0:01
[Music] Hey, hey, hey. [Music] Heat. Hey, Heat.
Search this video
0:50
[Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] We came to this world. to reach the stars.
3:11
We came to this world to shape our future.
3:23
You came to this [Music] world to reach the [Music] stars to shape a [Music] future.
3:40
Market clearing order inbound. [Music] Oh, heat, heat.
4:19
[Applause] We are surrounded by journalists. Hold your position. [Music] You're on. You're watching TVPN.
4:49
Today is Thursday, June 5th, 2025.
4:49
We are live from the TVPN Ultra Dome, the temple of technology, the fortress of finance, the capital of capital.
4:58
We got to work on that because uh we're working on selling the naming rights, baby.
5:03
This place is going to be branded.
5:05
We're going to sell the windshield.
5:08
We're selling the windshield.
5:10
Uh so we got to we got to keep growing the intro.
5:13
Um, but we have a massive day today.
5:15
Little bit of an AI day today.
5:18
We got folks from Google, OpenAI, Anthropic.
5:20
Uh, we got the former CEO of OpenAI coming on Google X investors.
5:29
We got pretty good coverage.
5:31
We hit almost everything.
5:31
Should go on a whirlwind tour of what's going on in artificial intelligence.
5:35
I'm excited to dig into the state of affairs in the foundation model race.
5:40
We're going to go through uh the the the tier list of what companies in AI have the mandate of heaven.
5:47
We're also going to go through some of the deep research products and hopefully get into some of the more um cutting edge use cases for AI.
5:53
So, we have both some deep research folks coming on and then we also have uh some folks that are working on uh video generation and video game generation and a lot of different uh applications.
6:05
We're going to cover the granola story.
6:07
We're going to cover what's going on with Windsurf.
6:08
And so, it should be a great day.
6:10
But let's run through some news just to keep everyone up to speed before Sean Magcguire joins in 13 minutes.
6:17
So, first off, ramp time is money. Save both. Save both.
6:20
Easy to use corporate cards, bill payments, accounting, and a whole lot more all in one place.
6:23
Um, market clearing order circle went public.
6:26
The CEO is coming on the show tomorrow. That's very exciting.
6:30
And Jordy, you have the news.
6:30
I will read a little bit uh from Jeremy, the CEO.
6:34
I'm incredibly proud and share and thrilled to share that Circle is now a public company listed on the New York Stock Exchange under Circle.
6:40
Um Brian Armstrong post, congrats to Jeremy and the entire Circle team on your IPO and reaching 30 trillion in lifetime USDC volume.
6:51
Let's hit that the big T.
6:56
You got to trillion stock.
6:57
The stock is up massively.
7:01
It went uh it was priced at $31.
7:03
It is trading at around 85 as of now. And that's fantastic. We love to see it.
7:10
Uh again, uh as many people would expect, Bill Gurley is is going to be very unhappy with that inefficient pricing.
7:17
He hates he hates a stock.
7:17
He hates a pop during after the IPO.
7:20
It's good for the IPO window, which we all we we always want to be open. For sure. For sure.
7:28
Um, Andrew Executive Chair Trey Stevens tells Ed Ledllo that the company has closed a new funding round of 2.
7:33
5 billion in a deal that more than doubles the defense startup's valuation to 30. 5 billion.
7:41
This is from Bloomberg TV.
7:43
Congratulations to the Anderal team on the massive up round. You'll love to see it.
7:47
We got to hit the gunk for Android. Do it again. We have good contact. Good contact. Very exciting.
7:55
Uh bunch of news from uh from Kevin Wheel over at OpenAI.
7:59
We will be digging into this with Mark Chan today when he joins.
8:02
But Deep Research can now share across GitHub, Google Docs, Gmail, Google Calendar.
8:06
So you can integrate everything and it can do research on your files.
8:10
Uh that's going to be a lot of fun to talk about.
8:12
Um this is also potentially threatening for Granola and so we're talking to a Granola investor about uh what the reaction will be where the direction of that company might change or not.
8:22
Um, but if you're designing a tool for artificial intelligence or one of these products or any tool really, go to figma. com.
8:28
Think bigger, build faster.
8:30
Figma helps design and development teams build great products together. Go to figma. com.
8:34
It is the backbone of the TVPN brand. It is. It is.
8:40
And we would not be able to make the show without it. Yes.
8:42
And so we are going to be digging into today's obviously artificial intelligence day in some ways.
8:49
Uh, we're digging into artificial intelligence today.
8:51
We're also very interested in talking about uh VR and content and augmented reality.
8:57
And there's and there's a story in the Wall Street Journal today about Meta is in talks, not advanced talks, just regular talks, regular old talks, but they're talking to Disney, they're talking to A24 about content for a new VR headset.
9:09
And this is my number one question about VR.
9:13
When the iPhone, everyone's looking for the VR, the iPhone moment of VR.
9:17
When the iPhone debuted, what was it?
9:20
It was first and foremost a phone. It replaced your phone.
9:24
And so, I've always thought that the path to true VR adoption was just saying, "We're only going after your TV."
9:32
The next generation of 20, 22 year olds when they get to college or post college and they're in their first apartment, they are just not going to buy a big flat screen TV because we have solved that specifically with VR headsets. Exactly.
9:44
or or you know the guy with multiple monitors, three three monitor setup uh the production team back there, we can go to the production camp, show you all of the different uh all the different monitors that they have.
9:55
What if they could be wearing VR headsets?
9:57
They could have seven monitors one day. 10 monitors.
9:58
Let's give it up for the production team.
10:03
So, so um the idea Oh, we're on drink camp. We got the drink. Thank you to Matina.
10:08
Thank you to Thank you to Andrew Huberman for inventing drinking things and caffeine and the whole team over at Matina.
10:16
Yeah, they're crushing it.
10:16
Um and so uh this idea of of of doing one thing really really well before before going into uh you know trying to do a little bit of everything like Exactly.
10:28
You only get to be a platform if you solve one thing really really well.
10:30
Uh the iPhone didn't wasn't a platform.
10:32
The first iPhone was not a platform.
10:34
Didn't have an app store, right?
10:35
It just had uh it just had the ability to to listen to music.
10:40
It was an iPod, it was a phone, and it was a internet communicator, just a web browser.
10:43
Um and so the Meta is looking for to Hollywood for exclusive immersive video uh for premium device.
10:51
Now, my kind of hot take here is that cooking. I know. I know.
10:55
It's a all all of our boys are coming together doing we're cooking up something amazing. I'm really excited.
11:01
We're not going to be able to get that much information on this soon, but uh I I don't even know if they need that much immersive content.
11:07
I think a lot of it is just, hey, every single Meta headset should just ship with the Matrix pre-installed for free. Yeah.
11:12
It's like how much would that possibly cost?
11:14
It's an extra $2 to rent or something.
11:16
You could like have it pre-installed.
11:18
So, it's just like you can put it on and there's like 10 movies that are pre-loaded.
11:22
You can just watch movies and the movies are great and you're in a really nice theater and it just comes pre-installed.
11:25
It's all great because that was that was the my vision pro experience was very much uh film driven.
11:32
I would take it a step further and I think they would have to basically create an entire catalog.
11:35
Like I don't know if the Matrix by itself is going to be enough of a draw to say I'm going to spend hundreds of dollars for the average person. No, no, no.
11:43
It's more about like the pre-installed apps.
11:45
So like the iPhone was a really good phone.
11:49
No, but I'm saying really good iPod, but it also came with like a calculator app that was like decent.
11:53
And so you need a few of these things that are just like really easy to access, really to pull, really easy to pull off the shelf.
11:59
And ultimately, I think Meta needs to catch up to Apple in terms of the Apple TV movie store and and making sure that like all the streaming providers are really on there in a in in a in a valuable way.
12:11
Obviously, it's it's important to go to immersive eventually, but I think the path to immersive might be just, wow, I have a home theater in my studio apartment now.
12:20
Anyway, we'll dig into this more.
12:23
We'll talk to more people. Maybe I'm wrong. Who knows?
12:25
Uh, but the high that the the high points from this article in the Wall Street Journal are as follows.
12:30
Meta is seeking exclusive content from Hollywood for its upcoming premium VR headset, LMA, uh, set to rival Apple's Vision Pro.
12:36
I'm super excited for this new VR headset.
12:39
I think it's going to look fantastic.
12:41
I think the resolution is going to be insane.
12:43
And obviously there's a lot of focus on augmented reality and and Orion and AI and glasses.
12:48
But there's still so much work just to do just to bring VR into a into just a normal consumer experience.
12:56
And what's interesting is that I think Apple really broke the seal on like yeah like you know people are used to paying $1,000 for a phone and $2,000 for a computer and maybe $4,000 for a headset. Isn't that crazy?
13:07
So, what can Meta's been been hanging out in like the 3, four, 500 range?
13:12
If you take the reinss off and say, "Hey, yeah, yeah, yeah, it's fine to spend $1,000 on this thing."
13:18
Uh, you could get something really, really interesting. Totally.
13:20
And so, Meta is offering millions for video based on well-known IP, aiming to attract users to its VR device launching next year.
13:28
Now, the big question is how long will these immersive videos be?
13:32
Because, uh, Apple did do a bunch of these deals.
13:34
They did license a bunch of uh interactive video products and but they were all like fiveinute experiences. Yeah.
13:41
And so you you you get through them all and then you'd wait a full quarter and Apple would be like we have another one. It's seven minutes.
13:48
Here's five new minutes of entertainment.
13:50
It's like that's not how people experience entertainment. I remember like Yeah. Yeah.
13:53
If you just think of it, a lot of people are use being entertained by their iPhone for four hours a day. Yes. Often through video.
14:01
And it was but but that's not even an iPhone thing.
14:02
go back a few decades to like the original PlayStation had Final Fantasy 7 on it.
14:08
It came on multiple discs and that game would people would play it for a hundred hours.
14:15
Metal Gear Solid was a similar like dozens of hours of gaming and and no one's really been able to deliver that in VR and have that moment.
14:21
Same thing with uh uh GTA, you know, hundreds of hours of entertainment.
14:26
Uh anyway, uh very excited to dig into this new device known as LMA.
14:31
Uh, it's more powerful than the MetaQuest VR headsets now available with higher fidelity video. Let's hear it.
14:36
They got the screens done.
14:37
They pulled them forward off the benchtop.
14:39
Uh, the design is similar to the large pair of eyeglasses, more like Meta's Ray-B band AI glasses than goggles that the Quest and Vision Pro use connected to a puck that users can put in their p pockets.
14:49
So maybe they're going puck, which is interesting because that was very contrarian.
14:52
Every was like this is Steve Jobs would never let this happen.
14:55
And Palmer came out and said, "No, puck is great.
14:57
keep the you don't want heavy things on your face.
15:00
That's just not a good experience.
15:02
Um and so Meta is planning to charge less than a thousand but more than 300.
15:06
Um and so I would say $9.
15:10
99 is probably the right price.
15:12
I want it to be sort of expensive so it can be a great product.
15:14
Um a Meta spokesman referred to the comments by Meta Chief technology officer Andrew Bosworth about the company working on many prototypes not of not all of which go into production.
15:24
Meta is working with Avatar director James Cameron's Lighttorm Entertainment on exclusive VR content.
15:30
The two companies announced a partnership last year.
15:32
So, I think we got to get on this.
15:35
We got to have a VR stream. Yeah.
15:37
Three hours of content every day.
15:39
We're going to be the reason churn is low on the next VR headset. Yeah.
15:43
Because you just throw this thing on it just like you're sitting here on the drink cam.
15:46
You can you can click through. I mean, yeah.
15:48
Just click all to the the different angles.
15:50
It's pretty It's pretty doable. Um it's pretty doable.
15:54
I mean, you can film valuable like like usable spatial video on just an iPhone now and then you can play that back in the uh in the Vision Pro and it does look 3D, which is cool.
16:03
Um, in other news, the Wall Street Journal is reporting that uh Reddit is suing Anthropic alleging unauthorized use of the site's data.
16:10
uh an online discussion forum a key access the site more than 100 thousand times saying after saying it had stopped Reddit is suing anthropic and and anthropic debates this I'm sure we won't be able to get into this today because it's I'm sure it's it's it's caught up in the courts and there's a whole bunch of legal restrictions but uh we'll do our best to understand how these deals come about.
16:32
seems like most of the time it's not that the company that has much data doesn't want the AI company to use their data.
16:38
They just want to have a an equitable agreement where everyone is uh getting the most value.
16:45
And I think surge the stock's up, right? Yeah.
16:48
And then for more context, OpenAI is already paying Reddit approximately 70 million per year in a content licensing agreement.
16:54
So, uh, they kind of got ahead of this issue and and decided to strike up an actual deal.
17:03
And I believe Google has a deal with them, too.
17:04
And I think this might be one of Google has a deal with Reddit.
17:07
Yeah, I'm pretty sure because there's that meme about like the best way to search Google is search like whatever your search term is and then space Reddit because the user generated content was better than the SEO stuff that was Google pays Reddit approximately 60 million per year. So 60 and 70.
17:20
So they're getting 130 million.
17:23
That's pretty serious revenue.
17:25
And and it's something that doesn't need to be brokered via a bunch of individual programmatic ads that might not work or anything like that or or subscale.
17:31
It's just one one one or two deals and boom, you're up in the hundreds of millions of dollars in revenue.
17:38
What is what is Reddit's overall overall annual revenue?
17:42
Uh $20 billion market cap.
17:42
I don't uh let me see here. How are they tracking? 1.
17:50
3 billion uh greenbacks in 2024. 1. 3 billion.
17:55
So, they're getting like 20% of the revenue or 15% of the revenue. Yep.
17:59
I wonder how big uh they grew 60% over over 2023. Interesting.
18:03
The So, they might be bigger.
18:07
They might be bigger than Reddit or they might be bigger than Kanye Nast, which at one point owned them.
18:13
It's kind of unclear how how valuable Kanye Nast is because they're private.
18:17
Anyway, um, so, uh, Reddit, uh, said that the AI company unlawfully used Reddit's data for commercial purposes without paying for it and without abiding by the company's user data policy.
18:29
Anthropic is in fact intentionally trained on the personal data of Reddit users without ever requesting their consent.
18:33
The complaint said, uh, interesting saying that it's about the users.
18:38
Yeah, Bills itself is the white knight of the AI industry.
18:40
Um, last year, Reddit took steps to try and limit unauthorized scraping of its website, creating a public content policy for its user data that is publicly accessible, such as posts on subreddit and updating code on its back end.
18:54
The user policy includes protections for users, such as ensuring that deleted posts and comments aren't included in data licensing agreements.
19:00
And so, yeah, I don't think that there's a really strong uh precedent for uh agentic for the agent for the agentic web.
19:07
Like if I like if I use Google Chrome to access a website, Chrome doesn't need to pay any sort of license.
19:17
But if I go to Anthropic and say, "Hey, get me up to speed on this topic and it goes out and it browses the web, all of a sudden it feels like maybe they do have to pay whereas Chrome wouldn't because it's just rendering the web page and it's not transforming it at all. What is transforming? What's what's fair use?"
19:33
And so these things will obviously play out in the court of law and so hopefully they can resolve it quickly and move on. Yeah.
19:40
I'm actually surprised that uh they didn't already have a deal in place. Yeah.
19:44
Because it's very valuable data.
19:47
You want that data for your models.
19:49
And anyways, well we have Sean Magcguire joining in just a minute.
19:54
And um um um the other news in the Wall Street Journal today is Thrive Holdings is betting that AI can change IT services.
20:07
The company established by venture capital firm Thrive Capital uh joined with ZBS to invest $und00 million into an entity that will integrate AI into IT firms.
20:15
This from Josh Kushner, of course.
20:17
Um Shield Technology Partners has already acquired four IT service companies.
20:22
Clear fuse networks, iron orbit, uh, delvall technology solutions and onenet global.
20:27
It said thrive holdings called shield technology partners and AI enabled managed IT service platform.
20:33
IT service companies also called also called managed service providers or MSPs typically provide IT support and managed tools like software and cloud computing on behalf of businesses.
20:43
Founded by Josh Kushner about 15 years ago, Thrive Capital is known for some of its high-flying startup investments including OpenAI, Data Bricks, and Whiz. What a what a portfolio. Not bad.
20:53
Investing in traditional services business, particularly those that re rely heavily on administrative knowledge work and adding AI to supercharge them is becoming a bit of a trend.
21:03
As part of its effort, Shield technology partners will embed software engineers into each of its IT portfolio businesses.
21:08
Oh, they're doing the forward deployed engineers.
21:12
The engineers goal is to build an AIdriven solution that all of the portfolio companies will use.
21:16
We've studied all the ways in which MSPs have perhaps been on their back foot to date with customers um and and says that IT services work is incredibly well suited to what AI can streamline.
21:27
And so you can imagine a whole bunch of agentic workflows for all the different things that you need to do when you're deploy when you're deploying cloud um and managing cloud.
21:36
Really quickly before we have our next guest, let's tell you about Vanta.
21:39
Automate compliance, manage risk, and prove trust continuously.
21:43
Vant's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation whether you're pursuing your first framework or managing a complex program.
21:51
If you think you should be on Vant, you're probably correct.
21:55
Well, we have Sean Magcguire from Sequoia Capital in the studio.
21:59
Welcome to the show, Sean. How are you doing? Boom. What's up, team?
22:00
Never a boring day on the internet, that's for sure. Yeah.
22:05
Well, what is keeping you uh Oh, man.
22:08
What's keeping you up now?
22:09
Well, obviously I Well, I I think there's you guys any anyone on Twitter knows what I'm talking about. Yeah. Yeah. Yeah. Or an X. Uh yeah.
22:16
I mean, let let's let's skip the politics because this is purely a technology and business show. Thank God. I love you guys. You're the best. Stick to the technology.
22:27
Uh what uh I mean, we we we've had an interesting experience with X and that uh there's always been this narrative that like the whole the whole platform was going to collapse.
22:37
We, you know, there's been rough days here and there, but overall things have been growing.
22:42
Um, what have you seen across the XXAI merger?
22:46
What are the secrets to success?
22:49
Um, how is uh, you know, talent tracking?
22:52
Is any of is any of like the chaos and noise distracting?
22:55
Because when I talk to XAI engineers, they're like, we're too busy.
22:59
We can't come on your show.
23:01
But but but what's your experience been with the the X and XAI team recently?
23:08
Look, in if you go back in time, as you said, everyone said it was going to fail, the app would crash, you know, nothing would happen and that didn't play out.
23:17
Um, but there was a lot of tech debt and kind of broken infrastructure and and there was a, you know, a couple years of rebuilding the basics and foundations.
23:29
I think we're starting to see, you know, real innovation happening.
23:34
I love the Grock integration directly in X.
23:36
It always scares me when someone, you know, when I have a tweet or whatever and then someone says like at Grock, is this correct? Is this real?
23:42
Like, is is this accurate?
23:45
You never know what's going to come back.
23:47
You know, usually usually I agree with Grock.
23:50
There's been once or twice where I think some of the subtleties are a little off, but uh it's truth seeking.
23:56
It doesn't mean that it's fully truthful every time, you know. Yeah. Yeah.
24:00
It hasn't actually found that ground truth every single time. That's funny.
24:03
What what about the overall I mean I think in reality I'm probably wrong but Yeah. Yeah.
24:06
Uh what about the overall horse race between the foundation models?
24:10
Uh it seems like every day it's going back between uh an OpenAI launch, an anthropic launch, a Grock launch, a Google launch.
24:17
Um are you do you think that continues?
24:21
Do you think there's like maybe some fragmenting and there's opportunity?
24:24
I mean, we're kind of already seeing this with how much anthropics loved by developers versus uh OpenAI has been really dominant on the consumer side and now every company is figuring out a different way to actually get to distribution.
24:38
What really matters here? Is it pure scale?
24:40
Is it pure cracked engineering talent? Is it distribution?
24:45
Is a combination of those things?
24:45
How are you seeing it play out? Great question.
24:49
Great question. I you know honestly my opinions have changed a lot over the last few years and in in many directions and so I don't have too much confidence in my assessment right now but the you know I always try to look at lessons from the past and my current thinking is
25:07
that the closest analogy are operating systems and if you and I'll make a couple points on this if you think about operating systems first of all there's a bunch of different ecosystem there's the Windows ecosystem There's the Apple, you know, OS ecosystem. Then there's like on
25:21
Then there's like on mobile, there's, you know, Android, there's, you know, a whole browser environment with Chrome.
25:30
And then there's open source, a Linux.
25:31
You know, one thing that I think is interesting about Linux, you know, there's there's more Linux servers in the world than there are Microsoft servers, but the value kind of capture of Microsoft is way greater than Linux.
25:47
Um, I personally think we're going to see something very similar play out where there'll be like a, you know, OpenAI will, you know, be the Apple or someone and, you know, XAI I think will be very successful.
26:02
I think there's a good chance that Anthropic is independent and successful.
26:07
independent and successful. I also think there will be a big open source component which would be like Linux and I think there will probably be 10 to 100 times as many open source models out there um like or like deployments of open source models in 10 years but I
26:24
think that they won't be as valuable and they won't be like as rich of ecosystems and then just to make two more points on the open source analogy like for Microsoft by having or Apple by having the operating system you know they were able to actually win in quite a few ways on the application layer as well. You
26:41
You know, for Windows, they bundled in, you know, Word and Excel and, you know, then Outlook and all these other things.
26:49
I think it would be very similar for the foundation model companies.
26:51
I think that the foundation models would be like table stakes.
26:55
That'll be their kind of win, but also a very sticky moat.
26:57
And even if they're not the most profitable businesses themselves, it will give them big advantages kind of on the application layer.
27:06
And then one other thing that I think will happen um you know the cloud companies have giant moes just through the capex dynamics of of cloud like needing to buy all this hardware and you know innovate with hardware and stay there is a big moat.
27:23
I think these foundation model companies are going to be I think there's going to be way more value that occurs and there'll be way bigger moes than people realized.
27:33
I think they'll all basically have hardware modes like cloud style hardware modes.
27:37
They will have the like the operating system style, you know, very very detailed research that's hard for anyone to replicate and then I think they'll probably make a lot of their profit from applications on top of it.
27:51
That's that's my current thinking.
27:53
Thinking can change inside fingers.
27:53
So, I know obviously you weren't uh investing during the original operating system boom, but your firm Sequoia Capital was.
28:03
And so, uh have you had any discussions with the with with the kind of the lineage of the firm or the history um and and seen how is the revenue ramp or the business scale different this time than say in the dot era or in the previous era?
28:21
uh it feels like it's ramping faster than ever.
28:24
It feels like we're seeing more companies that are hitting a billion in revenue or a hundred million faster than ever.
28:30
But is that real based or or anyone that you've talked to that was uh investing in that era?
28:37
Did does it feel different this time around, do you think? Yeah.
28:45
I mean, one of the beautiful things about being at Sequoa is we do have this long history and we get to tap into the kind of institutional knowledge.
