's not in something like AI or something in the >> even if it wasn't a that great of a performance, right? >> Yeah.
24:15
Even it was terrible performance, they have they have some record of past performance um that that gives them a leg up in the in you know in in the competition for that contract.
24:22
And so one of the things that the that the forge bill and the Senate, we just we just wrote an op-ed with Senator Wicker today that came out in the Washington Post talking about how important it is that the forge bill and that really the the House Armed Services Committee and the Senate Armed Services Committee are aligned on.
24:37
Removing the past performance requirement from these larger contracts is going to level the playing field where then you're going to have companies that have been around for a long time, if they have the best product, they can prove they have the best product.
24:48
best product. And then the companies that are just new on the stage in the last few years if they can perform at a higher level than the companies that have been around forever, like they should get that contract despite not having the past performance as written
24:59
into the requirements >> because historically a big a big prime would get a contract just because they can, but then it ends up being like kind of a rounding error for them and so they don't put that much re it's not like a flagship program and so they can underperform for years on it. Uh
25:12
Uh meanwhile, a startup might die even if they would have made it their main thing, right? >> Yeah. Yeah.
25:20
Or or it's just like they even if they have an open competition, they're they're they're because they have past performance, they're they're seen as as a more reputable company.
25:26
Um and so I think even just having that cultural shift, but having that language removed from the contracts is going to give a leg up to to all startups.
25:34
I mean, there's other things in it that I think are really important around changing the culture um inside of the Department of War.
25:41
Like right now if you are there are extraordinary people in in the Pentagon who stick their neck out trying to do this day in day out and they're not being celebrated.
25:49
They're not being rewarded for for caring about the war fighter.
25:52
And so there's language in forged that that changes the incentives around if you're a procurement officer and you procure the next greatest weapon system and it comes from a startup like you're going to you should be celebrated for that.
26:03
And that was something we heard a lot too when when I was in town last week talking, you know, across Senate and across the Pentagon, like they want to know who the heroes are.
26:11
And so, like, we're going to start being really loud about who these heroes are inside the Pentagon that are that have been, you know, changing the culture that have really been been pushing things um to get it to to where it is today where startups can work with work with the Pentagon in a much easier fashion than they could even five years ago.
26:27
Do you have any thoughts on uh there's this book by Dan Wang um break neck all about how chu America has the lawyerly society and we have maybe too many lawyers working in government?
26:38
There's been obviously a push at Doge to bring in more computer scientists.
26:40
We talked to Will Herd, he's also a computer scientist by training.
26:43
Uh has there been any shift in DC that you've noticed where I mean at a certain point it's going to hurt startup recruiting but maybe it's good for the country.
26:52
Um, but I'm interested to hear your opinion on where things are going or where things should be going around bringing engineers into the government. >> Yes. Yeah.
27:04
Well, well, just to go back to also to to Dan's book, like there there's a part of the book that I, you know, I don't agree with everything that he's stating when um, you know, when it comes to US versus China.
27:12
Um but I think the thing that is 100% in agreement that the Pentagon also agrees on um is that production is a real problem in in the US and we really have to rebuild our manufacturing.
27:24
>> Um not not only you know manufacturing across different industries but specifically focused on the defense industrial base and so a lot of our investments are prod what I call production investments.
27:32
They are things that they're you know investing in companies that are going to be producing things at scale.
27:37
Um and for a very long time Silicon Valley didn't touch that category of investment.
27:41
And so I think um you know the Department of War understands that they understand that we're on the back foot when it comes to things like ship building, when it comes to things of really producing in mass.
27:51
So I think you know to to to that point um it it is sort of a a crisis moment where we really do need to be you know beefing up our reindustrialization and really um thinking about how to how to produce more in terms of bringing engineers into the government.
28:04
Uh we we have to make that easier.
28:07
have to make that easier. um like it is it is something that I think you know I I always say like the the Doge legacy is is is much longer and I think misunderstood in terms of of what that that team did because I think that team really opened up the eyes of a lot of different departments who who now are doing themselves as they say definitely in the Pentagon that's happening but it's across um you know a wide array of
28:29
government realizing that we need to bring in people who actually understand engineering and who are actually talented and I think there's a lot of good people in this administration who are thinking through how do we normalize
28:39
um you know someone someone who is you know early in career coming in for a couple years and and kind of getting over that hump of them you know knowing that they can go back to industry after they've done their tour of duty. Um we
28:49
Um we just had Scott Cooper who's who's now the director of of OPM on the A16Z pod.
28:55
He had been you know the managing partner of A16Z for 16 years and he's now leading we joked you know we thought we thought he had a big job leading 600 people at A16Z. Now he's leading 2.
29:02
3 million people in the federal government.
29:06
Um but that's one of his priorities.
29:08
Um he has talked a lot about how do we bring emerging young talent into the government and it's actually a major opportunity for the government now because if you're looking at at employment trends and if you're looking at sort of where where are people having a hard time finding a job right now, college graduates with no experience um are are experiencing higher unemployment um than in recent years. >> Yeah. Even CS grads, right? >> Yeah. Which is crazy.
29:30
We never thought that would be the case 5 years ago.
29:32
It was like the guaranteed job.
29:33
was like the guaranteed job. it it yeah guarantee job and but it is a huge opportunity for the government because when you look at sort of the the demographics of government it is very topheavy it is very 55 and older heavy in terms of the people who are working in the government and young people don't consider that an option so I think there's a huge opportunity for the US
29:53
government to to go after those young people and say hey you have an opportunity here and it's not again it's not a forever career like we we want to change the tenure terms around if you go into government you're in government forever but if you come in for two or four years you understand AI, you work on very discreet projects and then you go into industry, it will be celebrated. And so I think we're going to see a lot
30:11
And so I think we're going to see a lot of experimentation around that and it's a massive opportunity for for college graduates who may be thinking, okay, what am I going to do in this job market?
30:18
Well, go work for the US government for a couple years, serve your country, and then after you have that experience, particularly if you can talk about the projects that you that you've really worked on.
30:26
Um, you should have a job.
30:28
Um, and we should make sure as the private sector that we we celebrate that that that work.
30:32
um and that we see it as as good and that we bring it into our companies after the fact.
30:38
>> On the topic of reindustrialization, I feel like one of the most underappreciated re-industrialization stories that's happened, there's a lot of projects that are really long-term.
30:48
Uh Adrian's doing fantastically.
30:50
It's a it's a it's a big tough nut to crack, but we've seen with the NeoClouds with Coreweave and Together and Nebus and Crusoe, like there are American companies building massive multi-billion dollar projects and they've kind of gotten lost in like, oh, they're an AI company, but I feel like there's something there about, wait, we're actually building something and what Elon's doing with Colossus, too.
31:15
too. like we are building crazy huge infrastructure when we're excited about it and the capital markets are you know marketing it at AI multiples but uh is there some sort of narrative reframing or or even just taking the lessons from a core weave or a cruso and saying if if
31:33
Crusoe can build Stargate or if Elon can build Colossus 2 certainly we can make a massive drone manufacturing facility or a boat manufacturing facility and just kind of breaking that that narrative and then also just bringing those skill sets back into other pieces of the manufacturing economy. >> Oh, totally. And I and I think you know >> Oh, totally.
31:52
And I and I think you know Cruso is a very good example, but but there's other companies that are really thinking about this from like a modularity perspective, right?
31:57
Like how do you build a system as quickly as possible and then build 10,000 of those systems so you can build the data center of the future so that you have, you know, modular energy systems that you can stack on top of themselves.
32:08
can stack on top of themselves. And I think that that like that there isn't actually that much of a difference in terms of like manufacturing theory um that if you build a modular system and you scale it up and you have the manufacturing lines that you can produce
32:22
at scale as quickly as possible, you know, these companies like Crusoe, like Exawatt, these companies that are really focused on the data center, how do we get energy and how do we, you know, how do we do it in a a very modular way um as quickly as possible? I think that I
32:34
I think that I think they're going to to to to have lots of learnings that are very applicable for everything that's happening with how do we build drones as quickly as possible in a modular way?
32:44
How do we build missiles in a cheaper way that aren't, you know, million-dollar missiles, but maybe there are 50k missiles, right?
32:48
Like, how do you bring down the cost because you're not building to these bespoke requirements or building these massive infrastructure projects where you have to work with EPCs and re reinvent the wheel every time that you do it.
33:00
um and and and you know you get halfway through and it it's not a working system.
33:05
So I think the sort of the I mean this comes back from like you know the SpaceX principles of manufacturing which I think are relevant for any company um that's building in in the world of atoms.
33:16
You build small and you build it in a modular way and you build it as simply as possible and then you build as many as possible.
33:22
Um, and whether you're building uh, you know, a a Starlink constellation or whether you're building um, you know, you're powering um, you know, powering data centers as as Crusoe is doing, like you're you're you're doing it in a way where it's modular where you can just build the the manufacturing systems and then watch watch it scale build as many as you possibly can.
33:44
Build as many manufacturing lines to build the same thing over and over again.
33:46
Um, and I think that's the future.
33:48
That's that's attractable systems.
33:49
That's what the DIB is doing.
33:51
Um and it's across every ecosystem and across everything that whether it's the department of war or whether it's these massive hyperscalers like this is how they want to be able to acquire um systems so that they can so that they can work on the important projects they have.
34:05
>> Over the last couple months there's been some not so good jobs data around manufacturing.
34:10
A lot of people wanting to take a victory lap on and and just say like you know re-industrialization is impossible.
34:17
At the same time, we talk to founders every single day that are building manufacturing facilities that are scaling all the all these good things.
34:26
How do you think of timelines around uh you know with with all the sort of like geopolitical uncertainty and chaos of this year?
34:34
you can see why there would be some contraction around manufacturing, but then there's also bright spots early stage and and you can see how there can be expansion over time as as companies like you know Hrien really scale up and and you see that across >> even if you're a company that manufactures internationally freaked out during the tariff war.
34:53
It's like you're not going to be hiring people to make stuff in America within six months.
34:58
Like it just takes time to actually lease a place and grow.
35:02
So we could still see that.
35:03
I'd love to hear what you have to say. >> Yeah.
35:05
No, I mean I think the one I mean the changes that are happening in this year's NDAA to open up the the production capacity for startups.
35:10
production capacity for startups. um it it's a sea change like everything we're seeing I always say like early stage investors and even you know even late stage investors on the private side they see where the capital is flowing and
35:22
sometimes that data doesn't make it to the numbers until a little bit later as you're saying um or it doesn't make it to public markets where public markets are shifting how they're thinking about where they're going to manufacture um you know over over several months or years. Um, but what we're seeing on the
35:34
Um, but what we're seeing on the early stage side is that production companies are are hot, right?
35:38
Like investors are investing in companies that have figured out how do I produce one of something and go to 10 and then go to 10,000.
35:47
>> And those companies like they you there used to be sort of even 5 years ago there was sort of this fear of the middle of the market does not want to invest in anything to have to to do with hardware.
35:56
That is not the case anymore.
35:58
Um, and these companies are moving a lot faster to use your lingo of the valley of death and and what the Pentagon sees.