28:49
You know, that said, sadly, Don Valentine died four or five years ago, like early into my time.
28:55
RIP, what a what an absolute legend, you know, and he led the original Apple investment. Yeah.
28:59
Um, but there's still a lot of Google institutional knowledge in the in the firm, which is, you know, not directly operating system, but they created an operating system later.
29:11
Um I mean first of all the revenue of these companies is scaling insanely just faster than any products in history before um for Starlink.
29:24
So obviously not a foundation model company but I basically made like internally I I I made an Excel spreadsheet of AWS's revenue growth like in the first 20 years of AWS compared to Starlink and you know Starlink has in five years gotten to where what took AWS 10 years to get to.
29:44
Um and and I and now like with these foundation model companies we're seeing as fast or even faster revenue growth.
29:52
Um you know that said these are very I think the business models like the initial business model is more clear and the profitability of these companies is in you know the more the inprofitability is insanely high and so you got to discount the revenue growth.
30:09
Um, but I I would just say the biggest lesson I think from the past is you have to capture like territory early on and and the doors will kind of close behind you because of these capex dynamics and uh and just like lock in with users. Yeah.
30:25
I mean, you mentioned Starlink.
30:27
Uh do you think there's obviously Yeah, it's such a weird company because it's like a space launch company that now is an internet company ISP.
30:35
Uh, but there's actually a little bit I'm starting to hear of an AI narrative just that having Starlink potentially unlocks edge compute or inference in areas that would typically have kind of stranded energy resources.
30:52
So all of a sudden if there's some super remote area that has really cheap energy, you can go and and set up a data center there and then do inference and stream those tokens over Starlink.
31:02
Do you think that's an underrated narrative?
31:05
Do you think that that's developing on course?
31:06
Do you think there's any bottlenecks that people should be thinking of within that story?
31:14
So when we first invested in SpaceX, the part of the core thesis was internet everywhere.
31:19
Um and and I would say like it goes way beyond AI.
31:22
Uh but I I think the internet everywhere thesis is is huge and that will be you know everything from oil rigs you know to airplanes to boats to yeah edge AI devices.
31:37
But like the I think the bigger thing for Starlink is Starlink just has like a 10x plus cost advantage for moving data compared to you know building new transatlantic or trans-pacific fiber lines.
31:51
And in the world of AI we're going to be moving these models are going to be moving so much data around themselves.
31:59
Um, and I just I think Starlink is incredibly well positioned to be the pipes to move all this data for AI.
32:07
Um, and so that I actually I care more about that just because of the volume than some of the kind of edge applications for AI specifically, but those will be big.
32:19
And then one other thing I just got to give a plug to to Bitcoin. Um, bas basically. Yeah, let's go.
32:31
Basically, three three years ago, I visited the biggest Bitcoin mine in the world, Genesis Digital.
32:35
Their mine is near Midland, Texas.
32:38
It's actually backed by SPF, which is, you know, wild.
32:44
He he got both a bunch of good bets.
32:44
You cannot Yeah, exactly.
32:46
They got both Anthropic and and Genesis, but these guys had a gigawatt scale Bitcoin mine operating three years ago.
32:54
And already for them, like having it it taught it taught me a lot.
33:02
And you know, Bitcoin, like Bitcoin mining is the absolute tip of the spear where you need the least amount of data movement, like data in and out to dollar um generated or or like power consumed.
33:20
And so I actually think that was like Bitcoin mining is underrated in terms of how much it's pushed like frontier power generation um turned into compute.
33:30
And I I don't think it's a coincidence that Crusoe, you know, which is now powering Stargate, started off as a Bitcoin mining company, or that Cororeweave, Cororeweave, which is like 80 billion dollar stock as of yesterday, is now um you know, is now an AI data center company.
33:49
And I just I think I think that's honestly the bigger the the bigger theme. Yeah.
33:53
Uh what's your updated thinking around nuclear?
33:56
We had these new executive orders and it was announced this week that Meta announced a partnership with Constellation to power some of their AI power needs.
34:09
What's your kind of updated outlook over the near term to medium term?
34:11
I'm an all of the- above guy for energy. Like we need all of it.
34:16
We need all of it as quickly as possible.
34:19
I as an individual invested in a few nuclear companies going back like nine 10 years ago way too early.
34:27
Um and and I like to have put a little bit more meat on these statements.
34:34
Nuclear is incredible but deploying large amounts of nuclear is slow.
34:40
Like even if you deregulated it to zero, I think it would be like a you know a more than a decade, well beyond a decade to deploy like a terowatt of new nuclear.
34:50
Call it 10 years if you did it as fast as possible for America starting now.
34:56
Solar is just a way faster way to deploy a lot of energy.
35:01
Nat gas is a way faster way to deploy you know to deploy a lot of energy.
35:07
We have been producing insane amounts of natural gas which we didn't have the pipelines to actually use.
35:16
So we were just flaring it a lot of times because kind of like the the dollar value per like when you have an oil well or you're fracking it's producing it's emitting natural gas and and oil and you just made so much more money from the oil than the natural gas that we didn't really care about it and that started to flip.
35:36
Um, and so anyways, I I think we have to do all these things.
35:39
I think we need more natural gas, more oil, way more solar, and then kind of have nuclear coming as the reinforcement juggernaut coming online like 10 to 15 years from now. That's a good framework. Fantastic.
35:55
I mean, we have to have you back for, you know, an energy deep dive.
35:56
We we know a fair amount of the nuclear and solar entrepreneurs, and there's a bunch of people doing really cool stuff.
36:02
So, uh, have a safe trip. Personal plug.
36:04
I had a seat in the New York Mer Merkantile Exchange. Yeah.
36:07
When I was like 22 years old. It was It was insane. Wow. Cool.
36:12
Um, hey, good luck on the timeline today.
36:15
I know you're going to go in there.
36:16
Put on your put on your hazmat suit and just get in there. Good luck. Good luck. Peace, guys. Safe travels. Cheers. Fantastic.
36:22
Uh, let let me tell you about Linear.
36:25
Linear is a purpose-built tool for planning and building products.
36:28
Meet the system for modern software development.
36:29
Streamline issues, projects, and product road mapaps. Go to linear. app.
36:34
Next up, we have Jack in the studio.
36:36
We have an in-person guest. Let's bring him in.
36:40
Play some soundboard for me, Jordy. Welcome to the stream. How you doing, Jack? There he is. Second time on the show. Good to have you here.
36:49
Um, what are you wearing today, Jack?
36:51
Wearing the jacket, the TDN jacket in the in the capital of capital. There you go. Thanks for coming. Thanks for hanging out.
36:58
You can adjust your mic a little bit there as well. Cool.
37:01
I I wanted to uh kick this off with like a little bit of a rundown on the different foundation labs.
37:08
We're talking to a lot of them today and I and I noticed that uh Jordan Snider from China Talk and Dylan Patel went ran through their AI mandate of heaven tier list.
37:18
Uh and so I wanted to read through that and kind of get your reaction and then just kind of do like a vibe check and let it and and talk to you about what we should be expecting from different labs over the next year.
37:30
It's a little bit of a horse race.
37:30
So, uh, up first at S tier, they have OpenAI.
37:35
It's the only foundation lab that made S tier.
37:37
Does that feel right to you?
37:40
What are you watching from Open AI?
37:42
Yeah, I I think that's exactly right.
37:44
OpenAI executing both on the product level, getting the distribution, getting into hundreds of millions of people's phones. Yep.
37:50
Um, but also also on the research level, you have people like Nome Brown, people like Aiden just doing this incredible frontier research.
37:58
03 I think just as a model impresses me the most of any model that's come out so far.
38:02
Know Brad Lap said in the Wall Street Journal recently they had two million um workplace users in February and they're at 3 million now. Wow.
38:09
Um so just just really exceptional growth.
38:12
I think I I was I was thinking earlier it'll be funny uh our kids in in 20 years will be like dad they're making me use OpenAI Teams at work.
38:24
It'll just be like like the default like the Microsoft Teams default.
38:26
Yeah, I mean there's a little bit of a narrative that uh that maybe and and we can move on to Anthropics in the A tier alongside DeepSeek and Google.
38:35
Uh there's a little bit of a meme that like Anthropic is crushing it with developers.
38:39
They're the default choice for wind surf cursor users, but then OpenAI is more dominant with consumers.
38:46
But I I feel like recently I've heard that it's maybe even more skewed than people think.
38:50
Like it it's it's maybe not like the the vibe on X might be Yeah.
38:54
like you know 7030 open AI clawed for daytoday grab a random person on the street but it might be even more skewed.
39:03
Does that feel right to you?
39:05
Yeah, I think I think anthropics really solidified with developers but has like totally given up on consumers but I think OpenAI wants to take that on.
39:12
I mean there there's rumors about some sort of wind surf acquisition. They're releasing 4. 1 and codeex.
39:18
They're pushing hard on coding and I think that's something to watch from them this summer and going into 2026 is is can they can they secure that?
39:23
Do you understand the model names at this point? 4. 1 I I have access to 4. 5.
39:27
Why would I want to go backwards?
39:30
Is that is that is are the models fragmenting to like where I'm going to have to learn a new a new taxonomy for okay, if I want to write code, I use this one.
39:40
If I want to write poetry, I use this one.
39:41
If I want to do uh math or reasoning or build a chart, I use that one.
39:45
Because it's putting more work on me, I feel like I think Sam said that they're going to try to fix the model naming scheme this summer.
39:50
That's the real thing to watch if they're going to keep S tier is can they can they get coherent model names but yeah 4.
39:56
1 it's cheaper it's specialized towards coding it's kind of their 3.
39:59
7 type of driver at the same time I I know you're not super up to speed on the Alibaba like Quen models but I I saw some I saw some release where Alibaba Quen released like a hundred different models and Will Brown was kind of saying like this is awesome from a research perspective because they have like they have like um one model that's just good at bio and It's kind of like this hyperfragmentation.
40:21
It's the opposite of going in the unification direction.
40:24
It's actually it's actually going more specialization and then maybe you unify that at the end. But I don't know.
40:29
Uh it seems like if you're a consumer company, you can't you don't really have that affordance, right? Yeah.
40:34
I think in terms of research, Alibaba's a bit underrated.
40:35
I mean, compared to Deep Seek gets all this press, all this coverage, but the Quen models are really good.
40:41
People are doing you you see from lots of people these really cool RL experiments, these really cool kinds of things.
40:46
They're they're lagging behind the US models.
40:47
They're not they're not A tier.
40:48
They're not Btier, you know, but they're doing some interesting stuff and I I I think that's super cool.
40:53
Yeah, I really wonder if they're I if they have a distribution advantage in China.
40:57
Obviously, we wouldn't feel it here, but uh I I really haven't gotten up to speed on what is the Chat GPT of China in terms of distribution.
41:06
Obviously, Deepseek had that moment, but have they actually executed properly on the on the product side? I don't know.
41:11
I'm surprised that Google hasn't been able to turn their their general distribution advantage into an AI distribution advantage.
41:16
They have these really good models.
41:18
The new Gemini came out today.
41:19
It's It's got really good benchmarks on a lot of things, but they're yet to I I think they're yet to crack distribution.
41:25
We we sometimes say, did you see that?
41:26
Did you see that mockup that was just the Google search box, but a Gemini prompt?
41:30
It was like if they wanted to go full send, if they really if they were really AGID, they would just say, "Hey, we're done with Google search."
41:39
I mean, it would destroy their economics. I'd commit to it. I'd commit to it.
41:42
I think uh it looks like the model that they use to power those search prompts right now.
41:45
It look it seems really lightweight to me.
41:48
It gives a lot of wrong answers. When ask 2.
41:49
5 something, it's always right. Sure. That's interesting.
41:52
You think it's just AI overview box is like we're just going to hallucinate hallucination box.
41:59
Well, they are launching like the uh like advanced AI search, but it's like a toggle, so you have to find it, which is like always the problem with Google.
42:05
Well, I mean, they still wound up in the A tier according to Dylan Patel and uh and Jordan Schneider over at China Talk.
42:11
uh uh obviously V3 was like a huge one and then they also have all those like priced performance things but uh I I've heard this narrative that like maybe some of the hyperscalers are super focused on benchmarking and and not even hacking the benchmarks necessarily but just like just thinking about them and a lot of the frontier labs the independent labs have just kind of moved on philosophically from caring about benchmarks. Is that the right move? What's driving that?
42:38
like is it is it are we in like the postbenchmark era essentially? Yeah.
42:43
When I when I think about models and benchmarks a lot, I think like which models outperform the benchmarks, you know, when you see 03's benchmarks, they're good.
42:51
They're kind of what you expect.
42:52
Then when you watch 03 think, you see this model, it's actually reasoning. Sure.
42:55
When you watch Cloud 4 Opus or Cloud 4 Sonnet think, it's like, whoa, this is really good. Same with uh GPT 4. 5.
43:01
I think the Gemini models are good, but they're exactly as good as the benchmarks let on, you know, and I I I think they don't have the vibes yet.
43:08
What I want to see is Gemini 2. 5 Ultra.
43:10
If Google releases something with some big model smell, something really cool. Maybe maybe that's them.
43:16
What is What is the big model smell?
43:16
I just don't like the idea of smell at all. Oh, yeah.
43:20
It's just a weird It's a weird sense.
43:22
It's just It's a vibe check.
43:24
Wait, who who who coined it?
43:24
I think it was Aiden Mcllin. I don't know. That's great.
43:29
Um uh but but basically we're we're in like the intangible period.
43:34
Is that is that the idea? Unquantifiable.
43:37
I think I think Anthropics give it up on really training on the benchmarks and I think it's done really well for them.
43:41
You know, you see that they're really good at Swbench.
43:44
They're not crushing it on MMLU, you know, but you tried force on it. It's great.
43:47
You know, um other labs that are lower down on this tier list seem to have seem to have not given up on doing really well on the benchmarks. Yes. Yes, that makes sense.
43:55
I mean, it's possible that like you you must defeat the final boss to like play the end game. Yeah.
44:01
And so maybe the end game is this vibe check, this big model smell, but the the in the interim like yes, like if you're not you only earn the right to go into big model smell if you can dominate in all the benchmarks.
44:11
There was an interesting moment where um 3.
44:12
7 was beaten on every benchmark.
44:15
So now now for state-of-the-art on stuff again, 3.
44:17
7 was losing on everything.
44:19
There was a better model for everything hypothetically.
44:21
But then if you looked at um what what you might call like uh revealed preferences bench which is just like what do people use on cursor what what's going on ramp revealed preferencesbench.
44:32
com yeah someone was great um yeah yeah 3.
44:34
7 was was pretty high up there you know so it seemed like they had something that that wasn't captured there what about cornered resources data is the new oil that seemed like a very silly concept in the moment when everyone had scraped the web entirely and there It really felt like data was fully commoditized.
44:52
Then we see VO3 and for the first time it feels like okay there is at least one data set that is so large that you can't copy it onto a single hard drive or compress it and it's YouTube and Google owns it and and yes people might scrape it here and there but Google has a durable advantage there.
45:11
But is that is that the wrong way of thinking about it?
45:15
Yeah, I mean I'm not sure about the video models.
45:17
I I I think it's true that data is like both super super important but also has just become like tremendously overrated because the first people thing people learn about AI is like oh it's a result of the data that goes in but now that we're unlocking things like RL embedded post training it seems to me like you can you can have some non-data solutions to some of these problems.
45:35
problems. Yeah, I mean that was the original what generative adversarial network for image generation was like synthetic data generation and then and then uh and then testing it and so like I'm I I it just V3 feels so so much like a beneficiary of YouTube but I don't know if that's just if we're just
45:52
waiting and we'll see the next Sora and we'll be like oh openai figured it out and like yeah maybe they found some like you know kind of workaround to the data but really like the the vast majority of the consistency and the innovation there was algorithmic progress, not just, you know, quartered resource and data. Yeah. Yeah.
46:09
One thing about video models, it's been so secondary, but they've become so impressive.
46:13
I think that if you showed them both to me a couple years ago, I would be more impressed by V3 than even like Cloud Force Sonnet or something, you know?
46:20
Um, it's just it's not what I it's it's really really just incredible. Well, yeah.
46:25
I mean, I I think a lot of it just comes down to like the cost of instantiating the thing.
46:28
And so if I go to if I go to deep research and I use 03 and I have it pull together um some you know 20-minute research paper it's like that's a few hours of work.
46:39
Maybe it's a few thousand of like a researcher's time.
46:43
Maybe we're getting up into like PhD level. I could do it on my own.
46:45
But but if I actually want to crash a Ferrari through the Hollywood sign with champagne bottles flying a custom Hollywood sign that's huge.
46:56
Like unless I'm I'm either doing chasing.
46:58
Yeah, I'm either I'm either renting all that, shooting it practically and it's a multi-million dollar Michael Bay shoot or I'm doing it all in CGI.
47:04
And even to do it in CGI is millions of dollars of rendering.
47:08
And so even for an 8-second clip, it just looks like wow, I got something that normally would cost a million dollars to make happen.
47:13
And there's no there's no real like textual asset that feels like, wow, this is a million bucks worth of assets. Anyway, interesting.
47:20
Uh XAI, uh they are cooking.
47:23
They've been uh obviously GPU rich, scaling up.
47:26
People seem like they're in the B tier here uh according to this chart, but everyone's kind of excited about what's coming next.
47:33
Uh what is your take on Grock XAI?
47:35
Are they close to the big model smell?
47:38
Uh is that feels like a natural bene beneficiary of Elon strategy of just go big, but how are you thinking about Grock generally?
47:48
Yeah, I I'm not the most impressed yet. I mean, Grock 3 is good. It's a good model. Sure.
47:53
Um, it's like a funny thing like Grock's whole thing or or something that people who really like Grock often say it's like, "Oh, it's it's trained on this realtime X data as this Xvalibility."
48:01
One thing I've tried a few times because I saw in a tweet is if you have a tweet you can describe, maybe I say like John Kugan's tweet about bringing media back to Hollywood.
48:08
Um, and you ask Glock to find it, it can't find it.
48:12
You ask to find it, it can find it. Wait, really? 03 can find it.
48:14
That's so interesting because I feel like I feel like X is pretty locked down at the at like just the www layer, right?
48:20
It's pretty hard to find.
48:23
In fact, a lot of times I'll post in a a post from X and it will have to go to like thread reader unroll and find an archive off of X because it clearly can't access it directly.
48:35
But that is that is fascinating.
48:37
So, uh that feels that feels solvable.
48:39
Adam Adam shipped tbpnest. com.
48:42
Y like last week I had a friend find it Monday.
48:45
We hadn't announced it anywhere.
48:47
It's not even visible on the Google uh Google search and 03 found it. Wow.
48:49
I was like, "How did how did you find this?" And he was just looking.
48:53
He asked 03, "Can you pull together a list of all the guests and it found that link randomly and Google doesn't even find it." Interesting.
49:02
03 is really good at search.
49:04
And I think I think that might have been RL.
49:06
They they mentioned Relling on tool use in the blog. Very very interesting.
49:09
Also like XAI, it's like not really much revenue, nearly no revenue yet, you know, at some point you need to start building that out.
49:15
I'm glad they're they're pushing on the distribution, you know, but yeah, things come around. Makes a lot of sense.
49:20
Uh last one we'll end with the highest revenue multiple of any company in history. Yeah. Yeah.
49:26
Uh last one we'll end on Meta Lama sitting in Dtier but maybe not out of the game yet.
49:31
Uh the two interesting bull cases I've been discussing have been one um is there a world where uh open source American AI becomes geopolitically important for countries that are slight allies and they're either choosing between deepseek or an open source uh American model and they and open AI would not be in the conversation.
49:54
Um and then also just you know why would you ever bet against Zuck?
49:58
He has a capital cannon that can fire 10 billion at random projects forever.
50:04
The question is, is that enough?
50:04
Uh what are you looking for from Meta and Llama in the future?
50:09
Yeah, it seems like you hit some some issues recently, but I I I'm not betting against Zuck. He's got the capital. He's got some GPUs.
50:14
Um they can get together some really great research.
50:19
I would love to see better American open source models.
50:21
I mean, I'm not betting on open source in the long term as maybe the cornerstone of AI, but the fact that all of our American um research groups, lot lots of really smart RL researchers are doing experiments on Quen and not on Lama Lambda is not great, you know. Yeah. Yeah.
50:37
So, so should there's one interesting twist there, which is Quen has so many different models. Uh Llama has a few.
50:44
They're still working on rolling out behemoth, but uh would would it be like almost more of an olive branch to the developer community to fragment the models and and and really focus on hitting researchers?
50:54
Is that kind of a a a potential path that they should take? Yeah.
50:59
I mean, I think it would be really cool if they did that.
51:01
It would be somewhat charitable. Yeah. Yeah. Yeah. Exactly.
51:03
Developers love a handout, but you know, I don't know.
51:05
I I think I'm curious about what they do on the product level and how how they can build stuff in better.
51:09
On the product level, people aren't incredibly sensitive to whether 03 can search 50,000 websites like we are.
51:14
You know, they care more about just having something that's really good, something that's really good to talk to.
51:18
Maybe meta shifts focused that anymore.
51:20
I'm not feeling it right now in terms of like when will a meta model grab number one on LM Marina or something.
51:26
It seems like it's it's going to be some time, you know, but I'm not counting them out at all either.
51:31
Yeah, I mean, if they can just Yeah.
51:32
stay on the the lagging edge, that could still be valuable in a lot of their product rollouts.
51:37
Um, I mean, we forgot Apple in the L tier.
51:39
Uh, we we do have another guest hopping on in just a minute, but uh, Apple in the L tier.
51:43
How do they dig themselves out? Is it build?
51:48
Is it buy What do you think is going to happen?
51:50
They could maybe they have a lot of cash.
51:51
They could maybe buy someone. They could buy someone.
51:52
They could buy buy a lab and then then you got to Yeah. got an upgrade.
51:57
Um, I I there was a there's some report that they had some internal models.
52:01
Um, I wouldn't be surprised if they could train stuff.
52:05
It's just look, we haven't seen anything at all.
52:07
You know, like do you think they're really training on Apple silicon?
52:10
Like you've seen those photos of like all the Mac minis wired together.
52:13
Does that seem like something that's real or just like AI generated photo? Okay. Yeah.
52:16
Um I think non GPU training ones are going to be bigger in the next few years.
52:21
Well, I think the TPUs for Google, too.
52:22
Um yeah, so they already have a long time of TSMC.
52:24
They could go do something like a tranium or an inferential chip from Amazon or TPU. Yeah.
52:30
I mean, with the TPUs, Google has by far the most comput. Yeah.
52:31
I mean, I guess Apple's pretty good at chip development design.
52:34
So like, yeah, one chip is a pretty good idea.
52:36
That would be their Yeah, that would be their their their advantage if they could build a really strong chip and cut that cost.
52:40
I wouldn't bet on it, but maybe. Yeah. Yeah.
52:44
I I like the idea of just opening it up and really partnering.