36:04
They're moving a lot faster through a production valley where they're able to produce at scale.
36:07
It also unlocks debt, right?
36:10
Like these companies raise massive equity rounds and then they can they they are credible enough to go out and have debt partners where they can scale their facilities.
36:16
And so the early crop of companies doing this, you know, maybe they were started in in, you know, around Andural's time of of 2017, 2018.
36:26
Gen 2 really started around, you know, I I would say early 2022.
36:28
Um, but I think you're seeing those companies grow up and scale faster than than even the Gen One did.
36:35
And and it's leading to this renaissance and now that we have sort of these regulatory changes where these companies are going to be able to compete on a level playing field that's going to lead to what what I genuinely believe is a manufacturing renaissance.
36:47
Um, that doesn't mean all companies are going to succeed, right? Like this is the thing.
36:51
We always go to the Pentagon and we say like we're not saying that all of these companies are going to be valuable to you.
36:55
But instead of just one company being valuable or no companies being valuable, we think that there's there's a very good chance that dozens of companies are going to be able to be the next primes of America.
37:05
And that's going to lead to a host of manufacturing jobs across the country.
37:09
And is a perfect example with their facility in Ohio.
37:13
Seronics is a perfect example.
37:15
They bought a shipyard in Franklin.
37:15
Um the speaker of the house was there to to open it, right?
37:19
like they're putting people back to work where they used to have jobs.
37:22
Um, and so I I think that's one of the things that's really exciting about this is that this capital is moving very quickly into places that need it and to places that have talent where they they didn't have manufacturing opportunities before, but but now they do because of private capital. >> Yeah. >> Yeah.
37:38
That that that was my read too.
37:40
It's like give it give it a little bit of time.
37:41
You just have to look at how many companies are now extremely well capitalized.
37:46
They can't just press a button and hire a thousand people overnight, right?
37:49
You need to scale, scale, scale.
37:52
So, >> well, thank you so much for taking the time to hop on the show.
37:55
We really appreciate talking to you.
37:57
>> Let's not let it go another few months until the next one.
38:00
>> Yeah, let's do it again soon.
38:03
>> Peter gave a a set of lectures.
38:03
It was at Stanford uh and it became the backbone of the bestselling book 0ero to1.
38:09
Uh the talks were focused on entrepreneurship, but it could be seen as a history lesson or political treaties or referendum on American culture.
38:17
There was a lot that came into that lecture series and there's a lot of facts in that book, a lot of stories in that book that aren't strictly directly applicable to just building a company, even though the the the the course was literally called How to Start a Startup CS183.
38:33
Um and 0ero to1 kind of turned into the playbook that founders fund ran for a long time and still runs and has produced a bunch of alpha the ideas of being founder friendly.
38:42
The monopoly thesis the definite optimist like these tealisms became real investment strategies uh for the following decade and are still holding true today.
38:49
Uh, it was a bit shocking.
38:52
It was always shocking to me that like the be founder friendly don't fire founders like stuck around as durable alpha for as long as it did.
38:59
But VCs just love firing founders. Like it's just nothing.
39:04
I Yeah, I've never done it personally, but I imagine you've always wanted to.
39:08
>> I imagine the thrill must be electric because uh even though >> somebody from their creation it must be because it clearly doesn't produce alpha because if you look at the companies that are founder led, they tend to outperform.
39:19
So, uh, you're sacrificing a lot of a lot of financial gain when you fire a founder, but the rush the rush must be incredible.
39:26
It must be better than even returning billions of dollars to your LPs.
39:30
Um, but uh um so this new lecture circuit is about the antichrist and this series doesn't fit neatly into the previous C Stanford course calendar.
39:43
That's not listed as CS 184 Antichrist.
39:46
No, it's not a computer science course at Stanford.
39:48
It's being put on by Michelle Stevens at Acts 17 um and is much more focused on religion specifically in Christianity.
39:55
Uh so the Wall Street Journal characterized one of Teal's core thesis this way.
39:59
Teal draws on a theory that the Antichrist could be an individual or entity that is incredibly charismatic but talks repeatedly about the end of the world, thereby convincing society to give it the power it needs to regulate the existential risks from science and technology.
40:15
Um, and there are lots of people that come to mind when you think about talking about dumerism.
40:18
And Teal cites Greta Thunderberg.
40:21
She said, "Our house is on fire.
40:23
We must act like the house is on fire."
40:25
Uh, she said also, "We are in the beginning of a mass extinction."
40:30
This is in September of 2020 uh 2019 and all you can talk about is money and fairy tales of eternal economic growth.
40:37
She told this to I think uh this is reported by PBS at the UN climate action summit.
40:41
She's been very focused on >> her uh her campaign to increase human suffering.
40:47
>> It's been a mixed bag, but uh Eleazar Udicowski said other doomer things. He has a new book out. He's on a book tour.
40:55
He says, "If we go ahead on this, everyone will die."
40:58
He's referring to AI, not climate change in this case, including children who did not choose this and do not do anything wrong.
41:03
Uh Sam Alman has also said some stuff that's uh sort of apocalyptic.
41:07
You know, he told the US Senate in his testimony in 2023, >> $200 trillion.
41:16
>> If I can't if I can't automate enterprise workflows, >> if I can't get 500 trillion million billion zillion999 I need999 billion. >> Yeah. Round numbers are out.
41:32
You need to go to a VC and say, "For this round, we're raising $999999999. Please, >> yes. >> Yes."
41:41
Um, >> don't make me choose. Don't make me choose.
41:45
>> Uh, >> don't make me choose between curing cancer and free. >> Don't make me choose. I will do it.
41:50
>> Saving time and saving money. Just save both.
41:53
>> We can save both with ramp. com.
41:56
>> Anyway, uh, no, but he he did he did give, you know, a bit of a doomer take.
42:00
He was talking about uh how this could go wrong.
42:02
He said if this technology goes wrong it can go quite wrong.
42:04
Uh Elon Musk also said something similar.
42:08
He said with artificial intelligence we are summoning the demon.
42:09
And the original SpaceX thesis was >> did he say that by the way?
42:14
>> He said that in uh October of 2014 at the MIT Centennial Symposium.
42:20
>> And what did he mean by that?
42:20
Because now he's got a >> summoning a demon.
42:24
You know, I think he thinks that uh if you shape the demon portal in the right shape, the demon comes through and it's kind of friendly.
42:33
Maybe >> friendly demon >> kind of nice. I don't know.
42:35
>> Friendly demon that increases enterprise value through business automation, increased efficiency.
42:41
>> Hopefully, hopefully and and and vibe coding 3JS apps.
42:43
That that's the demon that I want.
42:46
Um but uh even SpaceX you can view as a as a rebuttal to the apocalypse.
42:54
What would be apocalyptic?
42:54
A humanity remaining on the earth alone.
43:01
An asteroid comes hits the earth. We're all done.
43:03
But if we become multilanetary, uh we are no longer a single point of failure.
43:07
If there's humans on Mars and the moon and earth, if something bad happens to Earth, humanity continues.
43:14
Um, the stakes that that Elon framed SpaceX in were world consequential.
43:20
>> By the way, if you go to XAI.
43:20
com, >> it's this random company called UNK AI, unk, and they feature popular AI brands like Gro for Chat GPT. >> Okay.
43:34
>> Deepseek and Xiaomi and Uni Tree. >> Weird.
43:39
>> Seeming seemingly a Chinese company. >> Okay. Odd.
43:40
that's using it to promote a range of products. Yeah.
43:44
Anyway, it's very strange.
43:46
>> So, my question was there's a lot of focus on uh if you speak in millinarian terms, if you speak of apocalyptic consequences to not building your technology, you might be the antichrist. It might be bad.
44:00
But at the same time, I'm just looking at history here and looking at these folks who have become popular and and thought leaders on apocalyptic scenarios.
44:10
Elon, Sam, Udicowski, Greta.
44:12
Uh if you backed these folks as a basket, you would have done quite well.
44:18
You're in SpaceX, you're in uh you're in Open AI, you you've done quite well.
44:23
and and I'm wondering where the line is for uh you know using uh speaking about apocalyptic consequences to the lack of technology to the lack of not solving a problem and then balancing that with uh delivering something that isn't authoritarian but is actually just a really really big ambitious project.
44:42
So if you if you if you read into, you know, uh we're summoning the demon, we need to go to Mars because humanity could be wiped out on Earth is cooked, right?
44:51
Um there's one world where it's like that is apocalyptic thinking.
44:55
There's another which is just like this is a reframing on the classic like I want to save the world.
44:59
I want to change the world.
45:00
We want to make the world a better place.
45:02
all those terms have kind of fallen out of fashion as as a lot of founders just kind of wound up, you know, building automating manual workflows, right?
45:11
And so they they stopped saying we're going to change the world with a better database.
45:15
But there's still something if you're a VC and you want to back a moonshot, you want to back something that's either going to be zero or trillion dollars that where I feel like your ear >> Augustus is is a great example of this pitch because >> um everyday people experience rain all the time.
45:35
They're not necessarily like droughts are not something super tangible, right?
45:39
because you just you turn the faucet and the water runs regardless of if you're in a drought or not.
45:45
Obviously, farmers feel this much more intensely.
45:47
But he's come out and said, you know, we need to be able to control the weather.
45:51
We need to be able to increase precipitation. >> Mhm.
45:56
>> And you should fund me so that I can do this, right? >> Yeah. Yeah.
45:59
There's something about these bold missions that rally employees, they rally investors, they rally media attention.
46:08
You look at like what Augustus is doing is deeply controversial, but he's on podcasts that are so much bigger than what a normal what is he got? Series A.
46:18
Like a series A founder typically is not on multiple million plus subscriber podcasts. >> Yeah. >> Doing a tour.
46:29
We talk a lot >> going and talking to people that are are actively disagree with what he's doing. >> Totally. Totally. Yeah.
46:35
I mean, we talk to founders all all day long who come on and, oh, they raised at a billion dollar valuation or they raised a billion dollars.
46:41
We had five series E companies on Tuesday.
46:43
Uh, those folks are not getting calls from the biggest media platforms in the world on day one consistently on a regular basis.
46:51
And I think it's because of the that they're going after more tractable problems.
46:57
they're going after things that are less controversial, but also there's just something interesting about actually uh refocusing on tackling those really really big issues.
47:06
And so um there's something where I I'm I I feel like the fear of like targeting the antichrist turns into like uh you know Gerardian scapegoat.
47:15
You got to find someone to blame everything on.
47:17
I I'm worried about how all that shapes up.
47:19
But uh I am optimistic about this idea of venture capitalists returning to the super high risk.
47:28
It's either zero or trillion dollars.
47:30
It either is a completely useless company that doesn't get anything done or they cure cancer or they build a flying car or they build the rocket that goes to space.
47:37
Uh and even if Elon hasn't gotten us to Mars, we're not multilanetary yet.
47:41
He's still able to ship Starlink. It's great business.
47:46
And then also keep everyone still motivated on how cool would it be if if actually got to Mars. That's amazing.
47:52
Another we need we need AI to benefit humanity very oriented around safety, right? This is a nonprofit.