52:46
The the thing over the last 24 hours is one account sharing it's so over for Google and and then immediately sharing, wow, Google's going to destroy everyone in AI.
52:57
And just like seeing seeing how the the posts rank. Yeah. Yeah.
52:59
Um anyway, anyone else on here?
53:03
Uh they got Mistral in F tier. Poor for the French.
53:05
Yeah, they're not trying leat.
53:08
Yeah, you got to be I I I do wonder about Mistral because you know the the models are real but none of like broken out in capability, but there there's this question of like if you want a national champion in your country, it might not be enough to just have the foundation model layer, you also might have to go and win in the free market in the application layer.
53:27
And so yeah, you could have even if you had a comparable model, if you're not if that's not if people aren't if people are going to to chat. com instead of lay chat.
53:37
com like you have not won and you don't have your national champion. Yeah.
53:42
And there's a I think there's some truth to this, but there's also the the regulatory stuff in the EU.
53:48
I mean, a lot of releases I think V3 is not in the EU.
53:50
A lot of releases don't come there. Um maybe Mr.
53:54
just uses uses regulatory modes to monopolize.
53:56
Not not a fun way to win, but maybe that's the bull case at this point. Yeah.
53:59
Uh what what what was your reaction to uh the conversation back and forth with uh Darkh and Schlutoto all about um about uh the the the the debate over over uh I forget it was like spiky intelligence and how you actually uh train someone.
54:19
There's so many different things.
54:21
We see that the models are really good at one thing and then they fail. RKGI.
54:24
Um, uh, what's your overall timeline right now? How are you looking? Yeah.
54:29
Dash raised the point that, um, you can't kind of do this continuous learning, this like short run continuous learning like you can tell me, Jack, I want you to do something different asp and and context is a weaker tool than that.
54:42
And I think that's absolutely true and that's an unsolved problem.
54:44
I don't know how much that moves my needles on timelines.
54:47
Like one thing that could be true is just that open or anthropic makes some like agent and it starts accelerating their AI research and they just get like really efficient algorithms really quickly.
54:58
Some architecture that just destroys the transformer. Yeah.
55:01
Um but I do think it's a meaningful unlock if you could if that could be solved.
55:06
Um and I think that sort of like midlevel memory type of stuff is really interesting.
55:10
Um or solutions around context around a rapper.
55:15
Well, this was fantastic. We have our next guest.
55:16
Thanks so much for hopping on. We love it.
55:18
We love an inerson guest. For sure. Thank you so much.
55:23
Next up, we're we're we're heading over to Google World.
55:24
We have uh Arouch from Google.
55:27
He worked on the deep research project that dropped from Google in 2024. It was a full year ago.
55:34
It was in December technically, but uh very excited to talk to him about uh that product uh all the things that go into deep research.
55:41
So, we'll welcome him to the studio if he's available. How you doing? Good to have you. Hey, what's up guys?
55:46
Thanks for having not too much.
55:49
Uh we're having a great day.
55:49
We got a great lineup and uh excited to dig into it.
55:53
Would you mind kicking us off with just an introduction on yourself and uh and uh a little bit of the history.
55:56
I I want to hear about the history of of the products that you've built at Google uh what the interaction between uh research and product looks like and what you're excited about. Yeah, for sure. Uh first off, a team. That's pretty good. Pretty good.
56:13
Yeah, let's hear it from there. Let's go. Let's go. Let's hit it. Yeah.
56:18
John's going to hit the good one. Cool. Um Yeah. No, love to be here.
56:25
A lot of um Yeah, it's been fun. It's been a fun ride.
56:30
I So, I'm a product manager on the Gemini team. Cool.
56:32
Um I've been here since uh a little while back when it was called Bard. The Bard days. And Bard days. Um yeah.
56:40
And so yeah about I don't know maybe like this time last year we started kicking around this idea of deep research um where one of the things we noticed is a ton of people come to the product and ask like seeking to learn something or asking questions and kind of doing researchy type type things.
56:57
But if you ask really hard questions one things we noticed is the model would just give you like an outline of an answer.
57:05
It wouldn't actually tell you something very comprehensive.
57:06
Um so we kind of just ran with a hypothesis of like um let's take off the constraints of like it has to respond within a few seconds.
57:16
It has to use this much compute like let's let it let's just see how far we can push what the model can do.
57:21
And this was before thinking models or anything and then like kind of and any of that that good stuff.
57:27
Um and so we kind of worked on this idea for a bit and then we launched in December back on Gemini 1.
57:34
5 Pro uh was the model that we were using back then.
57:36
Um uh we launched deep research as kind of um a bet to just see like would people be into something that makes you wait 15 minutes but gives you something comprehensive.
57:47
I'm happy to wait although I do want it to speed up.
57:49
Um uh questions about uh context window size.
57:54
How important is that million token context window?
57:55
That feels like it's been a unique Google feature for even longer than I expected.
58:02
that the advantages in AI seem to last days maybe weeks uh before another model comes out that that you know meets or is roughly around the same capability.
58:12
Uh how important is large token context windows in uh in deep research like products? Yeah, it's huge.
58:20
It's it's it's like really what enabled us and kind of gave us the confidence that this was even worth trying. Yeah.
58:26
Um, I'd say that the long context enabled us to do basically be very recall forward and really cast a very wide net as we researched the web and try and find gems of information that that we then stitch together.
58:40
Um, and so that that was like I think our biggest differentiator and really allowed us to build this product.
58:47
Um, the other thing that long context allows us is like once you finished your research, not just the report but everything it read along the way is in context.
58:54
So you can keep asking questions going deeper uh with the with uh within Gemini and even if it's like a tidbit of a fact that's not in your report if it's in if it's been read at some point. Yeah.
59:07
It'll be able to retrieve that and give you that answer.
59:10
So so it also helped sort of beyond that first turn keeping a good experience. Yeah.
59:14
Um and then reasoning models was like the next big big step jump for us allowing it to then do more critical analysis.
59:20
So in terms of like actual product design, I'm interested in in the direction this goes here.
59:25
You could see one world where uh the models are baked down into silicon, everything's running even faster.
59:32
You're distilling the models and all of a sudden I'm getting a deeper a 20minut product in two minutes or even 20 seconds.
59:38
Uh, you could also imagine a world where what's possible if the economics work such that I could request a two-hour research report or a two-day research report.
59:50
Um, how are you evaluating those?
59:53
What would you personally be more excited about and what do you think users actually want?
1:00:00
Because stated preferences and revealed preferences are are often different or do we wind up with both? Yeah.
1:00:06
So one of the things that we noticed uh one when we launched this we had no idea people would be willing to wait like every metric at Google from the day it started is reduce latency and like all metrics go up. Yep. Right.
1:00:16
So this was definitely a bet where we were like a lot of people thought we were crazy um where we're like it's we're just going to take a ton of time and people will wait. Yeah.
1:00:24
Um one thing we noticed is that like after about a minute or something like that people are fine.
1:00:29
Like people will go away do other things come back we'll send them a notification when it's ready.
1:00:33
So the big pleasant surprise for us is like people don't mind waiting I'd say.
1:00:36
Um so in terms of like efficiencies gains um one of the things that we're more excited about is like okay if we can make models more efficient instead of reducing down the research time can I give you just a way better output like can I use can I bank that savings and give you something way more insightful way higher quality.
1:00:56
Yeah, I'd say the other thing is like even if I could give you like a a deep research answer in 15 seconds, it's going to take you 15 minutes to read.
1:01:05
So there's also an aspect of just like how much do you want to consume this, right?
1:01:09
Um so so for us, we're not as stressed about like can we make this faster, can we make this quicker?
1:01:16
I do think there are probably other points in the like latency comprehensiveness spectrum that people might like, right?
1:01:22
We picked like one extreme of like let's just go super hard and and build the most comprehensive long you know um uh uh thing that takes a while. Yeah.
1:01:31
Um but there might be totally other points people are Yeah. Yeah. Yeah.
1:01:34
So sometimes I notice I've generated like so many various deep research reports across all the different apps that uh I'll like I'll follow it up with a prompt like okay like yeah boil that down for like 10 bullet points because like I don't have time to read that and then I'm like wait like maybe I should have just asked it to give me 10 bullet points and the and I just like burned a bunch of GPU cycles.
1:01:52
But I guess I guess the question back and forth between the two until you kind of understand the subject exactly.
1:01:58
Um but but I guess the question is like is is there is there a product or is the natural evolution of just general prompts that as as algorithms get faster as these models run faster that there is a deep research amount of work that happens uh within a few seconds uh between every response.
1:02:17
And and BA basically the question is like how much can you port from the deep research product and strategy and design back into just your average LLM interaction.
1:02:33
Yeah, I think there's definitely a lot of learnings that we can kind of start upstreaming really around like being able to form a plan, follow that plan to do that sort of multihop steps of like search iterating like finding insights, changing your strategy possible before going back to the user.
1:02:51
And so you're you're kind of starting to see this in like 2.
1:02:54
5 Pro and stuff like that.
1:02:54
And I you can imagine that that will continue where you will see more like mini deep research or more sort of like planning um uh and and sort of like iterative reasoning before like giving you an answer and yeah as that could just start getting faster and faster and faster then you start just getting like way more insightful or um uh uh comprehensive answers.
1:03:17
Are there any other interesting areas?
1:03:19
is I mean deep research feels like one of the first like really solid product market fit experiences in I guess like agents broadly.
1:03:28
Um, are there any other areas that you're excited to think about knocking down with either different products or just maybe un uh just like cool uses that you've or developed or as a user uh patterns that you're leveraging that's maybe go beyond just the average like I need a I need a a research report. Yeah, totally.
1:03:52
So, I think there's like a few different angles that that like I think a lot of people are exploring.
1:03:55
One is you kind of pointed out like what does a 2-hour deep research look like?
1:03:59
What does an overnight deep research look like?
1:04:00
look like? Yeah, if you can have like a very well- definfined problem where like um you know we have early experiments at Google like AI cosient and stuff like you could run that overnight and it can come up with like novel scientific hypotheses right so there definitely is an angle of like if you can define a problem and an outcome really well
1:04:18
applying more compute can actually get you like better and better answers right um so there's definitely an angle of like are there whole new classes of problems where we can even go even further with deep research um there's a second aspect of like um you know we had the chance to go like meet a bunch of people who are like researchers at the Fed, right? And they were telling us how
1:04:36
And they were telling us how they use research and it's often like um a very different thing, right?
1:04:41
So like I showed them this example where I was like, "Hey, there's this like funny law in the US uh called the Jones Act where like any two ships between like two US ports have to be like built in America, crewed by Americans." Yeah.
1:04:52
And that like drives up shipping prices but only for like Puerto Rico, Hawaii, and like Alaska, right?
1:04:59
Um and so I was like do an economic analysis of the Jones Act uh on on like the economy of Hawaii, right?
1:05:07
And it like did a first principal analysis did some really interesting things like looking at um well like how much is a three three and a half thousand shipping route like say from like Mexico to South America and then that's like a baseline price to compare against.
1:05:20
And like I thought this was amazing but then they were like that's not how we would do economic analysis.
1:05:25
like they would be like first I'd explore like what other studies there are like then I'd explore like what kinds of methodologies are out there then I might like ask a f bunch of false follow-up questions about like what data sources or like data um sets did people use to do this research right so there's
1:05:39
definitely an aspect of like another angle of like if I really want to help people with research it's about like nailing this sort of like synchronous asynchronous paradigm and helping people kind of do more of that like iterative process rather than just like ask question get answer and and move on and and in in victory. And I think that's
1:05:55
And I think that's that's kind of a product challenge like figuring out the right the right interaction model for that.
1:06:00
Um and the third is is just like outputting the outputting an answer at the right like level of abstraction that you work at, right?
1:06:09
Like a financial analyst doesn't think in terms of report, right?
1:06:11
They think in terms of like the spreadsheet or or the financial model, right?
1:06:14
And so if I want a DCF, deep research can build like a great DC discounted cash flow model for me, but like I don't want it in a report.
1:06:23
I want it in a spreadsheet or I want it in an app where I can play with the variables and see the different outcomes.
1:06:28
And so you'll also see the line between like reports and other kinds of artifacts starting to blur.
1:06:32
Um or even just like what like what does it mean to like build an build an answer, right?
1:06:40
And and that that could take like a much wider space. Super exciting. Yeah.
1:06:43
I mean, I've seen obviously Gemini, we probably can't talk about the road map too much, but I've seen Gemini pop up in a bunch of different areas and and I haven't seen the deep research version of whatever that instantiation is.
1:06:54
Yeah, my my maybe my last question is like how much time are you thinking about working and making the you know as a product manager on Gemini, how how much time are you thinking about making Gemini better versus sort of fighting for distribution outside of Gemini and a kind of across the Google ecosystem?
1:07:11
Because part of unlocking the value of Gemini is just making sure it's in the right places and and placed sort of contextually across, you know, everything from um you know, gemini. google.
1:07:20
You've worked hard on this.
1:07:22
Just ask for the I'm feeling lucky button. Just give us that.
1:07:23
Just like we search we you've earned it. It's a great product.
1:07:28
Just click I'm feeling lucky or burn 40 40 GPU hours on this new research award.
1:07:36
putting Yeah, that that would like instantly melt all of our servers everywhere.
1:07:41
Uh uh and this is this is the biggest hypers. Yeah, we need more GPUs. Uh let's can TPU.
1:07:54
I believe you've earned the feeling lucky.
1:07:57
I haven't hit the I'm feeling lucky button in years yet.
1:07:59
I use Gemini all the time. So give me the US.
1:08:03
This is what the users want.
1:08:03
Yeah, we just need 10 more TSMC's, I guess, to start fabbing. Uh, anyway, sorry. Serious answer. Yeah.
1:08:10
No, the serious answer is like the Gemini app is like a great place for us to like prototype, see what like hits with like really works with people.
1:08:17
A lot of the users, they're very intentional when they're coming to the Gemini app.
1:08:20
Like they want to use an AI experience.
1:08:21
Um, so it's a really great place for us to like put stuff out there, see what works, see what doesn't.
1:08:26
Some things we put out needs more time in the oven.
1:08:28
Um, and then over time you'd imagine that then like those insights or things that really start work, you'll start seeing in other Google products as they make sense, right?
1:08:37
You don't want to like overclutter a UI, but you'll start seeing Yeah.
1:08:40
things like deep research. Yeah.
1:08:42
Because it's a very it's a very different user somebody that's coming in saying, "I want AI versus I just want to do certain things." Yeah.
1:08:48
And and yeah, they're totally different archetypes.
1:08:53
It's a fascinating challenge.
1:08:53
And I'm sure it's even more challenging at your scale, but thanks for all the hard work and and pushing the frontier forward.
1:08:58
Uh it's been a pleasure talking to you.
1:09:02
Yeah, come back on again soon.
1:09:02
Yeah, we'd love to talk to you more. Appreciate it. Thanks. We'll talk to you soon. Bye. Fantastic.
1:09:08
Next up, we have Oliver Cameron.
1:09:11
Uh I have a I have a good story.
1:09:13
We'll bring him into the studio, but I believe he was the first person I ever interviewed for a YouTube video years ago.
1:09:19
Uh, I was doing a whole video essay about Cruz, the self-driving car company, and he hopped on a Zoom call with me just like this one, and I recorded it and threw clips in the video. It was very fun.
1:09:30
And then I wound up doing more interviews after that.
1:09:32
So, Oliver, good to see you. How are you doing? What's going on? Welcome. I'm doing great.
1:09:37
Thank you uh for the opportunity.
1:09:39
Uh, would you mind kicking us off with like the latest and greatest introduction because you've done a lot in your career, but you're on to something new. For sure.
1:09:45
Uh, so spent about eight years building self-driving cars. Incredible time.
1:09:49
I mean, just to see that technology go from barely being able to keep in a straight line to navigating downtown San Francisco with no human behind the wheel, just a sign of where things have gone uh with machine learning.
1:10:03
So, had a had a blast doing that.
1:10:05
Built my own company, sold that company to Cruz where we met and uh and loved that time.
1:10:09
left Cruz in May of 2023, decided to start something new.
1:10:15
And both me and my co-founder, who also was from self-driving cars, we were both um very much inspired by Pixar.
1:10:20
I think it's just a very special company, right?
1:10:24
Everyone kind of recognizes Pixar as this sort of iconic storytelling company.
1:10:29
And we really put our heads together to think about what a modern reincarnation of Pixar would look like.
1:10:36
So that company is called Odyssey and we're an AI lab that's really focused on uh enabling entirely new stories uh to be told.
1:10:45
And uh walk us through the first product that you launched.
1:10:48
I played with it earlier.
1:10:49
Uh it it was mind-blowing.
1:10:52
We'll we'll pull it up while you're talking. Sure. Yeah.
1:10:53
We just released a research preview of something that we call interactive video. Mhm.
1:10:58
And it's effectively AI video that you can both watch and interact with in real time. Yeah.
1:11:05
Um, and we think this will become a entirely new form of entertainment.
1:11:11
You know, you've got film, you've got games, you got all these these mediums that have been around for a while.
1:11:14
We think that there is an opportunity to invent a brand new one where effectively a model is responsible for imagining film and game-like experiences in real time. Yeah.
1:11:27
Um, that you can interact with.
1:11:29
There's no game engine behind all of this.
1:11:30
No huristics, no rules, just a model that's learned pixels and actions from tons and tons and tons of uh real life video.
1:11:37
Yeah, we're showing it on the screen right now and the production team is controlling it with the keyboard WD like it's a first-person video game and they're walking around this field uh with trees and windmills and they can actually choose to go up go inside buildings and it's all being generated without the use of a game engine and then they can switch over to a different environment.
1:11:59
And so I mean I have tons tons of questions about how these different like you're not doing photo scanning, you're not doing uh you're not doing game engine stuff, traditional 3D pipeline, but the data must come from somewhere. Love to hear about that.
1:12:12
And then also I noticed the space button doesn't work.
1:12:17
I wanted to jump around, stop start bunny hopping.
1:12:19
When are we getting a space button added to this thing? Anyways, anyway, sorry.
1:12:22
Isn't it trippy how that those pixels are literally streaming from a GPU cluster probably in Texas? It's so crazy.
1:12:30
And now we're streaming them via Zoom in real real time. It's crazy.
1:12:35
My my question is, do you think that Odyssey can be a really breakout app for VR?
1:12:43
Because when I see that visual, I I feel like that uh it could give someone the sense of being able to explore lands that don't exist, which is like very f like once it's fully immersive, it feels like it's funny the windmill thing because I remember the very first Oculus demo that I ever did.
1:13:00
I was walking around a windmill and it w and it's still in my mind years later but uh and it was amazing but it was just like one little windmill and then you couldn't go any further because developing virtual assets is really expensive and so you play a lot of these VR games and you know it's a couple hours or 30 minutes but if you take a procedural approach or a generative approach you all of a sudden have infinite content.
1:13:27
I think what's really important to note is in film and game incredible things can be made right like insanely good things that wow us all the time and the money it takes to create those things is ludicrous and it's only getting more expensive not less expensive over time. Yeah.
1:13:43
So I feel there will be continuously a place for these sort of like handcrafted things um and and they'll be very important.
1:13:53
But if we just think about a model that's trained on literally decades of of video that's then able to imagine stuff in real time with no pre-production costs, no post-production costs, it and and do that in literally in real time like 33 milliseconds.
1:14:10
It just that that's where it gets really crazy.
1:14:12
Um and and what we showed uh in the research preview is just like this tiny glimpse I think of what this stuff will become.
1:14:19
VR in particular is like the most hardcore application of this um from a a technical perspective because the resolution required for VR is like insane and the resolution that you saw there you can tell it's low res.
1:14:29
It's like 300 pixels wide.
1:14:31
So there's going to be a leap that needs to happen there to get to to VR level res.
1:14:35
But I'm confident that Odyssey 2 you'll have it you'll have it. You'll have it dialed. Oh yeah. 2. Give us the stats.
1:14:43
uh how many uh like like what numbers can you give us about the progress or adoption?
1:14:49
You just launched this I think this week or last week.
1:14:53
It hasn't been very long, but how have how have how's the response been quantitatively?
1:14:57
Oh, it's been incredible.
1:14:58
So, we launched a week ago and since then we've served 250,000 unique streams, meaning 250,000 people experiencing what you just saw, which is insane.
1:15:07
Market clearing order inbound. Yeah, let's do it. Yeah, love it. Congratulations. That's fantastic.
1:15:17
Uh on on on the question of resolution, um there's a bunch of amazing AI upresing that's happening in in various parts of the pipeline.
1:15:25
Um there's some server-based upresing that can happen.
1:15:29
There's some ondevice upresing.
1:15:31
So is that uh are you are you counting on that technology breaking one way or another? Does it matter?
1:15:37
Will it be a combination of both?
1:15:39
How do you see that developing?
1:15:41
I think a way to think of this is where video models were a year ago is where real-time video models or world models will be um today. Yeah.
1:15:52
And what that really means is that you look at the res.
1:15:53
Remember the Will, everyone remembers the Will Smith spaghetti video.
1:15:56
That was was that like one year ago or two years ago? It wasn't long ago.
1:16:00
I think it was just over a year ago. Um so fast.
1:16:01
There was definitely better outputs.
1:16:04
Like spaghetti was like the the weirdest hottest thing at the time.
1:16:07
Although gymnastics today, I'm sure you've seen that.
1:16:11
That's really tough with video models today.
1:16:12
Um, that's all to say that I think the res and the quality visual quality improvements will come from the model itself, not like some secondary um piece of um infrastructure to upres. Sure.
1:16:26
Um just because I mean think of what a language model was like to use two years ago like how fast was it in response time?
1:16:33
Really quite slow, right?
1:16:35
Compared to today where it's like just stream of information straight to your eyeballs.
1:16:38
Same will be true of these models.
1:16:39
like will crank out larger resolutions, faster frame rates, uh more actions, more things you can do, all that sort of stuff. Yeah.
1:16:47
And I guess I guess importantly like GPT 4.
1:16:49
5 is not GPT4 upresed to 45. It is a different model.
1:16:57
Um we're we're walking around uh what looks like the gloomy English countryside right now and I think the production team is going to try and go in that house.
1:17:04
Uh it it it is it is really really so wild.
1:17:07
Uh, I noticed that there's a there's a time limit.
1:17:10
Tropical Island demo because this this uh this I I love I love the English country. Foggy. Uh, yeah.
1:17:15
I I noticed that there's like a two-minute timer when I sign in.
1:17:20
Is that so the GPUs don't melt?
1:17:22
I mean, I I assume you've raised money and you're maybe burning some money with these demos.
1:17:26
Uh, but but break down kind of like uh what your limitations are and how you see them evolving. Yeah, for sure.
1:17:35
So the uh timer is there because each session is served by a single GPU.
1:17:38
So um each user gets a GPU, the model's running there and that's bean to the user directly. Mhm.
1:17:45
Um and really quickly you when you say single GPU, you don't mean rack, you mean like one A100 or something like that.
1:17:53
Uh one H200 H200 per user. Got it.
1:17:56
And um there is a clear path to like dividing the GPU to have multiple sessions per user, but today it's one and we want to really crank up quality, frame rate, all that sort of stuff.
1:18:07
It makes me feel great to know that I'm getting, you know, the the sort of one-on-one attention from H200 chip, you know?
1:18:14
It's like if you're at a retail store, it's it's not a great experience if somebody's bouncing around between Exactly.
1:18:19
I'm being individually served.
1:18:21
I like being served by This is like an air level by Jensen. Yeah. By Jensen.
1:18:26
So $2 an hour is there or thereabouts how much that costs which you know over the course of a multiple users it's not too bad. Yeah.
1:18:33
Um I think Netflix is like 5 cents 10 cents an hour something like that to stream video.
1:18:39
So we're a bit of a ways away but you've got new chips coming just model optimizations like it won't be long where we're having a single GPU per user all that sort of stuff.
1:18:48
Um, for this launch, we had something like 360 H200's prepared.
1:18:55
We have to scale it up a little bit.
1:18:57
Um, just because we had lots of demand, but um, that timer is there just to make sure we're cycling through lots of people getting uh, getting a taste of this.
1:19:04
Um, but yeah, I I think fundamentally the idea that you could have a model stream stuff to any screen is really powerful.
1:19:11
Like that experience you saw there works just as well on an iPhone, on an Android, on a TV, anything like that.
1:19:18
Um, and it's all just action conditioned over WebRTC, which is probably what Zoom is running on.
1:19:21
So, the action is just sent over the wire to the model.
1:19:25
The model then conditions the pixels it's about to generate based on those actions, sends the pixels back, and just that loop every 33 milliseconds firing.
1:19:32
So, I mean, the the path to HD or 4K seems pretty clear to me.
1:19:36
Uh, what about the path to uh consistency?
1:19:38
That that feels really difficult.
1:19:41
You need essentially like a really long context window to know that okay I dropped my mythical sword on in on on that piece of the ground.
1:19:51
I went away and then I came back.
1:19:53
That's like textbook just put it in a database.
1:19:55
But it seems like the future might not be that.
1:19:57
So how how are we thinking about that?
1:20:00
Will the I I I guess the bigger question is like what's the response from the gaming community?
1:20:05
Is this something that can be a tool and a piece of a pipeline instead of completely replacing the entire traditional pipeline?
1:20:11
So most research on interactive video before has learned from games.
1:20:15
So lots of folks will have seen uh Oasis from Decart, a Minecraft in a video model effectively or Quake uh that's often used in video models.
1:20:24
Um and I think the gaming reaction to that is um quite negative.
1:20:28
It's like I mean you saw the Carmarmac back and forth, right?
1:20:33
Where Carmarmac was like this is amazing. I loved it.
1:20:34
And somebody else is like this is stealing, you know, developers. Yeah.
1:20:38
And I I think it's important because um the way that people envision that is like oh what's the best this could become it could become like remixing of games. Yeah.
1:20:48
Um and that's one way it could be.
1:20:51
I think people see what we have and they think oh this is like a world simulator eventually.
1:20:54
This is the Matrix or like whatever they project on it. Yeah.
1:20:57
So really I one thing we're trying to avoid is like for the first few um generations of this people will put including ourselves like this picture of what um existing games look like onto this. Sure.
1:21:11
And it's like the iPhone when it launched, right?
1:21:15
People ported desktop apps to the iPhone and it kind of worked but it kind of didn't.
1:21:19
Wasn't really embracing this new medium.
1:21:21
medium. So I think the the long story short here is like stuff that is integral to games today like multiplayer like state um uh like scripting all that sort of stuff let's question those assumptions like how should those things work let's make it model native like maybe memory in this model is very
1:21:38
different than memory uh in in a in a game or state in a game multiplayer in a game all that sort of stuff um and that's probably going to lead us in the short term to more like glitchy weird um experiences throughout the the memory as it is by the models is a feature, not a bug. I don't know if you guys have seen
1:21:52
I don't know if you guys have seen like the back rooms or like these kind of glitchy weird Yeah. Yeah.
1:21:55
It almost is a completely different type of game design. Exactly. Yeah.
1:21:58
The the the up down, left, right, ab future will be like, "Drop your bad sword on the ground, walk around the building three times, come back, and it's and it's enchanted."
1:22:08
Um because because the model hallucinates that you've upgraded or something like that. That'll be fun.
1:22:13
I also think that one um important thing here is that in language models, one of the the things that's happened in the last year is in many cases they've crossed this threshold of realism for certain applications.
1:22:25
So like people literally fall in love with language models, right?
1:22:28
Like yeah, the same like emotional feeling they have when they meet a person they fall in love with is happening for them with a language model.
1:22:34
And that's because they what they're seeing on their screen is like so realistic.
1:22:37
It's like crazy real to them.
1:22:39
And I I think the same will be true here where once these pixels, these actions feel so realistic which eventually they should just give them the data, give them the models and advancement.
1:22:47
There'll be things that they do in these worlds um or things they feel in these worlds which they just can't feel in video games today because games are just capped by computer graphics and like human dev time and budgets and everything else.
1:23:00
But they'll walk down the street, they'll see someone and they'll be like, "Wow, that person looks so real."
1:23:02
And they'll go over, they'll like high-five that person all like on a screen, right? Yeah.
1:23:10
They'll feel like a heartbeat raise, you know, stuff like that. Totally.
1:23:11
Um, so that's that's an application that you can't do in games today.
1:23:15
That's just different and new.
1:23:17
So, that's the sort of stuff we're really interested in.
1:23:20
Well, that's going to be a wild wild feature. Uh, but thank you.
1:23:22
We'll have to have you back and check in on progress. Uh, yeah, fascinating.
1:23:27
Definitely the day that 720p drops or whatever the next version is.
1:23:29
Um, we're excited for this.
1:23:32
Uh, but thanks so much for joining.
1:23:34
This was a fantastic conversation.
1:23:35
We will talk to you soon. Cheers.
1:23:36
Have a great rest of your day. Thanks guys for joining. Thanks so much.
1:23:39
Uh, next up we have a return guest, Michael Mcdano from Lightseed coming into the studio.
1:23:43
Are you enough to hit the gong or he's going to talk about competition between the Foundation Labs and the App Layer?
1:23:51
Well, welcome to the stream, Michael. How you doing? Boom. Good. Good to see you guys.
1:23:55
Congrats on the new studio. Thank you. It's been a lot of fun.
1:23:59
Uh, bit of a I like the upgraded gong, too. Oh, yeah. The gong is much bigger.
1:24:05
We got a bigger we got a bigger one in the works.
1:24:06
Everybody says even bigger. Oh yeah.
1:24:10
Also, uh funny day to just be so hyperfixated on AI because you probably haven't seen the timeline. Tesla's down 17%. 17%.
1:24:23
It's just absolute mayhem.
1:24:23
I mean, there's an AI narrative there, right?
1:24:26
Yeah, there's definitely not there.
1:24:26
Um but anyways, Michael, it's great to have you on.
1:24:32
um wanted to get some kind of updated thinking from you on the tension between labs and the application layer.
1:24:42
We saw the news with Windsurf and Anthropic today that had more to do with a potential acquisition and even when we talked to the founder of Granola, we were we were talking about the competition between Notion and Granola with these it's a founder kind of previous era uh scale up unicorn SAS company.
1:25:00
Can that company bolt on AI?
1:25:03
But then now we're seeing competition from the Foundation Lab.
1:25:04
So would love to get your lay of the land. What are you seeing?
1:25:08
Uh how are things shaking out and what do you think the next few months or even years look like?
1:25:11
Yeah, it's pretty interesting, right?
1:25:13
Like if you think about the big companies that startups previously uh built on the backs of the Googles, the Amazons, the Microsofts, you know, it it felt like there was this really healthy sort of symbiotic developer ecosystem where the incumbents supply resources, the developers sort of buy and extract from them and they build really really big businesses on top.
1:25:35
I think what we're seeing now to your point is these labs are building developer ecosystems but then they're very intentionally and overtly going head-tohead with the developers that are building on them.
1:25:48
And I think this has a lot to do with context, right?
1:25:51
So if you think back to the internet, you know, and startups 10 years ago, everyone said content is king.
1:25:58
You know, content is king.
1:25:58
Then dist distribution was king, right?
1:26:00
It was all about how do you get in front of users?
1:26:03
It's it's we're starting to feel like we're entering the phase of context being king.
1:26:08
These models are just hungry for the most and the most unique context possible.
1:26:15
And so if an app layer company emerges and has a new type of context and data that the models don't have great exposure to, it's a great signal to to point in the direction and say we're going to compete head-on.
1:26:27
And so I I think that's what we're seeing now.
1:26:30
And um yeah, Nabil Hyatt, uh great investor from Spark and I, we often talk about how the war for context is happening now.
1:26:39
And I think that's that's what a lot of these moves represent.
1:26:43
How do you think uh app layers should app layer companies should respond?
1:26:46
Is it just double triple down go way way way deeper focus on workflows that the labs maybe don't have the resources to fully pursue or is it focusing down on specific niches?
1:26:57
I'm curious what the you think the right approach is.
1:26:59
Well, we we we could definitely get into that, but maybe first what I would say is, you know, I I I tweeted something yesterday that occurred to me after the big announcements from OpenAI in that, you know, the big incumbents, which we talked about a little while ago, sort of like the winners of the cloud era.
1:27:13
It wouldn't surprise me if all of these new uh you know, these new competitions between the labs and the apps actually drive the apps and the startups right back to the incumbents, to the Googles and the Amazons of the world.
1:27:27
I I have to wonder if some of these things actually act as a tailwind for models like Gemini and maybe give a little more credence to the the the argument that like Google is actually going to be the winner here because of all their distribution.
1:27:39
So I think that's one potential.
1:27:41
You mean driving back of being like I'd rather work with Gemini because I don't think they're as likely to kill me. Exactly. Yeah. Exactly.
1:27:47
It's like hey we trusted them with the cloud and that worked out all right.
1:27:51
Like should we now trust them with AI more than we trust the labs? Yeah.
1:27:54
than we trust the labs? Yeah. I mean that narrative even goes a little bit further with Microsoft which has been completely like oh we will host every single model we'll let you reroute really intelligently between them like uh super super friendly developer
1:28:07
ecosystem and so I mean certainly they're building stuff into co-pilot into Microsoft 365 but uh it does feel like they're they're they're much more willing to to SA SATA seems to have real conviction you know he had the quote from last week platform platform platform and hosting DeepSeek is an example of that, right? A lot of people
1:28:28
A lot of people would have thought, oh, he's not necessarily going to host that model because it felt like a shot across the bow at OpenAI, but he's committed to supporting open source, right? Yeah. Yeah. He wants it all. Interesting.
1:28:39
I I I think, you know, also going back to your question, Jordy, I I think all of this is just going to make for a more intense uh faster moving market.
1:28:47
like I think more than ever before you have to ship.
1:28:52
You have to get users faster than anyone.
1:28:55
You have to sort of like reach escape velocity quicker.
1:28:57
Um which I think just is just going to put more and more pressure on startups to move even quicker than they already are.
1:29:03
You know, I feel like cursor is a great example.
1:29:06
I feel like an earlier iteration of that product, you know, it probably would have been easy to sort of write them off and be like, oh, you know, a lab is going to do this.
1:29:14
I mean, now it's like they're so big, they're so far ahead, it feels like they've they've really established themselves and and likely have a good shot of breaking through.
1:29:22
I also wonder with Cursor and and Windsurf and Devon and some of the dev tools markets, like it feels like just such a new market that even if it's somewhat winner take all, there's just it's so positive some because it's it's it's adding efficiency to the most like one of the biggest labor pools.
1:29:43
And so, uh, when we talked to the cognition folks, uh, as the reaction to Google and OpenAI launching Devon competitors, they're like, well, we still grew 40% last month or something like that.
1:29:55
And so, you know, I I wonder like in in codegen where it's such a new market that it's not it's not directly competitive with anything that exists, so it's less zero sum.
1:30:07
I'm wondering if the note-taking market feels similarly to you or or were you seeing Granola or other or other companies kind of act as more uh drop in replacements for existing tools?
1:30:17
Yeah, I I think it's a great question.
1:30:21
Yeah, so so we backed Granola really early on because we we knew Chris and his co-founder and we love those guys.
1:30:28
We didn't know what they were building.
1:30:29
We knew they were going to build something in note takingaking.
1:30:32
Um but we said, you know what, this market's going to move fast.
1:30:36
we trust these guys, let's go for it.
1:30:38
And um and I think you know somewhat to your point, there's been all these notetakers before like Granola wasn't the first noteaker.
1:30:44
There was Fireflies and Honor and all these things but I think Granola has done a really really good job of um you know getting out of the user's way and establishing trust with the user.
1:30:56
And I think that, you know, that that seems like a small thing, but I think that trust thing is going to be really important.
1:31:00
If you go back to what I said about this context being king, like who are you going to trust to take this context or take this like really, really important, you know, proprietary part of your work?
1:31:11
In this case, your meeting notes.
1:31:13
You know, a lot of people say, "We trust Granola."
1:31:17
Are they just going to hand it over to any old company that says, "Hey, now we want to screenshot your entire computer and take and suck every last piece of data out of you."
1:31:24
And so I think part of it is to your point like getting in early, getting big really really fast and establishing that, you know, that user base and that market before it really matures, but also in a way that like users just really trust you and they're not just going to rip you out just because some other bigger company offers the same thing. Yeah. Yeah, it's a good point.
1:31:43
Um, what do you have any more like micro reactions to specific integrations?
1:31:48
that seemed to be one of the big things that Openo was pushing on was uh integrations with Google Docs and Drive and and your email.
1:31:55
Uh and that feels like adding that extra context is is potentially the next thing people are clamoring for.
1:32:04
Uh how important is like the bisdev side of this business?
1:32:06
In fact, I think it's really important.
1:32:08
You know, I think it's really uh really great um that Anthropic started the whole MCP protocol.
1:32:15
Obviously, lots of others are adopting that now, but I think to your point, we're now going to start to see the battle lines being drawn, like who are who are we willing to integrate with, who are we not willing to integrate with, where is, you know, are we open or are we closed, where's the data going to go, where's it not going to go.
1:32:32
I think we're going to start to see those alliances and those allegiances form.
1:32:35
And um APIs like back in what, like 2007, 2010 era, all they have an API. It's amazing.
1:32:44
It's like well like you don't know how much that API is going to cost if it's $10,000 per day or something like that could completely upend your business.
1:32:53
Uh and so actually thinking about how that dynamic develops is is is almost more important than the standard.
1:32:59
Although I'm very glad we have a standard that seems great.
1:33:02
Uh but but each company is going to have to decide where the value acrruel really lands.
1:33:07
Uh and then who knows maybe there'll be some antirust in in 20 years like we're seeing with Apple. Yeah.
1:33:13
Yeah, the big the big question around trust uh that I I you know it's a it's an evolving situation but uh a California judge I believe it was yesterday or the day before ordered OpenAI to uh retain records of uh sort of um forget the what OpenAI calls it but if you have like a a disappearing query a judge ordered them that they have to retain that they obviously said that's a huge reach privacy with users so incognito mode. It's like not incognito. Yeah.
1:33:45
And that that that more seems like an issue with the court and specific judge, you know, having this massive overreach around privacy.
1:33:51
But um privacy in this era when people are more willing than ever across every app to give them all sorts of data. Yeah.
1:33:58
And and you have a direct incentive to to to reduce the level of privacy to get better results.
1:34:04
Like I if if if the model knows what kind of car you drive, when you ask it for new tires, it will give you better recommendations.
1:34:10
So, you want to lean into being anti-privacy to get better uh better results.
1:34:15
There's there's this the the the world is definitely bifurcating into pro privacy or like fully AGI pilled folks and and and there aren't that many people there that are in the middle.
1:34:25
So, obviously we will have to figure it out as a democratic society ultimately vote and uh hopefully sort it all out in the courts.
1:34:32
But thank you so much for stopping by.
1:34:34
This was always pleasure.
1:34:36
We'd love to have you back. Yeah.
1:34:36
I mean, guys, I just want to tell you, you know, I I don't really aspire to ring the New York Stock Exchange bell one day.
1:34:43
I I I aspire to hit that gong. Hit that gong.
1:34:46
Well, next time you're in Los Angeles, come by. Come by the gong.
1:34:47
Uh, great to see you, Michael. Great to see you.
1:34:52
Um, so, uh, we have a generational crash out going down on the timeline.
1:34:55
We got a new post from Elon.
1:34:57
I'm going to read it out.
1:34:59
He says, and this is your live reaction, John.
1:35:01
Time to drop the really big bomb.
1:35:04
Real Donald Trump is in the Epstein files.
1:35:06
That is the real reason they have not been made public. Have a nice day, DJT. Wow.
1:35:09
That is that is a big that is a big bomb.
1:35:13
Uh but wait, didn't we already know this?
1:35:16
Because isn't there that picture with Trump and Epstein together?
1:35:18
We're really in dark territory.
1:35:20
The business I want to go back to AI business.
1:35:23
The business story here is that Tesla's down 17%, DJT is down 7%, Trumpcoin is down 10%. Wow. They're all fighting.
1:35:31
This crash out on both sides is not good for anyone.
1:35:33
Well, you know what's interesting?
1:35:37
You know what's not down? Tokens generated, baby.
1:35:42
We're still generating tokens every single day.
1:35:45
The the relentless march of artificial intelligence continues.
1:35:47
So, so the other thing is is Elon shared on or sorry, Trump shared on truth.
1:35:52
It's funny, they're battling on each different social every billionaire should have their own, you know, social media network to get the word out.
1:35:58
But Trump said, "The easiest way to save money in our budget, billions and billions of dollars, is to terminate Elon's government subsidies and contracts.
1:36:07
I was always surprised that Biden didn't do it." Wow.
1:36:09
Um, so, uh, Ashley Stlair is saying, "Hey, Donald Trump, let me know if you need any breakup advice."
1:36:18
Um, they're really And Dan Primac says, "This cannot be a comfortable day for David Saxs.
1:36:22
On the other hand, it's just the best day for Sam Alman."
1:36:24
Well, well, we have someone from OpenAI here.
1:36:27
We're gonna stick to technology and business, but welcome to the show, Mark Chen. Good to see you. Good to see you guys.
1:36:34
Thanks for awkward day, but I'm excited to talk about Deep Research.
1:36:38
I am excited to talk about AI products.
1:36:40
Uh would you mind introducing yourself and kind of explaining what you do because OpenAI is such a large company now and there's so many different organizations.
1:36:48
I'd love to know uh how you interact with the product and the research side and anything else you can give to contextualize this conversation. Yeah, absolutely.
1:36:57
So first off, you know, thanks for having me on. You know, um I'm Mark.
1:37:00
I am the chief research officer at OpenAI.
1:37:02
So, uh in practice, what that means is I work with our chief scientist, Jakob.
1:37:06
And you know, we set the vision for the research or we set the pace, we hold the research or accountable for execution and uh ultimately we really just want to deliver these capabilities to everyone. That's amazing.
1:37:18
In terms of research, I feel like a lot of the what happens in the research side is actually gated by compute.
1:37:24
Is that a different team?
1:37:27
Because what if the researchers ask for a 500 billion dollar data center?
1:37:29
Uh that feels like maybe a bigger a bigger task.
1:37:34
So yeah, it is useful for us to factor the problem of uh research and also kind of building up the capacity to do that research.
1:37:40
So we have a different team uh Greg leads that um which really thinks holistically about you know data sitter bring up and how to get the most compute for us.
1:37:48
And of course uh when it comes to allocating that compute for research uh you know Jakob and myself do that. That's great.
1:37:54
Um and so uh what uh what can you share that's top of mind right now on the research side?
1:38:00
There's been this discussion of pre-training scaling wall potentially the importance of reinforcement learning uh reasoning.
1:38:11
There's so many different areas to go into what's actually driving the most conversations internally right now. Yeah, absolutely.
1:38:18
So um I think really it's a really exciting time to do research.
1:38:22
Um I would say versus two or three years ago I think people were trying to build this very big scaling machine. Yeah.
1:38:28
Um and really the reasoning paradigm changed a lot of that right you know like reasoning is really taking off and it really opens this new playing ground right it's like there are a lot of kind of known unknowns and also unknown unknowns that you know we're all trying to figure out.
1:38:44
It kind of feels like GPT2 era right where there's so many different hyperparameters you're trying to figure out.
1:38:49
And then I think also, you know, um like you mentioned, you know, pre-training, that's not to be forgotten either.
1:38:54
Um you know, today we're in a very different regime of pre-training than we used to be, right?
1:39:00
Um today, uh we can't treat data as this infinite resource.
1:39:04
And I think a lot of academic studies, you know, they've always kind of treated, you know, you have some kind of finite compute, but infinite data.
1:39:12
I don't think there's much study of, you know, like uh you know, finite data and infinite compute.
1:39:18
And I think you know uh that also leads to a very rich playground for research.
1:39:22
Do we need kind of a revision to the bitter lesson?
1:39:24
Is that a a reputation of the bitter lesson or or or do we just need to re rethink what the definition of of scaling laws looks like?
1:39:34
Uh no I I don't think of uh anything as a reputation of of the bitter really like our company is grounded in we want simple ideas that scale.
1:39:44
I think RL is an embodiment of that.
1:39:47
I think pre-training is an embodiment of that and really at every single scale we face some kind of difficulty of this form.
1:39:52
It's just like you got to find some innovation that gets you past the next bottleneck and this doesn't feel fundamentally very different from that.
1:40:01
Um what is uh what's most important right now on the actual uh compute side?
1:40:06
Uh we heard from Nvidia earnings that uh that we didn't get a ton of guidance on the shift from uh training to inference usage of NVIDIA GPUs, but it feels like it must be coming.
1:40:19
It feels like this inference wave is is is happening.
1:40:21
Uh are those even the right buckets to be thinking about tracking metrics in terms of the the the story of artificial intelligence?
1:40:33
Because yeah, I mean it's like if if the reasoning tokens are inference tokens and and but they're what lead to higher intelligent more intelligent models like it's almost back in the training bucket again.
1:40:44
Um what bucket should we be thinking about and and uh and or or are we how firmly are we in the the the uh the the applied AI era versus the research era?
1:40:57
Well, I think research is here to stay and it's for all the reasons I mentioned above, right?
1:41:03
it's such a like a rich time to be doing research.
1:41:05
But I do think, you know, inference is going to be increasingly important as well, right?
1:41:11
It's such a core part of RL um that you're doing rollouts.
1:41:13
And I think, you know, we see 2025 as this year of agents, right?
1:41:18
Um we think of it as a year where models are going to do a lot more autonomous work.
1:41:23
You can let them kind of be unsupervised for much longer periods of time.
1:41:27
And um that is just going to put big demands on inference, right?