47:58
This is work that needs to be done that's >> not going to get funded through the capital markets and it turns into an internet software.
48:08
>> And so and so I don't know what I don't know what your timeline is for SpaceX getting to Mars.
48:13
It might be 20 years, maybe 30 years.
48:15
uh but at least it keeps them focused on you know chopping wood and working hard to actually advance the underlying technology that we get a lot of value on through just satellites and stuff.
48:26
Uh, and I think that the same thing could be true for the AI companies where yeah, maybe we don't get the the the AGI god ASI, you know, super soon, but there's just like a ton of value and you keep working at it because uh it's delivering value in the short to medium term, but you're able to keep the sites really really really high.
48:42
Uh, and that drives just like uh that's just what you need to actually marshall all the energy to go back something really huge.
48:48
This is from uh two days ago by Berber Jinn in the journal.
48:54
Open AAI unveils plans for seemingly limitless expansion of computing power in Texas prairie startup showcases ground zero of AI boom and its plans to shepherd 1 trillion in infrastructure spending.
49:06
So the trillion dollar numbers getting thrown out thrown around for the first time because Stargate was originally 500 billion and now we're up at 1 trillion. Is that right? >> Yeah.
49:17
And remember even at 500 billion Elon was basically saying hey nobody involved here actually has the cash to do this.
49:24
So, >> just wait until these stocks run up and Oracle's up 300.
49:26
Larry says, "Hold my beer. >> I'm good for my 10." >> Hold my green juice.
49:34
>> I'm here for I'm I'm I'm good for my half trillion.
49:37
>> Um but yeah, once you get into the trillion dollar range, you're approaching the GDP of uh you know, some some serious countries, right?
49:45
>> Some serious countries.
49:45
Open disclosed it would ultimately need more than 13 times the computer the computing power of its first nent site which is raising which is rising out of the Texas brushland.
49:56
Frenzied construction here has turned a sea of red dirt into eight hyper futuristic data centers bringing online roughly 900 megawatts of capacity.
50:06
More than 6,000 workers labor on the project each day, including electricians, plumbers, and steel welders alternating between two 10-hour shifts 7 days a week. Wow.
50:14
They are they are uh what is that? Uh 2 to 107 210 7 2 a. m. to 10 p. m. 7 days a week something. >> Yeah.
50:26
People in the data center business see uh people with computer jobs saying, "Oh yeah, I do 996." >> Yeah.
50:33
I mean, I wonder if they have two shifts, 10 hours, does that mean they have two crews that are working 70our weeks?
50:39
Are people taking days off or are they rotating their shifts so that you get a day off every day?
50:44
>> I think they would get into pretty hot water.
50:46
I don't think you could like run and say you're working seven days a week forever.
50:50
I don't think I don't think employment law works like that.
50:53
>> Gray towers of gas turbines have dotted the landscape since the spring, offering backup power.
50:57
On a tour with reporters Tuesday, Oracle and OpenAI executives showcased the 1,100 acre site, calling it the largest AI supercomputing complex in the world.
51:07
Outside workers wearing dust masks and sunglasses shuttled around in buggies trying to shield themselves from the 100 degree heat.
51:13
One of the buggies flew a flag that said Jesus is the answer. >> Not very hgi pill.
51:19
I thought we were getting God in a box. What's going on?
51:26
Just Eleazar Udicowski shows up. He's like, "No, no, no.
51:30
Like the they're building God here. You're building God. You're doing the thing.
51:34
Like what are you doing here?"
51:36
Uh there was literally nothing here a year ago uh said uh Anous who works on the open AI computing team.
51:44
They also announced five new data sites data center sites across the US built with Oracle and Japanese tech conglomerate SoftBank.
51:50
It said the new facilities would help bring online nearly 7 gawatt of power.
51:53
Enough for almost 8 million homes.
51:56
Enough for a lot of API calls from Devon, from Cognition, the makers of Devon, the AI software engineer.
52:05
crush your backlog with your personal AI engineering team.
52:07
Company executives made it clear that the Abene site was just the beginning, noting they envisioned a need for more than 20 gawatts of computing capacity to meet the explosive demand for chatt which now has more than 700 million weekly users.
52:23
Um, I was laughing about, you know, how in the chat GPT if you fire off pro like GPT5 pro, it kind of thinks for like 20 minutes and everyone's saying like, oh yeah, this is good.
52:35
Like, we want the the agents to be reasoning for longer and longer.
52:38
I I feel like there's an opportunity for some startup just to be like, we've cracked the code.
52:42
Our model reasons for two days straight.
52:44
So, send us your request and we'll get back to you in two days.
52:50
>> And the agent just like is pretending.
52:52
So it's just like twiddling.
52:53
>> Yeah, it just has a progress bar >> twiddling it some until the last 20 minutes and then just starts going. >> No, no, no.
52:58
Of course, what you do is you is you have the loading bar there and then secretly you have a human doing the work and you have a human that's like dealing with this.
53:07
>> Well, I think the thing is like the longer the longer the the the period, the higher the expectations.
53:11
And I just don't know that I don't I I think that uh I guess the thing that I would want to know is like if you're running a deep research query for 20 minutes and then if you decide to run it for 2 days, how much better is it? Right. Yeah.
53:26
If it's >> if it's 10 times better, >> yep, >> could potentially be worth it for different types of tasks.
53:34
If it's >> but 30% better, is it worth the wait?
53:39
Yeah, there there is a world where you just route it to the human.
53:42
There's also the world where for for the for the really low >> don't encourage this.
53:47
People have already gotten into trouble. >> Yeah, they have. They have.
53:50
Uh the the other funny idea is like for a lot of the there have to be a certain set of queries that just get asked the same like every single day.