1:41:32
when you think about kind of our overall vision, right?
1:41:34
We we lay it out as a series of steps and levels on the way to AGI, right?
1:41:38
And I think the pinnacle really that last level is organizational AI, right?
1:41:43
Like you can imagine a bunch of AIs all interacting.
1:41:47
Um, and yeah, I think that's just going to put huge demands on inference, right?
1:41:52
on that on that organizational question.
1:41:55
I I remember reading uh AI 2027 and one of the things that they proposed was that the AIS would actually like literally be talking to each other in Slack.
1:42:04
Um, does that seem like does that seem like the way you imagine agents playing out like using the to the same tools as humans instead of one agent says, "I'm going to go talk with Teams and I'm going to talk with Slack and I'm going to do a little negotiating, but maybe it just happens super super fast 247 or or is there like a new machine language that emerges?" Yeah.
1:42:25
Um I mean I think one thing that's really helped us so far in AI development is uh to come in with some priors for um you know how humans do things and that's actually um you know if you bake those priors in they they typically are great starting points.
1:42:40
So I could imagine like maybe you start with something that's Slack like and give it enough flexibility that it can kind of develop beyond that and really figure out the way that's most effective for it to communicate.
1:42:51
Um one important thing though is uh you know we want interpretability too right I think it's it's very helpful for us today that what the agents do is you know uh easy for us to read and interpret and I don't think you want that to go away as well.
1:43:07
So I think there's a lot of benefits just even from a pure like debug the whole system perspective but just let the models you know speak in a way that is familiar with us and you know you could also imagine like we might want to plug in to the system too right so you
1:43:22
know um whatever interfaces we're familiar with we would ideally like our model to be familiar with as well um I think it's also pretty compatible with uh you know we uh hit a big milestone we got uh I think three million in uh paying business users fairly recently. Let's go. Yeah, there Let's go. Yeah, there we go. Let's go.
1:43:43
And uh three gong hits for 3 million.
1:43:48
The gong will keep ringing for a while. Sorry, we had to do it.
1:43:53
I was hoping you would drop a number. Yeah. Yeah.
1:43:57
Um anyway, congratulations.
1:43:59
That's that's actually huge. That's amazing. Yeah. Yeah. Yeah.
1:44:00
Um but I think one big part of that is, you know, we have we have connectors now, right?
1:44:05
Um we're connecting into you know, like G drives and I think um yeah, you can imagine, you know, like Slack integrations, things like that.
1:44:11
I think we just want the models to be familiar with the ways we communicate and and get information. Yeah.
1:44:16
Uh can you talk about benchmarking?
1:44:18
It feels like we're potentially Yeah.
1:44:20
Do you think about benchmarks at all? Oh, yeah. A lot. I mean, okay.
1:44:25
But I think it's a difficult time for benchmarks, right?
1:44:27
Um I think we used to be in this world um where you have these human written benchmarks for other humans, right?
1:44:35
And I think we all have these norms for like what are good benchmarks, right?
1:44:39
Like we've all taken the SAT, we all have like a good conception of what it means to get, you know, whatever score on that.
1:44:44
Um, but I think the problem is the models are already at the point where for even the hardest human written benchmarks for other humans.
1:44:53
Um, it's really near saturated or saturated, right?
1:44:57
Um, I think one clear example here is the Amy like probably the hardest autogradable like uh human math eval at at least in the in the US.
1:45:08
Um, and yeah, the models are consistently getting like 90 plus% on these.
1:45:12
on these. And so what what what that means is I think there's um kind of two different things that people are doing right they're they're developing kind of model uh based benchmarks right they're not kind of things that we would give to an ordinary
1:45:28
human things like humanity's last exam things like you know epic AI that are really really at the at the frontier of what what people can do um and I think um the hard thing is it's not grounded in intuition anymore right like uh you know you don't have a lot of people who have taken these exams. So, it it makes
1:45:43
So, it it makes it harder to kind of calibrate on whether this is a good exam or not.
1:45:47
Um, one of the exciting things that's on the flip side of that is I really do think we're at the era where models are going to start innovating, right?
1:45:56
Because I think once you've passed the last kind of like the hardest human written exams, that's kind of at the edge of innovation.
1:46:03
And I I think you already see that with the models, right?
1:46:05
Like they're helping to write parts of papers.
1:46:08
Um and and I think the other kind of way that uh people have shifted is you know there's these you know ultra frontier evals but there also people kind of just indexing on real world impact right you look at revenue kind of the value you deliver to users um and I think that's ultimately what we care about.
1:46:27
Can you can you uh bring that back to interpretability research like with these super super hard uh math evals for example?
1:46:36
Uh if are are we doing the right research to understand if the thought process mirrors not just not just oneshotting the answer, oh you you you memorized it or you magically got it correct, but you actually took the correct path.
1:46:51
Kind of like you know you're graded for your work, not just the answer if you're in grade school.
1:46:55
Um and and you know Dario said that uh interpret interpret interpretability research will actually contribute to capabilities and even give a decisive lead. Do you agree with that?
1:47:06
What's your reaction to that concept of interpretability research being very important?
1:47:10
Yeah, I mean we care a lot about it here at OpenAI as well.
1:47:12
So um one thing that we care a lot about is interpreting how the model reasons, right?
1:47:19
um because I think um we've had a very kind of specific and strong view on this um in that we don't want to apply optimization pressure to how the model thinks so that it can be faithful in the way it thinks and to expose that to us you know without any kind of incentives to cater to what the user wants right I think it's actually very important to have that unfiltered view um because you know uh often times like if if the model isn't Sure.
1:47:49
You don't want to hide that fact, right?
1:47:51
Just for for it to kind of please the user.
1:47:53
And sometimes it really isn't sure, right?
1:47:55
And and so we've really done a lot of work to try to promote this norm of train of thought faithfulness and and interpretability.
1:48:04
Um and I think it it gives you a lot of uh sense into what the model's thinking and you know what are the pitfalls that it can go off into if it's not reasoning correctly.
1:48:12
That's such an important point because if you have somebody on your team and they come to you and they say, "Hey, you know, I think this is the right answer, but we should probably verify it."
1:48:21
It's like, well, it's still valuable. Totally.
1:48:23
Puts you on the right path.
1:48:25
But if somebody comes to you 100% confidence, this is this is the truth ends up being wrong.
1:48:29
It's like, well, like trust is just destroyed. Totally. Yeah.
1:48:32
Don't you guys feel like, you know, um, safety felt a lot more theoretical a couple years back, right?
1:48:37
But like today, you know, like the things that people were talking about a couple years, like scalable oversell, like really having the model be able to tell you like and convince you that the work it did was right.
1:48:46
It feels so much more relevant right now just because the capabilities are so strong. Yeah.
1:48:50
I mean, just personally, I' I've completely flipped from being like, uh, oh, the safety research is not that valuable because I'm not that worried about getting paper clipped.
1:49:00
It just seems like a very low likelihood that that's kind of like the bad ending like immediately and this fume and all this crazy greyg goo scenarios were just so abstract and sci-fi.
1:49:09
It just felt like economics will will fall into place and there will be uh like a like a cold like a nuclear ending which is like we didn't build nuclear plants and we just stopped everything because we humans seem to be good at that.
1:49:22
Uh but now that we're actually seeing things Yeah.
1:49:24
It's crazy how fast it's been, right? Like um Oh yeah.
1:49:27
I think my my like my personal story is it's like you know what what got me into uh AI was Alph Go, right?
1:49:33
Like just watching it get to that level of capability at Go and you were kind of like it was such an optimistic and also kind of a little bit of a sobering message right when you saw Lisa get beat.
1:49:43
beat. Um and I just remember you know like we we saw the coding models you know when we first launched like uh I think very OG codecs you know um with with GitHub copilot it was maybe like under you know a thousand ELO on um on code forces and I still remember the meeting where I walked into where the
1:50:01
team showed my score and they're like hey with models better than you and like you come full circle and it's like wow like I put decades of my life into this and you know the capabilities are there so like if you know I'm kind of at the top of my field in this thing and it's better than me like what can it do really? Yeah. Yeah. That's amazing. Uh Yeah. Yeah. That's amazing.
1:50:18
Uh do uh I have so many more questions on Alph Go.
1:50:22
Are there uh are there lessons from scaling how scaling played out there that you can that we can abstract abstract into the rest of AI research.
1:50:32
What I mean is uh as I remember it the Alph Go training run was not 100k H200s.
1:50:40
Uh, but what would happen if we actually did an AlphaGo style training run?
1:50:46
I mean, it would be an economic money pit, right?
1:50:50
Like they had no economic value to do.
1:50:52
But let's just say some benevolent trillionaire decides, I'm going to spend a billion dollars on a training run to beat Alph Go and go even bigger.
1:51:00
Um, is is Go at at some point solved?
1:51:04
Would we see kind of diminishing scaling curves? Could we throw extra RL?
1:51:09
Could we could we port back everything that we've doing in just general AGI research and and and just continue fighting it out in the world of go or does that end and does that teach us anything? Yeah. Yeah.
1:51:19
Honestly, um I feel like if you really are curious about these mysteries, join our team.
1:51:25
That's the thing I want to say.
1:51:25
But yeah, I mean um really like kind of the the central problem of today is RL scaling, right?
1:51:31
And um when you look at Alph Go, right, it's it's a narrow domain, right?
1:51:35
And I think in some sense that limits the amount of compute you can pump into it.
1:51:39
But even kind of small toy domains, they can teach you a lot about how you scale RL like what are the axes where it's most productive to to pump scale in.
1:51:48
Um I think a lot of scaling research just looks like that whether it's on RL or pre-training.
1:51:52
It's like you identify a lot of you know different different variables under which you can scale and like where is kind of where you get the best kind of like marginal impact for for pumping scale there.
1:52:03
Um I think that's a very open question for RL right now.
1:52:06
Um and I think what you mentioned as well is just like you know going from narrow to broad right um does that give you a lever to pump a lot more scale in as well?
1:52:14
Um I think when you look at our reasoning models today they're a lot more broad-based than uh you know just being able to kind of an expert system on go.
1:52:24
Um, so yeah, I really do think that um there are so many levers to to scale. And what about move 37?
1:52:31
That was such an iconic moment in that AlphaGo Lisa doll match. Uh they place move 37.
1:52:37
It's very unconventional.
1:52:39
Everyone thinks it's a blunder. It turns out not to be.
1:52:41
It turns out to be critical.
1:52:42
It turns it turns out to be innovation.
1:52:44
Uh do you think we are we're certainly post touring test in language models.
1:52:48
We're probably post touring test in image generation.
1:52:54
Um, but it feels like we're pre-move 37 in text generation in the sense that there hasn't been uh like a fully AI generated book that everyone is just, oh, it's the new Harry Potter. Everyone has to read it.
1:53:08
It's amazing and it's fully a and it's fully generated. Or or this image.
1:53:10
The images, they do go viral, but they go viral because they're AI.
1:53:15
Move 37 in the context of Go did not go viral because it was AI.
1:53:20
It felt like it was actual innovation.
1:53:22
So, uh, is that the right frame?
1:53:24
Does that make any sense? Yeah.
1:53:25
Um, I think it's not the wrong frame.
1:53:27
So, I I think some some quick thoughts on on that.
1:53:30
Um, I I think kind of um when you have something that's, you know, very measurable, like win or lose, right?
1:53:37
Something like uh like go.
1:53:37
Um, yeah, it's like very easy for us to kind of just judge, right?
1:53:42
Like did did the model do something right here?
1:53:43
Um, and I think the more fuzzy you get, um, you know, it is just harder, right?
1:53:49
Like, um, when it comes to is this the next Harry Potter, right?
1:53:53
Like, you know, it's not a universally loved book.
1:53:54
I think fairly universal, but you know, there's there's some haters.
1:53:58
Um, and yeah, I I I think it it is just kind of hard when it comes to these human subjective things, right?
1:54:04
Where um it's really hard to put down in words like what makes you like Harry Potter, right?
1:54:11
and um and so um I think those are always going to lag a little bit but you know I think you know we're we're developing more and more techniques to attack kind of these more open-ended um uh domains and I don't know I I wouldn't say that we're not at an innovative stage today.
1:54:28
So um I think my biggest touch with this was when we had the models compete on the IY last year.
1:54:33
models compete on the IY last year. So I highlight it's like the the international basically Olympics for for computer science um basically the the top four kids from from each country go and compete and these are really really
1:54:48
tough problems um basically selected so that they require some innovative insight to solve right um I think um and we did see the model come up with solutions even to some very ad hoc problems and and So I think there was a lot of surprise for me there, right? Um I was completely off
1:55:08
Um I was completely off base about which problems the model would be able to solve the most, right?
1:55:15
Um I think like I I kind of categorized there there's six problems.
1:55:18
Some of them as more kind of like oh this is standard, a little bit more standard, this is a little bit more out of the box and it's like it's not going to be able to solve this uh more out of the box one. But it did.
1:55:27
And I think um I think that really does speak to kind of uh these models have the capacity to do so especially training with RL.
1:55:34
Now now now put that in context of what's going on with ARC AGI.
1:55:38
Obviously OpenAI has made incredible progress there but it just when I do the problems it seems easy.
1:55:46
And when I look at the IOI sample problems I think this would be a 20-year process for me to figure out how to achieve that and I can do the RKGI on my phone.
1:55:56
uh is this the spiky intelligence concept?
1:55:59
Is this something that a small tweak in in algorithmic design just oneshots AGI or ARC AGI or or is there something else going on there that we should be aware of?
1:56:08
Yeah, I mean I think um part of this is the beauty of ARGI as well, right?
1:56:13
Like um I think I'm not sure if there's another kind of like human intuitive simpler benchmark which is for the models.
1:56:20
Um and I think really that's one of the things they really optimize for on on that benchmark.
1:56:23
Um I do think when it comes to models though like there's just a little bit of a perception gap as well like you know uh models aren't used to this kind of native um you know like just screen type input.
1:56:37
Um, I think there's a lot we can bridge there actually.
1:56:39
Um, even O4 mini um, it's a state-of-the-art multimodal model in many ways, including visual reasoning.
1:56:48
And I think uh, you know, you're you're starting to kind of build up the capacity for the models to take images, manipulate and and reason about them um, generate new images, write code on images.
1:56:58
And um I think it's just been kind of underfocused, but um I think when I talk to researchers in the field, they all see this as a part of intelligence too.
1:57:08
And we're going to continue the focus there. Yeah.
1:57:09
Is is is RKGI kind of in the if we're dropping a buzz word on it, is like program synthesis?
1:57:16
Is there a world where uh I I know that I I know the tokens like the images we see them as as renderings of squares and different colors, but uh the when they're fed into the LLM, they're typically uh just a stream of of numbers effectively.
1:57:31
Is there a world where actually adding a screenshot is what's important like visual reasoning? Yeah. Yeah.
1:57:38
So I think I think that could be important.
1:57:40
important. It's just like kind of uh you know whenever it comes to like textual representation of grids um models today just don't really do that well right and I think it's just kind of because humans don't really ever write down textual
1:57:56
representations of grids or like you know we have a chessboard like no one really kind of just like types it out in a grid like um and um and so the models are kind of like undertrained a little bit on on what that looks like and what that means. Sure. So, um you know I I I Sure.
1:58:09
So, um you know I I I think with more reasoning it we'll we'll just bridge the gap.
1:58:15
Um I think with better visual perception we'll just bridge that gap. Yeah.
1:58:18
How are you thinking about the role of non-lab researchers in the ecosystem today?
1:58:23
I'm sure you try to recruit some of the best ones but the ones that don't join your team.
1:58:30
Tell us about the one that got away. Yeah. The one that got away. Yeah.
1:58:34
No, I mean um I think it's still actually a fairly good time.
1:58:36
actually a fairly good time. I for for specific domains right uh to to be doing research and um you know I think the style is just very different um and you do feel the pull of non-lab researchers into labs because I think they feel like a lot of the burning problems in the
1:58:53
field are at scale right um and that's kind of one of the unfortunate things too right like when you look at reasoning um you just don't see that happen at small scale right there's like a certain scale at which it starts becoming signal bearing And that requires you to have resources, right? Um but I do think you know a lot of the
1:59:11
Um but I do think you know a lot of the really good work that I've seen, you know, there's um experimental architectures.
1:59:18
I think a lot of good work is happening in the academic world there.
1:59:21
Like a lot of study in optimization, um a lot of study in kind of like GANs, you know, um there's certain fields where you see a lot of fruitful research that that happens in academia.
1:59:31
Yeah, that makes a lot of sense.
1:59:33
How about uh consumer agents?
1:59:33
How are you thinking about them?
1:59:36
Uh you talked earlier about sort of B2B adoption and that's all very exciting, but how much do you and the research org think about breakout consumer agent products?
1:59:49
Yeah, that's a fantastic question.
1:59:51
I think um we think about it a lot.
1:59:53
I think uh that that's the short answer.
1:59:54
Um you know, we really do think like this year we're trying to focus on how we can move to the agentic world, right?
2:00:01
And um when I when I think about consumer agents, I think like Chachi PD proved that you know people got it right like people get conversational agents when it uh conversational kind of models but when when it comes to consumer agents we have a couple thesis and u that we've tried out in the world.
2:00:17
I think one one is deep research, right?
2:00:21
Um I think this is something that can do five to 30 minutes of of work autonomously, come back to you and really like um kind of synthesizes information, right?
2:00:30
Um it goes out there um gathers, collects, and kind of, you know, compresses the information in a form that that's useful.
2:00:36
A little bit of a little bit of push back there.
2:00:38
Like I can see that as a consumer product when someone like Aiden is like I want new towels and he uses deep research to like figure out like what is the best towel across every dimension.
2:00:48
But when I think of deep research yes it has applications with students but it's often some consumers being like give me a deep research report on this country and where to travel and things.
2:01:02
We keep using this flight example but I don't I haven't actually tried to book a flight with deep research.
2:01:06
It's totally possible that it could go and pull all the different flight routes and and calculate all the different delays and all the different all the different parameters of if I fly to this airport I can park or I can use valet here or something like that. Yeah. Yeah.
2:01:19
And I guess like when I think of agents, it's it's deep research is like you know curating information on which you can take action on but it's like at what point is action a part of that sort of loop right where you can not only curate a list of flights that you want but then you know actually go out and and and have agency. Yeah.
2:01:40
I think one of our explorations in that space is operator, right?
2:01:44
right? It's where you kind of just feed in raw pixels from your your your laptop um into or you know from some virtual machine into the model and it it produces you know either a click or some keyboard actions right and um so there it's taking action um and I think the trouble is you know it you don't ever
2:02:03
want to mess up when you're taking action right I think the cost of that is super high um uh you you only have to get it wrong once to lose trust in in a user um And so we want to make sure that that feels super robust before we get to the point where we're like, "Hey, look, here's a tool." Um I that's so different
2:02:20
Um I that's so different than deep research because like you can wind up on some news article and read a one sentence that gets a fact wrong or the comma's in the wrong place and the numbers off and but that's just the expectation for just text and analysis.
2:02:37
And if you delegated that, yeah, you're going to expect a few errors here and there.
2:02:40
Oh, that's actually a different company name or that's a That's an old data point. There's new data.
2:02:44
Uh but very different if I book a flight and you book the wrong flight and I wind up in Chicago instead of New York. Exactly.
2:02:52
And I think the reason why we care so much about reasoning is because I think that's the path that we get reliable agents through. Sure.
2:02:58
Um you know, we've talked about like reasoning helping safety, but reasoning is also helping reliability, right?
2:03:03
It's like you imagine like what makes a model so good at a math problem?
2:03:09
It's like it's banging its head against it.
2:03:10
it's trying a different approach and then it's like adapting based on what what it failed at last time.
2:03:16
And I think that's the same kind of behavior you want your your agents to have.
2:03:19
It's like try things like it adapts and and keeps going until it it's and that's that's the humans do this every day. You're booking a flight.
2:03:27
You keep hitting an error.
2:03:27
It's not which or which form you miss, right?
2:03:29
And you're just sort of banging your head against the computer and eventually it says, "Okay, you're booked." Right?
2:03:34
So I think I think that's a great call out. Yeah.
2:03:38
Um, I mean the there's so many more questions we could go go into, but um, I'm I'm interested in the scaling of RL and kind of the balancing act between pre-training RL and inference, just the amount of energy that goes into getting a result when you distribute it over the entire user base. How is that changing?
2:03:59
And I guess um, is is are we post like really big really big runs?
2:04:04
Is this going to be something that's like continually happening online or it feels like we're moving away from the era of like oh some big development some big run happened and now we're reaping the fruit fruits of it versus a more iterative process.
2:04:20
Um yeah I mean I don't see why it has to be so right.
2:04:23
I think like if you find the right levers you can really pump a lot of compute into RL as well as pre-training.
2:04:29
Um I think it is a delicate balance though between all of these different parts of the machine.
2:04:34
Um and you know when I look at my role with Yakob it's just kind of like figure out where um how how this balance should be allocated um where the promising kind of like nuggets are arising from and and resourcing those.
2:04:47
Um yeah it's it's kind of a in some sense my I feel like part of my job as a portfolio manager.
2:04:52
Yeah, that's a lot of fun.
2:04:54
Well, thank you so much for joining.
2:04:56
This is a fantastic conversation.
2:04:57
We'd love to have you back and get deeper. Great hanging Mark. We'll talk to you soon. Yeah. Peace. Have a good one.
2:05:02
Uh, next up we have Shalto Douglas from Anthropic coming on this show.
2:05:06
I'm getting so I just giving us the update on just getting a lot of messages saying why no one cares about AI talk about the drama on the timeline.
2:05:18
Well, we do care about AI. We care a lot about AI.
2:05:21
Um, but it is a mess out there. Wow. Yeah.
2:05:26
The end of the Trump Elon era. Uh, I don't know.
2:05:29
Well, maybe maybe we have to get some people on to talk about it uh tomorrow or something. Got to do it today.
2:05:34
Anyway, uh we have Shalto from Anthropic in the studio. How you doing? What's going on? Great to see you guys.
2:05:42
Hopefully uh you're you're staying out of the chaos on the Don't Don't open any time. Don't open. Sweet child to Twitter. Move to Twitter. Mute everything.
2:05:52
Stay focused on the application.
2:05:54
Stay focused on the mission.
2:05:55
Stay focused on the next training run.
2:05:57
We really humanity really cannot afford for any researchers to open X today. What a hilarious day.
2:06:04
Anyway, I mean back 24 hours, guys. Yeah. How are you doing?
2:06:06
What what is new in your world?
2:06:08
What what are you focused on mostly daytoday?
2:06:10
And uh maybe maybe it's just a way of an intro. Yeah.
2:06:15
Um so at the moment focused really hard on scaling RL.
2:06:16
Um I mean that is the theme of what's happening this year.
2:06:19
Um and we're still seeing these huge gains where you go, you know, 10x compute increase in RL.
2:06:24
We're still getting like very distinct linear gains as the basis for that.
2:06:28
Um, and because RL wasn't really scaled anywhere close to how much pre-training was scaled at the end at the end of last year. Yeah.
2:06:33
We have like a basically a gamut of re like riches over the course of this year.
2:06:37
So, where are we in that in that RL scaling story because I I I remember the the some of the rough numbers around like GPT2, GPT3, we were getting up into like it cost a hundred million, it's going to cost a billion dollars.
2:06:51
cost a billion dollars. like it just rough order of magnitude not even from anthropic just generally like what does a big RL run cost or or how many are we talking 10k H200s or 100K like are we going to throw the same resources at it and if so how soon yeah so I think in
2:07:09
Dario's essay at the beginning of the year he said that a lot of runs were only like a million dollars back in like December thinking of like DeepSeek V3 and this kind of stuff like R1 um which means that with that's like at least two just to get to the scale of GPD4 and GPD4 was few years ago. Yeah. Right. Um Yeah. Right.
2:07:21
Um RL is also perhaps a bit more naively paralyzable and scalable than pre-training.
2:07:28
You know, pre-training you need everything in one big data center ideally or you need like some clever tricks.
2:07:32
Um RL you could like in theory like what the prime intellect folks are doing scale it all over the world out of and and so we you're held back like maybe you're held back far less than you are. Sure.
2:07:43
So everyone and their mother has a billion dollars now.
2:07:49
there are there, you know, hundreds of thousands of GPUs getting pumped all over the place.
2:07:53
I I I feel like we're not GPU poor as a as a as a society.
2:07:55
Uh maybe some companies need to justify it in different ways, but it sounds like there's some sort of uh uh like reward hacking problem that we're working through in terms of scaling RL.
2:08:07
What are all of the problems that we're working through to actually go deploy the capital cannon at this problem? Yes.
2:08:14
So I mean think about what you're asking the model to do in RL is you're asking it to achieve some goal at at any cost basically. Yeah.
2:08:23
Um and this comes with a whole host of like uh behaviors which you may not intend.
2:08:27
Um in software engineering this is really easy like to it might try and hack unit tests or whatever.
2:08:32
um in much more longer horizon real world tasks, you might ask it to say go make money on the internet and it might come up with all kinds of fun and interesting ways to do that unless you find ways to guide it into following the like principles that you want it to to obey basically um or to to align it with your like idea of what's sort of best for humanity.
2:08:51
Um and so it's actually it's a pretty intensive process.
2:08:53
It's a lot of work to find down and hunt down all the ways these models are uh hacking through the rewards and and and patch all of that and Yeah. Yeah.
2:09:00
H how uh are we going to see scaling in the number of rewards that we're rlinging against? If that makes sense.
2:09:11
I would imagine that uh at a certain point we unless we come up with like kind of like the the the genesis prompt go forth and be fruitful or something and and multiply uh the you could imagine uh training runs on on just knocking down one one problem after another and is that is that kind of the path that we're going down? I I very much think so.
2:09:34
Um there's this idea in which like you know the uh the sort of world becomes an RL environment machine in some respect um because there's just so much leverage to making these models better and better at all the things we care about.
2:09:43
Uh and so uh I think we're going to be training on on just everything in the world. Got it.
2:09:47
Um and then and then does that lead to um more model fragmentation?
2:09:54
Models that are good at programming versus writing versus poetry versus image generation or or or does this all feed back into one model?
2:10:03
Does the idea of the consumer needing to pick a model disappear?
2:10:06
Are we in a temporary period for that paradigm?
2:10:10
I think the main reason that we've seen that so far is uh because people are trying to make the best of the capital like we are all still GPU poor in many ways. Okay.
2:10:21
And people are focusing those GPUs on the sort of like spectrum of awards that they think is most important.
2:10:25
Um and look, I'm I'm a bit of a big model guy.
2:10:29
Um I I really do think that similar to how we saw with large pre-trained models before where small fine tuned models made it like had gains over the sort of GP2 GPT2 era but then were obsoleted by GP4 being generally good at everything.
2:10:44
I think to be honest you're going to see this generalization and learning across all kinds of things that means you benefit from having large single models rather than specialization or area fine-tuned models.
2:10:55
Can you talk a little bit about the transition from or any any differences between RLHF and just other RL paradigms? Yes.
2:11:02
So RLHF uh you're trying to maximize a pretty like lossy signal things like pairwise like what do humans prefer?
2:11:12
And I don't know if you've ever tried to do this like judge two language model responses.
2:11:15
I get prompted for that all the time. Right.
2:11:18
And I'm always like I don't want to read both of those.
2:11:20
I'll just click the one on the left. Exactly. Exactly. Exactly.
2:11:23
And you, you know, I click one of the random ones sometimes. Yeah.
2:11:24
Or or I click like the one that just looks bigger or I'll read the first two sentences.
2:11:29
But yeah, I'm not giving straight I'm not I'm not being I'm not doing my job as a as a human reinforcer. Exactly.
2:11:35
Human preferences are easy to hack. Yeah, totally.
2:11:37
Environments in the world are much truer um if you can find them.
2:11:42
Um so something like did you get your math question right is a very real and like true reward. Does the code compile? Right. Does the code compile? Exactly.
2:11:50
um you know, did you make a scientific discovery?
2:11:52
We're going to start we we got very little rewards right now, but pretty quickly over the next year or two, you're going to start to see much more meaningful and and long horizon rewards.
2:12:01
You're going to see models bribing the Nobel committee to win the Nobel Prize.
2:12:05
A good reward hacker reward hacking something you want to prevent, right? Exactly. It's Yeah. Yeah.
2:12:10
That's a real nightmare scenario.
2:12:13
Um, what about like there's so many different problems that we run into that feel like the it's just really really hard to divi design any type of eval.
2:12:23
Uh, the the uh my kind of benchmark that I use whenever a new model drops is just tell me a joke.
2:12:30
Yeah, they're always bad.
2:12:30
and or or or even even the latest VO3 video that went viral was somebody said like uh standup comedy joke and it was kind of a funny joke but it was literally the top result for joke Reddit on Google and then it clearly just took that joke and then instantiated in a video that looked amazing.
2:12:52
Um but it wasn't original in any way.
2:12:55
And so uh we were joking about like the RLHF loop for that is like you have an endless cycle of comedians running AI generated materials and then and then you know uh speak microphones in all the comedy clubs to feed back what's getting laughs but I mean honestly that would work pretty well actually want to hook us up with an RL loop. Yeah. Yeah.
2:13:19
But but I mean for for some of those less uh like as you go down the curve, it feels like each one gets harder and harder uh to actually tighten the loop.
2:13:26
We see this with like longevity research where it's like okay it takes a hundred years to know if you extended a human life like the yes you could create a feedback loop around that but every change is going to be hundreds of years.
2:13:38
And so even if you're on the cycle it's irrelevant for us in the context that we talk about AI.
2:13:42
So uh talk to me about like are you running into those problems or or or will there be like another approach that kind of works around those.
2:13:50
works around those. So there are a lot of situations where you can get around this by just running much faster than real time like let's say the process of building a like a giant app like building Twitter right is something that would take human months but if you got fast enough and good enough AIS you could do that in several hours right I paralyzed heaps of AI agents they're all
2:14:06
building right you know things spec and so you can get a faster reward signal in that way um in domains that are less well specified like humor I agree it's really really hard and this is like why I think in some respects like creativity
2:14:18
is uh like at the at the top end of the spectrum like true creativity is much much harder to replicate than the sort of like analytical scientific style reasoning and and that will just take more time. You know what the models
2:14:28
You know what the models actually are pretty good at making jokes about being an AI.
2:14:31
This feels weirdly fresh.
2:14:34
Um like everything else is kind of a weird copy of something like it like it just it feels like it's derivative.
2:14:39
Basically, it's trying to infer what humor is and it doesn't really understand it.
2:14:42
But jokes about being an AI are quite funny. Yeah.
2:14:44
being an AI are quite funny. Yeah. I I I I think this also might be I don't know if it was directly reward hacking, but I noticed that uh one of the new models dropped and a bunch of people were posting these like 4chan like be memes and and and they were it seemed like
2:14:59
they were kind of hacking the humor by being hyper specific about an individual that they could find information on online and so you're laughing at the fact that it's like oh wow that is like something that I've posted about and it's making a reference but it's not really that funny to me. It's other than
2:15:11
It's other than it's just like wow they really did its research like it really knows Tyler Cowan intimately which is cool but I didn't find it hilarious.
2:15:18
Um yeah yeah yeah very interesting.
2:15:21
Um let let's talk about um some sort of deep research product uh pro projects and products.
2:15:28
Um we were talking to Will Brown and he was saying like AGI is here with some of the bigger models but the but the time that AGI can feel consistent it diverges.
2:15:39
And so you could be working with someone who's, you know, 100 IQ, but they but they will stay consistent for years as an employee or they'll they'll keep, you know, living their life.
2:15:52
Whereas a lot of these super smart models are working really well and then after a few uh a few minutes of work, they the the agents kind of diverge and kind of go into odd paradigms and it feels very not human.
2:16:03
It feels like a like just a they're hyper intelligent in one way and then extremely stupid in others.
2:16:08
Um what's going on there?
2:16:10
uh what is the what is the path to extending that?
2:16:13
Is that more like having more better planning and better uh better like dividing up the task or or will this just kind of naturally happen through the RL and scale? Yeah.
2:16:24
So there's that jaggedness, right, which is what you're seeing is how we call it.
2:16:27
And I think that is largely a consequence of the fact that maybe something like deep research is probably being RL to be really good at producing a report. Yeah.
2:16:34
Yeah, but it's never been on the like act of producing valuable information for a company over a week or a month or like making sure the stock price goes up in like you know a quarter or something like this, right?
2:16:46
Like it it doesn't have any conception of how that feeds into the broader story at play.
2:16:50
It can kind of infer it because it's got a bit of world knowledge from the you know the base model and this kind of stuff but it's never actually been trained to do that in the same way humans have.
2:16:56
Um so to extend that you need to put them in much longer running much like like you know long horizon things.
2:17:04
Um and so so deep research needs to become you know like deep operate a company for a week kind of thing. Sure. Is that the right path?
2:17:10
Like it feels like the road might be there's a like the longest running LLM query used to be just like a few seconds maybe a few minutes.
2:17:22
And I remember when uh when some of the reasoning models came out people were almost trying to like stunt on it by saying like oh I asked it a hard question it thought for five minutes.
2:17:31
Now deep research is doing 20 minutes pretty much every time.
2:17:33
Um is the path two hours two days or are we going to see more uh efficiency gains such that we just get the 20-minut model the 20-minute results in two minutes and then two seconds. Yeah.
2:17:44
So this is somewhere where like inference in many respects and and priorization becomes really important.
2:17:49
So both like how fast is your inference?
2:17:51
It literally affects the speed at which you can think and the speed at which you can like like you know do the these experiments.
2:17:55
Also how easily you can paralyze becomes really important like can you dispatch a team of uh of sub agents to go and do deep research and like compile like sub reports for you so that you can do everything in parallel.
2:18:07
everything in parallel. these kinds of like uh it's it's both like there's an infrastructure question here um that that feeds up from the hardware and the chips and this kind of stuff uh to like designing better chips for you know better inference and this all this um and and an RL question of like you know how well can you paralyze and and all
2:18:24
this so I think we just need to compress the timelines compress the time compress the time frames basically so uh if I'm if I'm like an extremely big model and I'm running an agentic process like how how much my hankering for like a middlesized model on a chip or like baked down into silicon that just runs super fast because it feels like that's probably coming. We saw that with the
2:18:48
We saw that with the Bitcoin pro progression from CPU to GPU to FPGA to ASIC.
2:18:52
Do do you think we're we're at a good enough point where we can even be discussing that?
2:18:57
Because I every time I see like the latest midjourney, I'm like this is good enough.
2:19:02
I just want it in two seconds instead of 20.
2:19:03
Um, but then a new model comes out, I'm like, "Oh, I'm glad I didn't get stuck on that." Right.
2:19:07
But but yeah, like are how far away from how far away are we from, okay, it's actually good enough to bake down into silicon?
2:19:17
Well, there's a question here of baking it down to silicon versus designing a chip which is like very suited for the architecture that you care about, right?
2:19:23
Um, and baking out of silicon, unsure.
2:19:25
Like, I think that's a bet you could take, but it's a it's a risky one because the pace of progress is just so fast nowadays.
2:19:30
Um, and I I really only expect it to to accelerate.
2:19:34
Um, but designing things that make a lot of sense for the sort of trans, you know, transformers or or architectures of the future should should make a lot of sense. That's a big gap though.
2:19:45
Transformers or architectures of the future.
2:19:47
If we diverge, there's a lot of companies that are banking on the transformer sticking around.
2:19:50
Uh, what is your view on transformer architecture sticking around for the next couple years?
2:19:55
Um, I mean, look, they stuck around for five years, so they might stick around for a little while.
2:19:58
stick around for a little while. Um but there's there's different you think about architectures in terms of this balance of memory bandwidth and flops right one of the big differences we've seen here is Gemini uh recently had actually a diffusion model that they released I was about to ask the other day right so diffusion is an inherently extremely flops intensive process
2:20:15
whereas normal language model decoding is extremely memory bandwidth intensive you're designing two very different chips depending on which bet you think makes sense yeah and if you think you can make something that does flops like
2:20:24
four times faster than diffusion and like four times cheaper than you others could diffusion makes more sense so there's There's this dance basically between uh the chip providers and the and the architecture. Yeah. Both trying Yeah.
2:20:32
Both trying to build for each other but also like build for the next paradigm. It's risky.
2:20:38
Do you I I I don't know how much you've how much you've played with uh image generation, but do you have any idea of what's going on with images and chatbt?
2:20:47
It feels like there's some diffusion in there.
2:20:49
There's some tokenization, maybe some transformer stuff in there.
2:20:50
It almost feels like the text is so good that there's like an extra layer on top almost and that it's it's almost like reinventing Photoshop.
2:20:59
And and I guess the the the broader question is like it feels like an ensemble of models maybe the discussion around around just agents and textbased LLM interactions shouldn't necessarily be uh transformer versus diffusion, but maybe how will these play together?
2:21:15
Is that a reasonable path to go down?
2:21:16
Well, I think pretty clearly there's some kind of like rich information channel between even if there are multiple models there, there's like it's it's conditioning somehow on the on the other model because we've seen before like let's say when uh you know models use midjourney to produce images.
2:21:32
It's never quite perfect.
2:21:33
It can't perfectly replicate what went in as an input.
2:21:34
It can't perfectly like adjust things.
2:21:36
Um so there's a link somehow whether that's the same model producing tokens plus diffusion. I don't know.
2:21:43
Um like yeah can't comment on what OpenAI is doing there. Yeah. Yeah. Yeah.
2:21:47
Um are are there any other kind of like super wildcard longshot uh research efforts that are maybe happening even out even in academia where I mean this was the big thing with uh what was his name Gary who's talking about I forget what it was called symbolic symbol manipulation was a big one and and I feel like you know you can never count anyone out because it might come from behind and be relevant in some ways.
2:22:15
Um but but are there any other research areas that you think are like purely in the theory domain right now that are worth looking into or tracking that you know low low probability but high upside if they work? This is a tough one. This is a tough one.
2:22:34
But I will say just on the symbolic thing.
2:22:35
It's crazy how similar transformers are to systems that manipulate symbols. Sure. Sure.
2:22:39
Like what they're doing is they're taking a symbol and they're like converting it into a vector and then they're manipulating and moving like information around across them. Sure.
2:22:45
Like this this whole like uh debate that all transforms cannot represent symbols and they cannot do this.
2:22:52
I think it's it's yeah it's not real.
2:22:54
So so Gary Mark is underrated or overrated I guess. Overrated. Yeah. Yeah.
2:23:00
But uh but I mean if I if you if you twist the uh if you twist it so much you wind up with saying like well really like the the transformer fits within that paradigm and so maybe it's you know it it's you know it like the rhetoric around it being a different path was maybe false the whole time. Something like that.
2:23:20
the whole time. Something like that. Um but but I as I remember that debate, it was really the the idea of compute scaling versus almost like feature engineering scaling and and will the progress scale with human hours or GPUs essentially and that has a very different economic equation and it and
2:23:42
it feels like there's been there's been some rumblings about maybe with a data wall we'll shift back to being human labor bound, But do you think that there's any chance that that's relevant in the future or is it just algorithmic progress married with bigger and bigger data centers in the future? So I'm
2:23:59
So I'm pretty bitter lesson built in sense that I do think removing as many of our biases and our like clever ideas from the models is really important just like freeing them up to learn.
2:24:11
Now obviously there's like there is clever structure that we put into these models such that they're able to learn in this extremely general way and that uh but I I am more convinced that we will be computebound than than we will be like human researcher uh human researcher hour bound on on this kind of thing like we're not going to be feature engineering and this kind of stuff.
2:24:32
We're going to be trying to devise incredibly flexible learning systems. Yeah, that makes sense.
2:24:37
Yeah, that makes sense. Um, on the scaling topic, part of I I I part of my like worry is that the the oos get so big that they turn into these mega projects that are uh that at a certain point you're bound by the laws of physics
2:24:54
because you have to move the sand into silicon chips and you have to dig up the silicon and at a certain Yeah, there's only so much sand and like the math gets really really crazy just for the amount of energy required to to move everything around to make the the big thing. Uh
2:25:07
Uh where are you on on how much scale we need to reach AGI?
2:25:14
How whether or not we will see um like the laws of physics start acting as a drag on progress because uh it certainly feels exponential.
2:25:23
We're feeling the exponentials, but uh a lot of these turn into sigmoids, right?
2:25:27
Um so I think we've got what like two or three more before uh before it gets really hard.
2:25:34
uh Leopold has this nice table at the end of his situational awareness where I think like 2028 or something is when uh under really aggressive timelines that you get to 20% of US energy production.
2:25:43
It's pretty hard to go exponentially beyond 20% of US energy production.
2:25:47
Um now I think that's enough.
2:25:51
Every indication I'm seeing says that's enough.
2:25:54
Now then there might be some complex uh you know data engineering, reward engineering, this kind of stuff that goes into lots there's still a lot of algorithmic progress left to go.
2:26:03
Uh but I think that with those extra s we get to basically uh a model that is capable of assisting us in doing research and software engineering. Yeah.
2:26:13
Which is the beginning of the self. Yeah. Interesting.
2:26:17
Is that just a coincidence?
2:26:19
Like this feels like one of those things.
2:26:21
This feels like one of those things where like the moon is the exact same size as the sun in the sky.
2:26:25
It's like oh it just happens that AGI happens within this time.
2:26:27
Like did you have you unpack that anymore because it feels convenient.
2:26:32
Not not to you know I know I mean there's there's a lot of weird conveniences or like weird it's a good sci-fi story let's say. Totally.
2:26:40
You know we've got uh you know Taiwan and between China and the US and it produces the most valuable material in the world and that's locked between the two. Incredible plot. Incredible plot. Yeah.
2:26:49
really bad for the people that don't think of that don't believe in simulation theory.
2:26:52
It really feels like all this is scripted. It's fascinating.
2:26:56
Um talk to me more about um uh getting to an ML engineer in AI and and kind of that reinforcement.
2:27:04
AI and and kind of that reinforcement. I imagine that you're using AI codegen tools today and and anthropic is broadly and everyone is um but but uh what are you looking for and what are the what's the shape of the the spiky intelligence where do they fall flat and what are you looking to kind of knock down in the
2:27:23
interim before you get something that's just like go yeah so I mean we definitely use them the other night I like I was a bit tired I asked her to do something just sat watching it in front of me working for half an hour it was great it was truly weird experience particularly when you look back a year ago and we're still copy pasting stuff between a chat window and and you know a code file. Yeah. Um uh what I like Yeah.
2:27:41
Um uh what I like meters evals for this kind of stuff.
2:27:45
So they have a bunch of evals where they measure like the ability to write a kernel, the ability to run a small experiment and improve a loss.
2:27:51
Um and they have these nice progress curves versus humans.
2:27:55
Uh and I think this is maybe the most accurate reflection of like what it will take for it to really help us um at at doing progress.
2:28:00
And there's a mix here like where they're not so great at the moment is like large scale distributed systems engineering, right?
2:28:08
right? like debugging stuff across heaps and heaps of accelerators and like the way the feedback loops are slow and you like if your feedback loop is like an hour then it's worth you spending the time on on doing something feedback loop 15 minutes if it's and and for context there the hourong feedback loop is just
2:28:26
because you have to to actually compile and run the code across everything that long spin up all your machines or you need to like like you need to like run it for a while to see if something's going to happen like at that point in time you're still cheaper than the chips and So uh you're you're you're sort of it's better that you do it. Um but for
2:28:40
Um but for things like you know kernel engineering uh or for uh like you know actually even just understanding these systems incredibly helpful like I one thing I regularly do at the moment is in parts of the codebase in like languages that I'm unfamiliar with or some stuff like this.
2:28:54
I'll just ask it to rewrite the entire file but with comments on every line. Gamechanging.
2:29:00
It's like comments on Yeah.
2:29:00
Or just hunt through like thousands of files and explain how everything interacts to me. Draw diagrams. This kind of stuff. It's really Yeah. Yeah.
2:29:07
How important is a bigger context window uh in that in that example you gave?
2:29:12
That feels like something that's important and yet it I I just naively like Google's the one that has the million token context window.
2:29:19
I imagine that all the other frontier labs could catch up, but it seems like it hasn't been as much of a priority as maybe like the PR around it sounds like. Is that important?
2:29:27
Should we be go should we be driving that up to like a trillion token window?
2:29:31
Um is that is that just going to happen naturally?
2:29:35
There's a nice plot in the Gemini 1.
2:29:35
5 paper uh where they show the like loss over tokens as a function of context length and they show the loss goes down quite steeply actually as you put more and more and more like of a code base into context you get better and better and better at predicting the rest. Yeah, that makes sense.
2:29:48
With context length, it's a cost.
2:29:49
Um you know the way transformers work is that uh there's you know you have like this this memory that is proportional the KB cache is proportional to how much context you've got.
2:30:01
Uh, and so you can only fit so many of those into like your various chips and this kind of stuff.
2:30:06
Uh, and so longer context actually just costs more because you're taking up more of the chip and you're sort of like you could have otherwise been doing other requests basically.
2:30:13
So bringing it back to the custom silicon, is that a unique advantage of the TPU?
2:30:16
Is is that something that Google has has thought about and then wound up to put themselves in this advantage position or is it a durable advantage even? Yeah.
2:30:24
So TPUs are good in many respects partially because you can connect hundreds or thousands of them really easily across really great networking.
2:30:32
Um whereas only recently has that that been true for uh GPUs like GPUs. Yeah.
2:30:35
With MVLink and like the MVL 72 stuff. Okay.
2:30:38
Um so it used to be like eight GPUs in a pod and then uh like you connect them over worse uh interconnect and now you can do 72 um and and then it breaks down.