54:01
And they should just store those in a database and cache them because if like you have to imagine that every single day someone is asking like just going to chat GPT and lighting the GPUs on fire to just ask like what's the capital of Illinois, you know, or what's the capital of California?
54:17
>> Can you run a deep can you run a a GPT like pro deep research report for what is the capital?
54:25
>> You can do it for anything.
54:25
Yeah, you can you can go to deep research and fire off 20 minutes of compute and say what is the capital of California answer in just one word and like it will just think for 20 minutes and then spit out Sacramento like >> No, it won't think that long.
54:39
It it like thinks based off how hard it is.
54:42
>> No, deep research will >> I just said what is the capital of California answer in one word and it's >> deep research.
54:49
>> Yeah, it's doing a deep research query.
54:52
>> Let's see how long this takes.
54:52
I I I it's not going to it's not going to come back in in 2 minutes.
54:55
It's going to it's going to cook for 10 minutes. >> All right.
54:58
I was a little bit late, but I just started a a timer. >> Okay.
55:01
>> So, what do you think?
55:02
>> Also, there's like a sense where like um the like embeddings of facts is basically just the model already.
55:07
Like you can just think of a model as like a compressed version of the internet or at least like non-reasoning models. Yeah.
55:13
So it's kind of like >> but wouldn't it be even more compute efficient to just do fuzzy lookup of of these questions and see if they >> in the chat says cachebased retrievals are >> oh yeah >> but apparently not here because >> GP5 fully implemented
55:30
>> okay okay it reasoned for 47 seconds >> okay >> and it >> this is this is pro you didn't do deep research though >> I want to see you turn on deep research and do it >> well this is the pro which is research search grade intelligence. >> No, no, no. That's not the deep research >> No, no, no.
55:43
That's not the deep research thinking. >> No, no, no, no.
55:46
You need to click on the little plus button next to the chat box and then check the box on on uh on deep research. Do they still have this? I don't know.
55:56
They might have they might have deprecated this.
55:57
I think it's still there. >> It's still there. Still there. Okay.
55:58
I'm doing a deep >> You're doing deep research. >> No, no, this is good. This is good.
56:02
Just to confirm, are you asking for the current capital of California or or are you interested in historical capitals as well? >> Yeah. Let's see. Current capital.
56:14
>> Tell it to also design a uh a website in Figma.
56:17
Think bigger, build faster.
56:17
Figma helps design and development teams build great products together. Get started for free.
56:23
Uh while we're waiting for that, let's keep reading from the journal.
56:26
Each gigawatt of capacity is expected to cost roughly 50 billion, Tyler. 50 billion per gigawatt.
56:32
That's the number in the journal.
56:34
Meaning the company is laying the groundwork for at least 1 trillion in infrastructure spending.
56:40
Demand is likely to reach closer to 100 gawatt, one company executive said, which would be 5 trillion, >> which is roughly the GDP of Germany. >> Or Japan.
56:53
>> Yeah, GDP of Japan is 4 trillion. Germany is uh 4. 6. >> Good luck.
56:58
Good luck, Germany and Japan.
57:00
You're getting left behind.
57:00
You're part of the permanent underclass now. It's over for you.
57:04
>> I feel bad for lighting the GPUs on fire for this.
57:06
It says >> it's >> got it.
57:08
I'll confirm the current capital of California for you and get back shortly. And it's >> it'sable. >> It's cooking.
57:15
Like this is the real pitch for the uh >> I'm getting my money's worth.
57:20
>> Brandon Jacobe in the chest is breaking.
57:22
Jordy discovers the UX issues of every foundation model company because >> fix it, Brandon. >> Yeah. Yeah.
57:27
Aren't you doing consulting now, Brandon?
57:29
You got to get in the in the trenches. Yeah. all these folks.
57:33
>> If the labs haven't hit you up yet, >> if someone who spends 10 hours a day studying this stuff, >> it's reading Wikipedia >> just like how many drops of water did this use?
57:44
How how many fish had to like go find some other it should come back and it's like thought for 15 minutes put made two species extinct >> lit lit seven gallons of diesel on fire.
58:01
Okay, it's it's checking all the sources. >> It's good.
58:03
I I like a detailed report.
58:03
I like to know that the AI did its job.
58:08
Don't just sit there clanker and riff off cashbased retrieval.
58:09
I want the deep research every single time.
58:14
>> Even for one word fact.
58:16
>> Even for one word fact. I want you to be sure.
58:18
I don't want any hallucinations in the >> Well, I mean, this is a this is a tough question, too.
58:22
A lot of people that aren't from the West Coast or the United States would say, "Yeah, what's the capital of California?" Yeah, San Francisco.
58:29
I was using this as an example about >> I remember being I must have been like seven years old asking like figuring out that Sacramento was the capital of California and I thought that was the funniest thing ever. >> Yeah, it is. It is hilarious. >> All right, it got it.
58:43
It thought for 101 seconds.
58:47
>> That's not that long for deep briefing.
58:48
>> It went for four sources. It did 19 searches.
58:50
So, this this is why this is why I can't trust, you know, everybody keeps screenshotting uh uh these Google trend reports. >> Yeah. Yeah.
59:01
>> I'm not sure that Google has figured out how to >> exclude >> exclude agents because this just did >> 19 searches for a single fact. Yeah. And so that's true.
59:13
>> If you're looking at the growth of queries on Google, you would think >> and and if a lot of people are asking this kind of question, every single >> keyword is going to be spiking.
59:25
>> Well, if you want to get your brand mentioned in chat GPT, you got to go to profound. That's right.