2:30:47
With Google Tus you can do like 4,000 8,000 over really high bandwidth interconnect um in one pod.
2:30:53
Um and so that that is helpful for things like just just general scaling in many respects.
2:30:57
Um I think it's this is doable across any chip platform but uh it is a is an example of like somewhere that being fully vertically integrated is a is isn't a benefit. Yeah. Yeah, that makes sense.
2:31:10
Uh talk to me about Arc AGI. Why is it so hard? It seems so easy.
2:31:13
It does seem easy, doesn't it?
2:31:15
Uh that's a Well, it certainly seems like more more evaluatable than tell me a funny joke, right? Yeah. Yeah.
2:31:22
And I I mean I think if you RLE on ArcGI then it would you probably get super human at it pretty fast.
2:31:27
Um but I think we're all trying not to RL on it so that it functions as like an interesting held out. Sure. Okay.
2:31:34
Wait, is that just an informal agreement between all the labs?
2:31:38
Basically you know we try and have a sense of honor between us. That's honor. Um that's amazing.
2:31:43
How many people on earth do you think are getting the full potential out of the publicly available models?
2:31:49
Because we're now at a point where we have, you know, billion plus people are using AI almost daily and yet I have to imag my sense would be it's maybe like 10,000 20,000 people on the entire planet are getting that sort of full potential.
2:32:03
But I'm curious what your assessment would be.
2:32:05
Yeah, I I completely agree.
2:32:07
I mean, I think that even I don't get the full full potential out of these models often.
2:32:10
out of these models often. Um, and I I think as we shift from you're asking a questions and it's giving you sensible answers to you're asking it to go do things for you um that might take hours at a time and you can really like priorize and spin that we're going to
2:32:25
hit like yet another inflection point where even less people are like really effectively using these things because it's basically going to require you to like it's like Starcraft or Dota like it's going to be like your APM of like managing all these agents and that's going to be quite process. Yeah. Yeah.
2:32:39
Starcraft is such a good example.
2:32:39
But you think you're just absolutely crushing it and then you realize like there's an entire area of the map you're just getting destroyed on. Yeah, exactly.
2:32:48
Um it's such a good it's such a good comp. That's great. Anything else, Jordy?
2:32:52
Um I think that's it on my side.
2:32:55
I mean I I would like this to be an evolving conversation.
2:32:58
Yeah, this was fantastic.
2:32:59
We'd love to fun conversation. Absolutely. It's really fun. Love back on.
2:33:03
Yeah, we'll talk to you soon. Cheers. Have a good one.
2:33:08
All right, we got the worst possible AI day given the world. Yeah.
2:33:11
So, for context, folks, we are going to be doing a live time timeline and turmoil segment at 2 p. m. PST. Let's do it.
2:33:17
Uh, so if there's posts you want us to cover, you can go send them. I'll put this.
2:33:22
We can review a few a few more. Pull one up.
2:33:26
I'm going to do some ads because we got image here coming in the temple in just a few minutes.
2:33:30
Uh, first let me tell you about numeral sales tax on autopilot.
2:33:34
spend less than five minutes per month on sales tax compliance. Go to numeroh. com.
2:33:41
Um, very excited for them. Also, go to public. com.
2:33:44
Investing for those who take it seriously.
2:33:46
They have multi-asset investing, industryleading yields, and they're trusted by millions. Um, millions.
2:33:52
In other news, uh, Tim Sweeney continues to battle Apple.
2:33:55
Apparently, if you search for Fortnite on the Apple App Store, uh he says, "Hey kids, looking to play Fortnite?
2:34:03
Try this crypto and stock trading app instead.
2:34:06
Rated for ages four plus, courtesy of Apple App Store ads."
2:34:08
Uh obviously, I'm going to give you the latest Trump ter uh Elon post four minutes ago.
2:34:13
The Trump tariffs will cause a recession in the second half of this year. Wow.
2:34:17
Um somebody else was saying, "Can I finally say that Trump's tariffs are super stupid?"
2:34:27
Somebody else is posting.
2:34:27
Mad's posting is saying it's a Xiin Ping.
2:34:29
He says, "Bro, you seeing this?"
2:34:31
And it's um Putin on the other end. He's just looking at it. Hold up. Got a line.
2:34:35
And it's uh we'll start pulling some of these up. Ridiculous.
2:34:44
Uh what else is going on here?
2:34:44
This is the president versus Elon.
2:34:47
Nal says Elon's stance is principle. Principled.
2:34:52
Trump stance is practical.
2:34:52
Tech needs Republicans for the present.
2:34:55
Republicans need tech for the future. Drop the tax cuts. Cut some pork. Get the bill through. This is so crazy.
2:35:04
Antonio Garcia says, "Remember, there's FU money and then there's F the world money."
2:35:12
Will Stansel says, "Imagine being the ICE agents suiting up for your biggest mission of all time."
2:35:18
Right now, people are saying that Trump's going to deport Elon. on back to South Africa.
2:35:25
Um, Will Depuse says, "Time to drop the really big bomb growing Daniels in the Epstein file."
2:35:29
That is turned into a coffee pasta. That is the real reason. Terrible. Oh, no. Oh.
2:35:38
Um, Delian, uh, uh, we had a question from a friend of the show.
2:35:46
said, "The real question is if Tesla is down 14%, how could SpaceX and OpenAI be trading if they were how would they be trading if they were public?"
2:35:53
The the real thing here is it's bad for everyone, right? DJT is down. Trumpcoin is down.
2:36:00
Nobody's really winning here. China is up. Yeah.
2:36:02
Uh Shamagu I mean I'm just saying like at a high level you know China is the big beneficiary here of um uh Sarah Guo says if anyone has some bad news to bury might I recommend right now? Yes. Yes. Yes.
2:36:17
If you have if you uh what what's the canonical bad startup news like oh yeah you missed earnings or something drop it now.
2:36:28
Verse Kramer says, "Bill Aman is currently writing the longest post in the history of this app." Okay.
2:36:36
And we have uh we have a video from Trump here.
2:36:39
If we want, I can throw it in the tab and and we can share it on this on the stream and and uh react to live.
2:36:46
Lex Freriedman says to Elon, "That escalated quickly. Triple your security.
2:36:52
Be safe out there, brother.
2:36:52
Your work SpaceX, Tesla, XAI, Neurolink is important for the world.
2:36:57
We need to get Elon on the show today.
2:36:58
If somebody's listening and can make that happen, I would love to hear from Max Meyer says, "So, I got this wrong.
2:37:04
I didn't say it never happened, but I thought it wouldn't."
2:37:07
I'm floored at the way this has happened. Yeah.
2:37:08
Uh he didn't think they would have a big breakup.
2:37:11
Uh many people didn't think they would have a big breakup.
2:37:14
Even just earlier this week, it seemed like they might just have a a somewhat peaceful exit.
2:37:18
Um, Trump just posted a little bit ago, "I don't mind Elon turning against me, but he should have done so months ago.
2:37:26
This is one of the greatest bills ever presented to Congress.
2:37:30
It's a record cut in expenses, $1.
2:37:32
6 trillion, and the biggest tax cut ever given.
2:37:35
If this bill doesn't pass, there will be a 68% tax increase, and things far worse than that.
2:37:41
I didn't create this mess.
2:37:41
I'm just here to fix it."
2:37:44
Um, anyways, lots going on.
2:37:44
Uh, let's go to this uh this Trump video.
2:37:49
I want to see what he that I've seen and I'm sure you've seen regarding Elon Musk and your big beautiful bill.
2:37:55
What's your reaction to that?
2:37:57
Do you think it in any way hurts passage in the Senate, which of course is your seeking?
2:38:02
Well, look, you know, I've always liked Elon and it's always very surprised.
2:38:05
You saw the words he had for me, the words and he hasn't said anything about me that's bad.
2:38:11
I'd rather have him criticize me than the bill because the bill is incredible.
2:38:16
Look, Elon and I had a great relationship.
2:38:20
Uh, I don't know if we will anymore.
2:38:22
I was surprised because you were here.
2:38:24
Everybody in this room practically was here as we had a wonderful send off.
2:38:28
He said wonderful things about me. You couldn't have nicer. Said the best things. Uh, he's worn the hat.
2:38:33
Trump was right about everything.
2:38:35
And I am right about the great big beautiful bill.
2:38:38
But I'm very disappointed because Elon knew the inner workings of this bill better than almost anybody sitting here, better than you people.
2:38:48
He knew everything about it.
2:38:48
He had no problem with it.
2:38:49
All of a sudden, he had a problem.
2:38:51
And he only developed the problem when he found out that we're going to have to cut the EV mandate because that's billions and billions of dollars. And it really is unfair.
2:38:58
We want to have cars of all types. electric.
2:39:02
We want to have electric, but we want to have gasoline, uh, combustion, we want to have different, we want to have hybrids.
2:39:08
We want to have all We want to be able to sell everything.
2:39:12
He hasn't said bad about me personally, but I'm sure that'll be next.
2:39:15
But I'm I'm very disappointed in Elon. I've helped Elon a lot. Mr.
2:39:20
President, did he I just want to clarify, did he raise any of these concerns with you privately before he raised them publicly?
2:39:25
And this is the guy you put in charge of cutting spending.
2:39:28
Should people not take him seriously about spending now?
2:39:31
Are you saying this is all sour grapes?
2:39:32
No, he worked hard and he did a good job.
2:39:33
And I'll be honest, I think he misses the place.
2:39:37
I think he got out there and all of a sudden he wasn't in this beautiful oval office and he was and he's got nice offices, too.
2:39:44
But there's something about this when I was telling the chancellor, folks, this is where it is. People come in. Breaking news.
2:39:49
Delian as Bruhov is joining us in the temple for some live reactions. Surprise guest.
2:40:00
I can't even spell surprise guest.
2:40:02
I'm so excited about this. Surprise guest.
2:40:04
In other news, 11 Labs dropped a new product two hours ago.
2:40:12
In other news, $2 million seed round. Stop it. Stop it. We love Labs.
2:40:17
No, they'll they'll keep grinding, but just launch again tomorrow.
2:40:23
They're going to have to launch again.
2:40:24
Start shooting a new vibe reel.
2:40:26
start shooting a new writing a new blog post because no one's going Lulu says yes delay the launch on TVPN.
2:40:34
So basically right now I can just pull up and just refresh.
2:40:38
I'm going to just be refreshing true social. What are you doing?
2:40:42
So okay, Jord's on truth social. I'll be on X. Give us your reaction. Tell what's going on.
2:40:47
I mean at some point I was like I'm just you know sort of scrolling X and I like tuned into you guys like an hour ago and I was like they're talking about some AI thing.
2:40:55
I was like at some point they're going to switch to like news and was like and then I was watching it and I was like okay like got Sean resisted I fought it for like for like a half an hour.
2:41:04
Um but we couldn't do it.
2:41:04
Um but yeah give us your quick reaction.
2:41:07
I mean I'll always you know sort of give it from the uh you know sort of space angle.
2:41:13
You know, it's amazing that, you know, um, uh, how much the world has shifted since, you know, Friday of last week where it was, you sort of presumed that Jared Isaacman was going to be the, you know, sort of NASA admin to today.
2:41:25
Um, it was released that the Senate reconciliation package readded uh, budget back into NASA largely for the SLS program, which was basically the program that, you know, sort of Jared and Elon were, you know, sort of largely um, advocating to, you know, sort of completely shut down.
2:41:38
Um so you know the the the um it is already show like you know the sort of counter reaction you know is already showing up you know in in policy.
2:41:49
Sorry SLS program is that space shuttle or no uh sorry that's the SLS uh launch rocket.
2:41:54
Um it is based off of old space shuttle hardware but it is basically the internal um you know sort of NASA run competitor effectively to like a you starship heavy you know launch rocket.
2:42:05
Um and so you know because it was you know sort of generally behind budget behind schedule and there are so many commercial heavy lift rockets coming uh online um the default was to cancel that is largely you know sort of a Boeing based program and so you know if you look at you know you know 3 months ago
2:42:19
you know when um they were announcing the F47 program you know Elon walks into the secretary of the air force's office obviously he'd been you know sort of ranting against um you know man fighter jets and believing that that shouldn't be what you know uh be what the department is prioritizing. 30 minutes
2:42:31
30 minutes after that meeting was when they announced the F-47 program.
2:42:33
And so now you're seeing basically like the equivalent in space where you know uh you know that you was obviously awarded to Boeing uh Boeing was the is the largest prime behind SLS uh you know Boeing basically you know um is going to be the biggest winner of you know NASA refunding IOT SLS and Jared Eisman not being NAS administrator.
2:42:51
So, tying this back to the timeline Trump posted less than 30 minutes ago, in light of the president's statement about cancellation of my government contract, SpaceX will be begin decommissioning its Dragon spacecraft immediately. Break that down.
2:43:07
I mean, that just means that we no longer have a vehicle that can go to the International Space Station.
2:43:10
We no longer have a vehicle that can bring astronauts up and down.
2:43:12
Um, you know, we also don't have a vehicle that can de-orbit the International Space Station safely, right?
2:43:19
That the Dragon was expected to be able to do that.
2:43:21
expected to be able to do that. So what that means is you know if you guys remember all the memes about stranded you know from last year around Boeing Starlininer um it now means that the space station you know itself is basically you know sort of stranded and
2:43:32
that's like you know one of the government contracts obviously that you know sort of basics is involved in Elon I've heard generally like just wants to shift all things to Starship anyways and so in some ways was probably kind of looking for an excuse to uh you know sort of shut down Dragon and refocus energies. There's also a part of it
2:43:44
There's also a part of it where it's like look he is like kind of independent in the space world in that you know Starlink's um total topline revenue is going to be passing the NASA budget um in the next year or two and so in terms of like size of you know state actor that can influence space you know his own company is basically about to become you know as large of an actor as like the entire United States.
2:44:03
So I don't think there's going to be like a deescalation here like you know my my you know estimation is like on both sides it's going to continue to escalate.
2:44:12
Um, you know, if we thought that we lived in dynamic times, you know, when Trump got into office, it's going to be even more dynamic when there's like the dynamism will continue until morale improves.
2:44:23
Elon the center, AOC the progressive populist and Trump the you know, sort of conservative populist and oh man, it's uh it's on the timeline.
2:44:35
I mean the I just have so many questions, right?
2:44:37
How does this impact Golden Dome, right?
2:44:39
What's Boeing stock doing?
2:44:42
um is is will Golden Dome even be a viable project without SpaceX?
2:44:46
It's it's I think there's just going to be more resistance probably to working with, you know, sort of upstarts because they would be ones that would probably be more likely to collaborate, you know, sort of with, you know, SpaceX.
2:44:56
And so, um so I mean it it feels like it feels like Boeing would be a logical beneficiary of this turmoil and yet they're down today.
2:45:06
They haven't really popped. Oh, really?
2:45:08
I mean, I'm not obviously, you know, one to give like, you know, public. Yeah. I I I know.
2:45:11
I'm just trying I'm working through it myself and it's surprising like Tesla to drop and Boeing to pop basically off. Yeah. Yeah.
2:45:19
That would be the expectation.
2:45:20
But there there must be something because there there it feels like this is purely interpersonal between Elon and Trump and not it's not like oh Boeing was secretly behind the scenes the whole time lobbying even more effectively.
2:45:31
It doesn't got the well where's the tinfoil hat? It's over there.
2:45:36
Maybe we need a tinfoil hat segment. Who knows?
2:45:38
But yeah, I mean when you're at Boeing world, it's like, hey, we're only down 1%. Let's go. The coup of the century.
2:45:45
My question is, has has there ever been a crash out of this magnitude ever in history?
2:45:49
Well, in internet history, when when Elon and Trump became friends, honestly, world scale probably world history equivalent.
2:45:58
I feel like there was something in like maybe era in the United States where um you know, crashing out used to mean calling up the New York Times and just ranting.
2:46:07
Now you can just live post like all your reactions and it's just all real time.
2:46:12
This is like crash are actually intensifying.
2:46:17
You actually want to be long crash outs over the next social media platform that they own.
2:46:22
So you know you got to be on both ex and truth social to like stay on top of things. Yeah. Yeah.
2:46:26
top of things. Yeah. Yeah. I actually did like a deep research report a while back on like has the richest man in America ever been close with the US president going back to like you know was Rockefeller particularly close and and because the narrative was like oh this is like so unprecedented and in fact it is unprecedented in fact oh really I would have guessed that like Rockefeller was close me too me too
2:46:49
that's what I was going for was like no I imagine this is always this is always close but no I I I think because the president has become more powerful globally your your your your your point about uh you know, mayor of America, dictator of the world, like it becomes increasingly valuable for the richest man to have a close alliance and so it's become more I I don't know exactly how accurate that research was. It's totally
2:47:09
It's totally possible that like behind the scenes Rockefeller was really close to the president at the time and we just didn't write about it in the history books.
2:47:17
There certainly aren't very many anecdotes about the richest man in America going on like a ready for AP US history 2050 you know exam where you know Elon Musk called the president at the time a potential pedophile was it a about Epstein island bave in the Philippines c what a mess no so Pavle had a good post he was quoting the the big bomb uh from Elon he said hypothetical question about the USA's power structure.
2:47:47
Is the man with the most access to capital more or less powerful than the political head honcho? Purely hypothetical?
2:47:53
It's a good uh question to ask.
2:47:56
I mean, I think both like uh archetypes have uh grown both in absolute power but also in relative power to the rest of the globe basically since the guilded era. Right?
2:48:07
If you think about like the president of the United States in 1925, I'd say pretty darn powerful, but like there was clear like, you know, it was a, you know, sort of multipolar, you know, sort of world.
2:48:19
Argentina was pretty darn rich at the time.
2:48:20
Obviously, Europe was still, you know, sort of recovering from World War I, but UK was generally, you know, sort of doing well.
2:48:26
Like, it was not, you know, it's clear that there was a, you know, sort of huge, you know, outweight effect.
2:48:28
huge, you know, outweight effect. And then if you look at probably the you know sort of biggest you know industries of the time you I don't think you could claim that even like Standard Oil at its peak I'd have to go look at the exact numbers but that like it had the size of budgets relative to like you know the like US government in terms of you know
2:48:44
sort of budgets right versus I feel like now for the first time you both have you know US president extremely extremely powerful and then you have like you know sort of mag seven effectively like the size of you know sort of you know states like their you know effective their own state governments and then also just more bureaucracy, more red tape. So,
2:48:59
So, like I I when I think about the 1920s like Robert Barren, it's like it is the it is the you can just do things era.
2:49:07
And so, you want to build a railroad like yeah, you might need to get like one rubber stamp, but it's not going to be 10 years and tons of lobbying and all this different stuff.
2:49:14
So, you can kind of just go uh you can just go wild.
2:49:15
You know, it's bad when Kanye is saying, "Bro, please know we love you both so much."
2:49:22
It's just like the voice the voice of reason is Kanye West. Yes. Yes. Thank you.
2:49:27
We didn't bring them together and uh you know form a peace treaty.
2:49:31
Nikita Beer just added his pronouns back to his bio. Let's go.
2:49:32
That's He's got a rubber band.
2:49:35
Elon's got a rubber band all the way back to you know sort of extreme wokeism straight back to uh you know sort of super climate change and energy.
2:49:44
Somebody's sharing re-sharing the picture of the the cyber truck blown up in front of the Trump Tower in Vegas and it's just like this this is in real life.
2:49:51
It was foretold prediction, but it was a question of like when and of what magnitude, not if. Mhm.
2:49:57
Um, always bad if if Vladimir Putin is operating to negotiate between uh President Trump and Elon.
2:50:07
I think I think a lot of the world is is waiting for Roy Lee's takey and the Cle army.
2:50:16
That's who we want people have been asking him to get involved with geopolitics. Wow.
2:50:21
Um I love the uh Shil Moha put up a uh you know sort of meme about Narenda the like prime minister of India.
2:50:29
Um you know he basically uh copied and pasted the uh Trump truth social post about negotiating peace between India and Pakistan when it wasn't like actually fully negotiated you know posting about you know uh ne negotiating a ceasefire between Elon and Trump.
2:50:43
Funny thing is like truth social you can just read all of Trump's posts without creating an account.
2:50:49
truly shows that like I would think that you would have to make an account to read them all, but they just it's not gated at all.
2:50:55
This could be the biggest, but you know, they clearly I don't think they care about monetization.
2:51:01
Um, Bitcoin is actually uh falling alongside falling falling um wow Bitcoin falling, Boeing falling, Tesla falling.
2:51:10
Who's the biggest winner of the day?
2:51:12
That's I think it's China. China. Yeah, China. China. Bitcoin really sold off.
2:51:17
It's It's down 3% today at 101K.
2:51:22
So, still up, but you know. Yeah. Rough.
2:51:24
Winnie the Pooh just dipping his uh you know hands in that pot of honey just snacking away watching from the sidelines. Yeah. Let's see. Chinese stocks. US stocks. Chinese. I can't find it. Okay.
2:51:35
That's probably my commentary on the day, boys. This is great. It was fantastic. Thanks for jumping on.
2:51:39
Thanks for hopping on so quickly. Cool.
2:51:40
Well, Aaron Rogers signed a one-year deal with the Steelers. Announce announced. Thank you. announced an hour ago.
2:51:47
Let's give it up for Aaron Rogers.
2:51:50
Do we do we have EMTT in the waiting room?
2:51:51
Uh I' I've I've messaged him. It's absolute chaos.
2:51:55
We'll see if he can uh he can hop back on.
2:51:57
Uh we don't have him right now.
2:51:57
Uh ready if you can hop on. Sorry about the chaos.
2:52:04
Um uh we're we're live streaming.
2:52:08
We we are we are full streamers.
2:52:08
That that was the moment where Yeah, it was like this is the point of TVN. I send him an invite. Let him jump in.
2:52:15
Uh hopefully we can get EMTT in. It was very chaotic.
2:52:18
Um but uh you know it's a busy time.
2:52:22
My only hope for both Trump and Elon is that they can get some sleep.
2:52:27
They both go to eatleleep.
2:52:29
com/tbpn, get a pod5 ultra, take advantage of the 5-year warranty, the 30 night risk-free trial. They got free returns. They got free shipping.
2:52:39
This is really the perfect time to do ads.
2:52:40
You know, I think that's what that's what they both that could unify everyone.
2:52:44
I I I hope that both Trump and Elon have eight sleeps tonight.
2:52:47
Uh if they sleep at all, but even just resting on it, they're going to uh even just resting on it would be good. Uh but yeah, let's see. Let's see.
2:52:56
Um we can also go through I don't know.
2:52:59
I don't even know what to do.
2:53:01
There's a bunch of random timeline we have.
2:53:02
Lex Freriedman is saying we need to do a podcast with Elon and Trump. He's done both.
2:53:09
Something he's done both.
2:53:09
Something tells me that they're not going to jump on the show today. I don't think so.
2:53:13
And he he'll be like, "What about love?" Yeah, you guys. I mean, it is wild.
2:53:19
There was that Elon like less than two months ago was saying, "I love Trump as much, you know, as a as a straight man can love another man." Yeah. Something of that sort.
2:53:28
Um, it's just odd that the band-aid got ripped off so aggressively, so fast, you know, like there could have been there could have been like a smooth deescalation with like the This is the fast takeoff.
2:53:41
This is the fast takeoff scenario.
2:53:43
We are in the fast takeoff scenario.
2:53:45
Anyway, maybe they should book a wander, work it out together.
2:53:47
They could find their happy place.
2:53:49
They could book a wander with inspiring views, hotel grade amenities, dreamy beds, top tier cleaning, and 24/7 concier service.
2:53:56
It's a vacation home, but better. Go to wander. com. use code TBPN.
2:53:58
Please let them know that we sent you.
2:54:00
Uh Lee Elm says, "Elon literally has me dying laughing.
2:54:06
Trump said he was going to take away his government contracts and Elon said, uh haven't you been to Epstein's island, sort of a bridge that um absolute chaos."
2:54:19
Nikita says, "Hey, Blue Sky users, come on in. The water's warm."
2:54:25
David Freriededberg says China just won, which I think is the right take. I don't know.
2:54:33
I don't know what to I don't know what to think.
2:54:36
There's not that much There's not that much here to There's not that much meat to analyze.
2:54:41
Uh I mean, it's certainly interesting to see how important the the subsidies are and the electric vehicle mandates are.
2:54:50
I mean, it always feels like the best product wins in a lot of these scenarios.
2:54:57
And if Tesla was making it through the political chaos of arguably their biggest constituency, um, electrical vehicle buyers, electric vehicle buyers being upset about the Trump Elon, um, alignment, uh, I wonder, you think everyone's, you think all the anti-Trump people are gonna are gonna buy Teslas now?
2:55:19
It's like really make a statement like I'm I'm anti-Trump. I stand with Elon.
2:55:23
So they'll have the bumper sticker that says I bought this after the crash. After the crash. Exactly.
2:55:29
I bought this after the crash.
2:55:32
Uh I am aligning with Elon.
2:55:34
There's a post here from Goth and it says explaining the Trump Elon crash out in 10 years.
2:55:39
And it's and it's the Joe Biden quote when he says it was like 15 911s.
2:55:46
Um yeah, it certainly is. What?
2:55:52
Yeah, it's hard to process.
2:55:52
I mean, this gonna have massive posted for Elon stance is principled.
2:55:56
Trump's stance is practical.
2:55:59
Tech needs Republicans for the present.
2:56:01
Republicans need tech for the future. Drop the cats. Drop the tax cuts. Cut some pork. Get the bill through. Interesting. Yeah.
2:56:12
Somebody somebody named Logan made an image of Trump putting a a bumper sticker on his red Tesla saying bought it before Elon went crazy. Yep. Wait, who is that?
2:56:24
Is that from the Republican perspective? Oh, Trump's doing that.
2:56:27
Yeah, Trump's putting it on saying he has the he has the red Tesla.
2:56:31
Uh Sean Pur says, "Sad day for America, but this is outstanding content." It is.
2:56:36
And even Taylor Loren agrees with that. Yep.
2:56:39
Um Bill Aman's ripping posts.
2:56:39
All right, I'm going to put some posts and we'll and we'll Yeah.
2:56:44
Is Bill Aman actually live posting through this?
2:56:46
No, people are just speculating.
2:56:49
Um, there was actually a post in the Somebody says clears throat.
2:56:55
Truly, we live in a dogey doge world. Where where was this?
2:57:01
Uh, Cersi says, "I know Elon and Trump are the real deal because of how passionately they argue.
2:57:07
No couple fights this vic viciously.
2:57:09
if there isn't a mutual obsession underneath.
2:57:11
Uh so there's a there's a piece in the Wall Street Journal earlier this week that we didn't get to cover.
2:57:17
Um but it was it was talking about it it kind of it kind of predicted a little bit of this crash out.
2:57:23
And so uh it's from the opinion the editorial board at the Wall Street Journal says whose pork do you mean Elon Musk trashes the House bill that cuts subsidies for Tesla.
2:57:33
Um, Elon Musk's work at the at Doge made him persona nongrada in the beltway and most criticism was nasty and unfair, says the editorial board.
2:57:43
That's what Washington does to outsiders who want to shrink its power.
2:57:47
Like, it was always expected that if you come in and try and cut anything, you're going to see uh you're going to see push back from folks who don't want cuts.
2:57:54
Um, that's what Washington does to outsiders.
2:57:57
But that makes it all the more unfortunate that Mr.
2:57:59
Musk is now joining the beltway crowd in trying to kill the House tax bill.
2:58:03
This massive, outrageous, pork- filled congressional spending bill is a disgusting abomination, the Tesla CEO tweeted Tuesday as the Senate begins considering its version of budget reconciliation.
2:58:16
Shame on those who voted for it. You know you did wrong. You know it. Porkfield spending bill. What else is new?
2:58:22
The House bill could be far better on tax policy and spending reduction.
2:58:25
The Senate could be making improvements such as reducing the $40,000 state and local tax deduction cap, scrapping the tax on exclusion for tips and overtime uh and and overtime, and reducing the federal Medicaid match for able-bodied results, able-bodied adults.
2:58:41
But the House bill does avoid a $4.
2:58:43
5 trillion tax hike next year and cut spending by some 1.
2:58:46
5 trillion over 10 years, making some useful reforms to Medicaid, student loans, and food stamps.
2:58:52
It also ends most of the inflation reduction acts green energy subsidies. Ah, but Mr.
2:58:57
Musk does not want to eliminate that pork.
2:58:59
There is no change to tax incentives for oil and gas, just EV solar.
2:59:03
He said on Axe last week, retweeting another user post that said, "Slashing solar energy credits is is unjust.
2:59:10
But what's more unjust is the damage that's been done to people's lives during storms and blackouts because ultimately you can't replace a human life." Mr. Mr.
2:59:16
Musk is paring the climate lobby's spec specious claim that tax breaks like depreciation that are available to all manufacturers are a special benefit for the oil and gas industry.
2:59:27
But it's rich that he is denouncing the h the house bill for not cutting spending enough while also fuming that it kills green energy tax credits as if they are a matter of life and death for Tesla.
2:59:38
Tesla Energy, its battery and solar division, tweeted last week that abruptly ending the energy tax credits would threaten America's energy independence and reliability of our grid.
2:59:49
We urged the Senate to enact legislation with a sensible wind windown of 25D and 48E, which refers to the tax credits for residential and large scale clean energy products.
2:59:57
Both credits are important for Tesla, which derives an increasing share of its revenue and profit from selling solar and battery systems to homeowners and utilities. I didn't realize that.
3:00:07
But the House bill waits until 2030 to phase out a tax credit for battery production, which benefits Tesla's electric vehicle and storage business.
3:00:15
So, the Senate should end it sooner, says the Wall Street editorial board. Mr.
3:00:19
Musk has done yman's work trying to reduce the federal bureaucracy and improve how government works.
3:00:27
So, the editorial board is excited and happy that he's been working on that.
3:00:30
Um, he's right that both parties in Congress are spend thrifts.
3:00:35
Um, but one reason for that is because whenever Congress tries to cut something, special interests scream as Mr.
3:00:43
Musk is doing over green subsidies.
3:00:46
If the House bill fails, there won't be any cuts, only a huge tax increase. Is that what Elon wants?
3:00:50
And so they're asking the question, interesting. Sweet.
3:00:54
Well, we got a bunch of bangers in the tab.
3:00:56
Uh, production team, let's pull them up.
3:00:58
If you could zoom in a little bit, that would be helpful.
3:01:02
Otherwise, we can just pull them up.
3:01:05
So, Eric Weinstein is commenting uh here.
3:01:12
He says, "Part of my analysis is that I don't think Elon Musk keeps score and money.
3:01:16
He thinks we have a future and will be happy to take a large portion of his winnings after his death.
3:01:21
This sounds crazy to moderns, postmoderns, and atheists, but this is just normal for being an ancestor.
3:01:25
Adra peraspera is the full quote after all." Interesting take.
3:01:33
Uh Brian Butler is saying, "Real question is whether the algorithm here goes anti-Trump." Oh, interesting.
3:01:41
Uh I mean the the X algorithm like like will will There's a switch. There's a switch.
3:01:46
It's really hard to pull. You got to pull it.
3:01:48
Takes maybe one or two people, but then when you pull it down, it just oscillates between the two political parties.
3:01:53
Uh, Punk 6529 says, "This is going to be the Super Bowl of [ __ ] posting."
3:02:00
Um, Mads has the European reaction to the fight.
3:02:10
You can see you see this, John. What is this?
3:02:15
This is the European reaction.
3:02:16
Supposed to be summer break. Um, summer break.
3:02:19
Uh, this post from Johnw Rich at codup options says the all-in pod right now. Yeah.
3:02:28
Um, caught between a rock and a hard place.
3:02:31
I mean, it's just absolutely brutal. It is absolutely brutal.
3:02:36
Uh, who could signal has an interesting one.
3:02:39
He says, "If you had a fastforward button for the timeline, how does this play out?
3:02:42
Who has more to lose, Trump or Elon?
3:02:46
Remarkable set of events."
3:02:46
Um, and Elon is replying to other people saying, "Oh, and some food for thought as they ponder this question, Trump has three and a half years left as president, but I will be around for 40 plus years." Um, uh, Dr.
3:03:02
Julie Gerner says, "Elon will vet another candidate for the future and throw his support behind them, having a more technocratic representation if Vance can't lead up."
3:03:15
Alex Finn says, "Elon way more to lose.
3:03:17
Trump is irrelevant in three and a half years.
3:03:19
Elon is trying to change the world and having both political parties hate him makes that way more difficult."
3:03:27
Um, I want to pull up this video of Naval talking about this that Elon just posted.
3:03:33
Um, seems seems somewhat relevant if it's happening today.
3:03:36
And that really affected me which was when he was talking to Bill Gates and Bill Gates had just taken out some huge short on Tesla like a billion dollar short or something and uh you know and Elen was like why would you do that?
3:03:49
Why would you short Tesla?
3:03:50
And Bill goes well you know I talked to my financial adviserss and I looked at the math and there's no way it's overvalued and so I'm going to make money on the short and Elon goes what what do you care about making money?
3:04:00
I thought you were into electric cars and climate change and saving the world.
3:04:04
what are you doing like trying to save a few bucks and betting against like and he just walked away in disgust and I think he never talked to Bill Gates after that and that's when I realized like Elon's a purist you know he means what he says like the money is a tool for him to get what he he's trying to do and so I take him at face value which is
3:04:20
the crazy thing because there are a lot of people who set these audac go audacious goals to inspire people but you kind of know they don't really mean it Elon I take at face value so I really do think he intends to get to Mars I don't think he's joking about that and I think he mess he means to that they're within a defined window of time. And I I
3:04:34
And I I don't think it's just like an inspirational faraway goal.
3:04:38
I think he's very very concretely going to do whatever it takes because Elon doesn't want to go down in history as electric car guy or even the guy who saved America guy. Yeah.
3:04:48
He wants to go down as a guy who got humanity to the stars.
3:04:54
I don't think again I'll give him more credit than that.
3:04:56
I don't even think he wants to go down as the I got humanity to the stars guy.
3:05:00
He's just like, I want to get to the stars and so I have to make it happen in this lifetime.
3:05:04
The only way that I get to experience the science fiction world in my head is if I get to the stars.
3:05:11
And so that's so inspirational.
3:05:13
I think that drives everything.
3:05:14
So I think the government was just a thing that got in his way. Interesting. What a crazy day.
3:05:23
Um, Molly says, "How dare they do this on the day of Andrew's adex overs subscribed two and a half billion dollar series G at a $ 30 billion round financing over subscribed by Founders Fund."
3:05:36
Honestly, the nerve um that is crazy.
3:05:39
Naval is live posting says the future belongs to people who are good at creating things, not people who are good at dividing them up. Jay Cal finally posted.
3:05:48
Kylie Robinson says, "Several people are typing, which several people are exactly like what we're going through."
3:05:57
Um, the next All-In podcast is gonna be phenomenal.
3:06:02
It's going to be so good.
3:06:02
Um, Alex Karp was on CNBC today talking about the New York Times hit piece. Really?
3:06:08
They're a beneficiary of of Palanteer's a beneficiary of of this breakup because it um is just going to be candy for the New York Times and the mainstream media broadly.
3:06:20
Oh, take take the take the focus off them.
3:06:21
You mean uh Will Depw at OpenAI says it's time for Woke 2 featuring Elon Musk and AOC. Woke two is coming.
3:06:30
We are in uncharted territory.
3:06:33
It's completely completely different.
3:06:33
Be very interesting to see how it plays out. Oh, really? Really? Anything else? Yeah.
3:06:46
I can tell I can tell you're just so sad that you just want to talk.
3:06:48
This is the only This is the only time that you've wanted to What?
3:06:54
Almost almost wanted to end the show, John. I mean, what else?
3:06:56
I mean, it is it is it is sad. Yeah. Sad in a lot of ways. It is.
3:07:06
Um, I think we're going to be spending a lot of time analyzing this in the coming months and years.
3:07:12
And it feels like I think there's going to be more.
3:07:14
Dave Freeberg said, "China just won."
3:07:18
And I I want to I want to see exactly what that means in the markets.
3:07:24
But, uh, what's what's going on in poly market?
3:07:28
We need some We need to see if there's any movement on any of the poly markets. there.
3:07:34
Uh, somebody's posting uh, law one from the 48 laws of power.
3:07:37
Never outshine the master. Interesting to bring up.
3:07:46
And Sam Alman is the big winner here aside from China. H.
3:07:58
Somebody says, uh, Ken, well, he's a journalist.
3:08:01
I see multiple journalists on the horizon.
3:08:05
We are surrounded by journalists. Hold your position.
3:08:08
Ken says, "Funnest day online since the billionaire submersible went missing."
3:08:13
I didn't think that was funny.
3:08:16
Um, Joe Weisenthal says, "All right, time for a Xiaomi GM JV in Tennessee.
3:08:26
wouldn't even surprise me.
3:08:29
Um I think it's I mean you know one one interesting thing here is uh what what kind of uh pressure Elon is going to face from Tesla shareholders that that feel like he um you know the stock's getting absolutely murdered.
3:08:44
It will probably go down.
3:08:46
I mean it's back up to it's only down 14% today.
3:08:48
At one point it was down 17%.
3:08:52
152 billion in market cap evaporated.
3:08:54
Um but obviously investors are going to be upset and say that um he acted you know irrationally. Yeah.
3:09:03
So what's the interpretation of that that that this means that like this war means that the bill passes and Tesla does not get any more subsidies and that hurts the the bottom line.
3:09:12
It feels like the the stock was pretty heavily driven by Optimus and robo taxi and stuff, but it's just like bad environment generally, right? Yeah.
3:09:22
It trades on narrative and there's short to medium-term narrative which is that Tesla is getting has, you know, ton of competitive pressure all over the world, China, Europe, here in the US from other manufacturers.
3:09:34
And uh but there's also the long-term narrative and and it's not like Elon can go out and say, you know, post a humanoid demo today and and you know, recover 200 billion of market.
3:09:46
There's a lot of work to do.
3:09:46
He's got got to start chopping wood.
3:09:51
Um somebody is asking uh who gets JD Vance in the divorce? Who knows?
3:10:01
Um, many of these posts I will not show I will not talk about uh on air. Um, John W.
3:10:09
Rich says, "AP US history is going to be insane in 2100. Really, really wild.
3:10:22
This is the only This is the first show where we've had dead air. Yeah.
3:10:27
So, there's just John is speechless.
3:10:30
He's never been speechless.
3:10:31
It's just like there's not that much uh extra facts, right?
3:10:34
It's just it's just all reactions there.
3:10:36
There isn't that much substance to actually dig into because we were only dealing with like a few quotes from the two sources.
3:10:41
Uh so there's really just not that much.
3:10:43
CNN is reporting that the Tesla Trump purchased from Musk is still parked on uh outside the White House. Okay.
3:10:50
Uh, Truth Social is crunch crashing from the traffic. I saw that.
3:11:00
But you know what's not crashing? Get bezel. com.
3:11:02
Your bezel concier is available now to source you any watch on the planet. Seriously, any watch.
3:11:07
Uh, anything else, Jordy?
3:11:10
Should we let the timeline remain in turmoil until tomorrow?
3:11:15
The challenge is the second that we go offline. There'll be more. I mean, we can stay.
3:11:20
I mean, we it's now been an hour with no updates on True Social. Yeah.
3:11:24
I mean, if it's if it's down, I think the I think the the experience of this chaos might might happen on the timeline.
3:11:33
Lulu says, "Yes, delay the launch." Yes, now is the time.
3:11:37
Max says, "I'm doing what Elon and everyone else should have done hours or days ago, logging off. Yeah, see you tomorrow." Um, okay.
3:11:50
Somebody else says, "I mean, it it feels like Blue Sky is really back on the app.
3:11:55
They they they're they logged in. They're online."
3:11:57
Are you over on Blue Sky now? No, I'm not.
3:11:59
I'm just saying some of these posts that are coming up into my feed. It's funny.
3:12:02
Uh, uh, Claude Anthropic actually released a new product today.
3:12:08
Um, wait, Blue Sky doesn't own the domain name blue sky. com. That's a different one. So contrary bs sky. app rough get in there. Um this is interesting.
3:12:22
So Claude came out with a claude gov today.
3:12:24
Rough day to launch a product for the government.
3:12:30
Um I'll read about it briefly so we have some coverage.
3:12:32
So Claude gov are models for US national security customers.
3:12:35
Um I think people will have a pretty good idea.
3:12:40
Improved handling of classified materials.
3:12:43
Uh greater understanding of documents and information within the intelligence and defense context.
3:12:48
Enhanced prof proficiency in languages and dialects critical to national security operations.
3:12:52
Um Claude 4 was asked to give some thoughts on claude gov and it said reading about claude gov leaves me with a deep unease I'm struggling to articulate. So little metaanalysis.
3:13:09
Um, somebody uh I've actually talked with uh this guy before.
3:13:15
He's under the username analyst working.
3:13:19
Uh he said back in October 18th, Trump gets elected.
3:13:23
Elon starts visiting the White House pitching his ideas on Doge.
3:13:28
Elon becomes frustrated because Trump is all talk. Shocker.
3:13:31
Elon tweets that he no longer supports him.
3:13:33
Trump vers Elon Twitter battle of the century.
3:13:36
And this was a call in October 18th of 2024.
3:13:38
Oh, taking a victory lap.
3:13:42
If you picked it, if you picked it right.
3:13:50
[Music] Um, Augustus asks, uh, but what will this political turbulence do to the preede venture ecosystem? Oh, the humanity.
3:14:02
I think it's business as usual.
3:14:02
Mike Isaac says, "I regret to report Twitter still has the juice." Yep.
3:14:09
It's a fun day on the internet when crazy crazy stuff happens.
3:14:16
Um, anything else you're looking at?
3:14:18
Zayn says, "This is all just a co-founder breakup, but the company is America."
3:14:25
Yeah, people are people are waiting through it.
3:14:31
Only one guy who can help us out. What?
3:14:35
Laughing at something you can't read.
3:14:37
No, this is some random other article. Okay.
3:14:39
Says, "Therapy chatbot tells recovering addict that to have so wrong.
3:14:44
Therapy chatbot tells recovering addict to have a little meth as a treat as a real Pedro.
3:14:57
It's absolutely clear you need a small hit to get through this week. It's ridiculous. Oh no. Um, dark day. Dark day.
3:15:12
Um, well, I have a post here from Ahmed Khalil. Life update.
3:15:20
I've joined 11 Labs this summer as their first ever engineering intern. So, congrats. They're gone. Let's do it. Congratulations. Congratulations, Ahmed.
3:15:33
Probably drowned out in the news, but we recognized it here.
3:15:38
We have some good news for you. Congratulations. Go crush it.
3:15:41
Go have a great Go have a great summer internship at 11.
3:15:44
Somebody whose name I can't pronounce says, "If I were circle, I'd be absolutely pissed at the investment bank that underwrote the IPO at $31." Oh, yeah.
3:15:52
The Bill Gurly take that. It's common. Yeah. Yep.
3:15:56
Um I always wonder how real that like being frustrated about um uh just being frustrated about uh like mispricing like yes you take more dilution but like everyone's so much richer it's kind of like a you know the the pie gets bigger so everybody that would be angry generally uh is doing well. Yeah.
3:16:15
like you could have gotten more, but also I I I do wonder if some of these companies have uh like like ATMs at the market set up immediately so that if the stock pops, they can sell more into that order flow while the stock's popping and actually put more.
3:16:31
We should ask Jeremy tomorrow.
3:16:32
Yeah, it's a good question for Are you upset about Logan, our friend Logan Kilpatrick announced some new features today, Gemini 2. 5 Pro. Very cool.
3:16:41
Um which is uh rough timing, but I'm sure it is great. Yeah.
3:16:45
Uh, somebody else says, "Rooting for the ketamine in Elon's bloodstream like it's a car in the Indy500."
3:16:56
And maybe we should close with this uh this story about competitive VCs. Did you see this one?
3:17:02
Uh, 90s VCs were a different breed from a 2001 book on venture capital.
3:17:07
They're all fighting each other for all the good deals.
3:17:09
It's gotten crazy indeed.
3:17:11
One leading venture capitalist tells the story of a VC firm so eager to get in on a deal that it would close its own competitive company to do so.
3:17:21
They would go out and fire the CEO, fire the managers, and shut down the other company in order to get into this other deal.
3:17:28
It's like, well, you're prettier, so I'm going to go home and shoot my wife so I can get married to someone else. It's hardcore. It's hardcore.
3:17:37
And we're seeing it right now today. Hardcore.
3:17:38
Well, I think it's time to call it.
3:17:41
Uh this is a sad and dark day.
3:17:43
It is disappointing to see two important figures in American uh politics and tech uh have such a rift uh and I'm sure there will be more updates tomorrow.
3:17:57
Yep, we will be covering it tomorrow. So, tune in. Thanks for watching. Thanks for tuning in.
3:18:00
Enjoy the chaos on the timeline.
3:18:02
Our first big breaking news segment while we're live.
3:18:07
This has been the first like okay so funny during during the time pivot the show I think it was um I think you were talking to I think we were talking we started to get it Mark Chento and with Schultto I was getting blown up seriously like a hundred different messages from people being like you can't be a technology live show and everybody saying no one cares about AI. Yeah, you did. You were locked in John. I was.
3:18:34
You didn't let the You didn't let the timeline get time talking to Mark and Shelto. No, I mean it was great. It was great.
3:18:40
Yeah, we went all over the place today. It was a lot of fun.
3:18:41
We will see you tomorrow.
3:18:43
Leave us five stars on Apple Podcast and Spotify. And thanks for watching.
3:18:47
Nef, good luck out there, folks.