59:28
Reach millions of consumers who are using AI to discover new products and brands and get a demo.
59:35
Um, I don't think we've figured out the final form of what financing for compute looks like, said Sam Alman.
59:41
But I assume, like in many other technological revolutions, figuring out the right answer to that will unlock a huge amount of value delivered to society >> and delivered to open AAI shareholders. >> For sure.
59:56
Yeah, it is it is an interesting question because you know Daario was talking about how like each training run looks great by itself but when you put them all in one structure it looks like you just have this neverending money pit because you spent a hundred million on a training run.
1:00:10
You made a billion over the next two years.
1:00:14
You spent a billion on a training run.
1:00:16
You made 10 billion over the next two years.
1:00:18
You spent a hundred billion on a training run.
1:00:19
You make a trillion over the next two.
1:00:21
And there's a question, okay, how long until it just completely maxes out?
1:00:24
Are there diminishing returns to these things?
1:00:26
But even if the trends hold, you still wind up with this like ever growing money pit that should act as like some sort of slowing down force.
1:00:36
You would think >> I think when you when you dig into this quote, I don't think we figured out the final form of what financing for compute looks like.
1:00:41
The question is like will this actually be a novel financial instrument or it will be remixing something that Wall Street has used in mortgages? >> Maybe it's debt. Maybe it's debt.
1:00:52
who knows >> GPU back securities.
1:00:56
>> Uh the Tuesday announcement made it clear that SoftBank once seen as a formidable OpenAI funding partner has scaled back its ambitions in the data center buildout which the Wall Street Journal reported in July.
1:01:04
Three new sites, one located near Abalene, another north of El Paso in New Mexico and a yet to be announced Midwest location combined with an expansion to the Abene complex will be capable of delivering 5. 5 gawatts of capacity.
1:01:16
Those will be be built by uh Oracle.
1:01:19
The other two smaller sites, one in Lordstown, Ohio, and the other near Austin, Texas, will be built in partnership with SoftBank and generate 1.
1:01:27
5 gigawatts over the next 18 months.
1:01:30
The first completed data center called building one painted a pristine white that contrasts with the reddish dirt surrounding the site is larger than two Walmart Super Centers.
1:01:41
Doesn't feel that big to me, but certainly huge. Uh entering.
1:01:43
Yeah, the the 1100 they said, 1100 acres earlier in the article.
1:01:49
That's under two square miles. >> It is.
1:01:52
Yeah, it is crazy how compressed this is.
1:01:54
I mean huge when you're walking around, but you know, we're not in Okay, we're we're we're losing Yusede over this yet.
1:02:01
>> Yeah, we're not in the, you know, >> in the full terraforming.
1:02:03
I mean, I said told the what was it the San Francisco Chronicle that we could he could imagine a future where the entire earth is covered with solar panels if the trends continue.
1:02:14
Gabe Gabe in the chat says, "Shout out to Lordstown, Ohio.
1:02:17
Factory used to build Chevys.
1:02:20
This is your your thesis or or take.
1:02:23
Re-industrialization is happening.
1:02:26
Just it's a lot less jobs and it's data centers." >> Yep.
1:02:31
Um >> but the the the uh white pill there might be that uh if we can build big data centers really quickly in in uh you know at at the scale that traditional infrastructure projects would have taken 10 plus years.
1:02:47
If we can do that in a few years >> it could create a precedent where somebody could decide you know what I'm going to build this this new rail line in few years.
1:02:56
Is a is a Walmart super center really one mile long?
1:02:59
Uh QBS rollover saying that um that the uh that the building's 2 miles long. That mean >> Well, no.
1:03:09
I was saying I was saying the site is is 1100 acres total.
1:03:10
It's not uh >> uh which is like 1 1. 7 huge uh square miles.
1:03:20
So obviously massive, but I don't think the building itself is uh anywhere near that.
1:03:24
Well, fiber lines snake across the data centers and underneath the ground, allowing the AI chips called GPUs, in case you were wondering.
1:03:29
The Wall Street Journal's got you covered uh to talk to one another and complete requests more quickly.
1:03:35
Proponents of the infrastructure boom say it will bring hund hundreds of thousands of jobs and revive American manufacturing.
1:03:41
In January, OpenAI unveiled a $500 billion data center uh data center project, Stargate, we know about this.
1:03:46
Uh the reality is more mixed.
1:03:48
While data center providers uh provide plentiful temporary construction jobs, far fewer people are needed once they once they are built.
1:03:56
Abalene mayor uh said residents had mixed feelings about the site and its power and water usage, though some of the concerns have been assuaged.
1:04:02
An Oracle executive said there will be roughly 1,700 permanent jobs on site once the construction ends.
1:04:07
I think the things that ends up being a little bit yeah disappointing is is these projects get announced and they're saying we're going to spend $20 billion here and then we're going to create like a thousand jobs or we're going to create 300 jobs and that is very significant.
1:04:26
that will be good for the local economies uh in these regions but not it's not hey we're >> we're replace you know you we we lost 87,000 manufacturing jobs last month I think it was >> we're not really making a dent in that on data center
1:04:43
>> it takes a lot >> Joe weisenthal has an absolute banger the emergence of quote unquote slop was foretold as soon as we started consuming content via the feed >> I didn't see this I put this this is how kicked off the newsletter today. Go to Go to tbpn. com, drop your email.
1:04:57
Uh, it says, "Why do they call them social media feeds?
1:05:02
Are they feeding a slop at a trough or seating us at a Michelin star restaurant?"
1:05:05
It's both and always has been.
1:05:06
You can use social media to follow the most insane craftsmen who spend years researching, writing, producing, and distributing content.
1:05:11
Or you can slop it up with some viral videos of scantily clad hot people in street fights, sometimes both, in the same video.
1:05:18
The distribution of between craft and slop has been growing for decades, maybe forever, and is continuing to grow with the development of AI content.
1:05:27
>> And this this post got 10,000 likes from 0005 seconds.
1:05:34
>> We at Meta are delighted to announce we've created the infinite slot machine that destroys children from the hit book.
1:05:39
Don't create the infinite slot machine that destroys children.
1:05:43
>> This is such a good coffee pasta.
1:05:43
I forget what uh the the torment nexus is the original one.
1:05:49
It's like everyone in tech is like we've created the torment nexus from the famous book don't create the torment nexus.
1:05:56
I just think torment nexus is hilarious.
1:05:58
Um we will see if this destroys children.
1:06:02
Um I don't know how much it will destroy children.
1:06:05
I don't know how much it'll destroy everyone.
1:06:07
There's an interesting dynamic.
1:06:09
uh most parents are aware of technology and and that it needs to be measured and released in in you know >> if if any parents are unaware of technology we would love to interview you >> for sure.
1:06:21
Um but but the iPad kids meme is a thing.
1:06:25
Are you going to give your kids an iPad anytime soon? >> Only on flights. >> Okay. Yeah.
1:06:31
Just as a replacement like watch a movie basically. >> Yeah.
1:06:35
It's basically it's basically like I know this is bad for my child but >> so limited >> I don't want to ruin you know four hours with you strangers.
1:06:45
>> So I mean there was a time when like I actually grew up uh my parents were art art majors studied fine art and uh film and they insisted that I don't watch TV. >> Yeah.
1:06:58
>> Like do not watch television.
1:07:00
>> I was only allowed to watch television in Spanish. >> Yeah. So I Oh, interesting. >> Like early on. So it was like Yeah.
1:07:04
If you want to watch >> Okay. Yeah. You're learning. Oh, that's cool.
1:07:09
Get ready to learn Spanish. >> I love it.
1:07:23
clearly worked perfectly.
1:07:23
Um, but but I had to watch a lot of the AFI 100, the top the best films.
1:07:28
I've seen Lawrence of Arabia like 25 times because of that, because that was one of the approved movies that I could watch.
1:07:33
Um, and and I think parents know that, you know, you need to let the dust settle on the new technology before you expose the kids to it.
1:07:41
And so, um, the phones are getting restricted.
1:07:45
There's screen time limits for most kids.
1:07:46
There's this idea of like let the let them cook, let them simmer, let their brains develop.
1:07:51
Uh keep them focused on books and writing and reading and then yes movies, but maybe a Pixar movie once a month, you know.
1:07:59
Uh the the I think it was Andrew Sorcin that was interviewing Peter Teal and said kind of put the screws to him and was like, "You're on the board of Meta.
1:08:09
How much screen time do you give your kids?"
1:08:11
And he was like, "90 minutes a week."
1:08:12
And that feels like a very reasonable place to be where you're not going to become >> Not all screen time is created equally, right?
1:08:21
You have a an old school animated show. >> Yes.
1:08:26
>> And it's handdrawn and it's sort of >> even just the moral lessons in one show can be wildly different than another.
1:08:34
Like there can be shows that have really com they grapple with really complex social issues or or instill, you know, they're retelling like if you're if you're watching something that's like essentially a a transposition of a Romeo and Juliet story or a biblical story or something that teaches you to turn the other cheek or the golden rule, right? The hero's journey.
1:08:57
these things are going I I feel like you know uh like letting a kid watch the original Star Wars is not going to make them like a worse person.
1:09:05
It's going to be like yeah be heroic save the princess like like you know fight the bad guys right like those are sort of good and you need to unpack it and you need to follow the story lines from the beginning to the end and stuff.
1:09:19
It's not just something that you can just mindlessly turn on like it builds a whole world and it helps you imagine things and explore the ideas of space travel and science fiction.
1:09:27
like these are these are not that bad for you.
1:09:29
I think at least in limited doses.
1:09:31
Um and I think this will be the same for this AI slop.
1:09:34
I think that uh most parents will say um you're not getting that much out of this so it should be limited.
1:09:42
This is very much candy and then there are >> the only thing I would say is I don't think most parents are engaged enough to yes actively control screen time. >> Yes.
1:09:54
And so there will be a barbell where certain parents will will will police it and see very good outcomes for their par for their children and other parents won't and they will see bad outcomes from their from their children.
1:10:05
And then there will also be dynamics within these groups where if you look around and you see that your peer group is all zombies because they've all been spending every waking moment glued to AI slop and they can't tell you anything because they're just watching like nonsense.
1:10:20
Well, then there's incredible alpha for going to the library and reading a book.
1:10:24
And so, someone will figure that out and realize, wait, if everyone else is completely non-competitive and I'm the only one that's read this book, I will have the ability to make a ton of money and get a huge house and live a wonderful life.
1:10:38
And so, I should lock in and I should turn that stuff off, resist that, and and become a more interesting, dynamic, creative, and you know, independent thinker.
1:10:47
And so there will be this natural game theoretic outcome where if everyone's falling for one thing, I mean, it's the land of the Lotus Eaters.
1:10:55
If everyone's doing drugs all day and you're the one that's not on drugs, you're going to be pretty powerful in that society, right?
1:11:00
Like if everyone's drunk all day and you're the only sober person, well, you're probably going to be, you know, the president or, you know, the the the ruler of the world potentially.
1:11:09
I have no idea if this uh stat is fully accurate, but one study showed that the average American 5-year-old watches 2 to three hours of TV a day. >> That's a lot. [Music]