0:10
[Music] Hey, [Music] hey, hey. [Music] Come on. Come on. [Music] Hey, hey, hey. [Music] Hey, hey, hey.
Search this video
1:05
[Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] [Music] We came to this world. to reach the stars.
3:26
We came to this world to shape our future.
3:37
We came to this world [Music] to reach the stars to shape our future.
3:50
[Music] We came to feel the new music.
4:16
[Music] Feeling the music. Let me [Music] feel.
4:53
[Music] You're watching TBPN.
5:03
>> Today is Tuesday, July 29th, 2025.
5:03
We are live from the TBPN Ultra Dome, the temple of technology, the fortress of finance, the capital day.
5:13
>> Uh we created a list of top AI researchers.
5:16
It's called the Metis List and it's available.
5:18
It's live now at metislist. com.
5:23
>> Uh these are the world's top AI researchers.
5:26
We're of course experts in AI and AI researchers.
5:27
So we are fully qualified to curate this list.
5:30
But we did have some help from some friends and we pulled some data.
5:34
We looked at >> it was interesting.
5:35
Many of the people that we asked to do kind of like an ELO style ranking did not want to be named because uh nobody wants to be picking favorites >> on the inside.
5:46
But uh I think we put together a great list.
5:50
>> Uh it it's honestly some of the most impressive.
5:52
There's people in the top 10 with no Darc uh appearances.
5:57
>> I don't know how that happens. It's crazy. How how did they do it?
6:00
>> You got to get on door cash.
6:00
Yeah, we we we tried to have some fun with this.
6:02
We pulled some Google citations, some Google research citations, looked at some of their previous companies.
6:06
This was of course based on the list, which apparently has been going around Silicon Valley.
6:13
Uh Mark Zuckerberg has been, you know, using this to recruit top AI researchers.
6:17
We've been following the the trade war, as Jordy put it earlier.
6:22
We should have put a we should have put a a buy it now price on here for for Zuck so he could just add to cart. >> Add to cart. Yeah.
6:28
Just just create the super intelligence team directly on metalist. com.
6:32
Maybe that's how monetizing.
6:34
>> Yeah, it was good to see Alan Turing make the list. >> Fantastic. Yeah. Underrated.
6:37
George Bool not doing so well.
6:40
Inventor of Boolean Logic, but he did make the list but I think he's sitting in the 80s not doing too well.
6:46
Tyler, tell us more about the list.
6:48
Uh what do you think stands out?
6:50
What did you learn from the process?
6:52
What was your methodology?
6:52
How'd you build this thing? >> Um, yes.
6:54
So, this is a lot of fun.
6:54
I uh I scraped most of Google Scholar, like the top 5,000 people maybe.
7:00
And then, uh, with some friends from from the big labs, I I kind of narrowed it down.
7:05
>> Um, you can see like a lot of the papers they've written, interest, stuff like that.
7:10
>> Um, I think another gem in there is Alan Turing.
7:12
He's doing pretty well right now.
7:13
He's, let's see, he's ranked number six. >> Six.
7:16
Yeah, people love touring.
7:16
I mean, we passed his test.
7:18
And how long did his test stand? What? 60 80 years. Not bad.
7:23
>> Yeah, something like that.
7:24
>> That's That's longer than most benchmarks in Evals.
7:26
>> Long Allen Turing versus Goldpost. Who wins? >> Yeah. Yeah.
7:30
Wait, what were you about to say? >> Yeah.
7:31
I mean, it's the longest enduring benchmark, I think, ever. >> For sure. 80 years. Yeah. But who knows?
7:36
RKGIV3 might be around for 80 years. We might be here.
7:39
I beat it in 20 in 2034 or something. Wait, no.
7:45
2,120 I guess would be like the the the Allen touring test benchmark. Um we got John Schulman. Who else?
7:52
Jurgen Schmid Hubers is on here.
7:55
Jeff Dean, one of the greatest to ever do it over at Google.
8:00
He's been at Google for a long time. Still no.
8:01
Yes, a Dark Cash appearance. Let's go. Let's go. >> It's good.
8:05
Built map produce, big table, tensorflow. Yan lacun's in here.
8:07
You got all the different countries.
8:10
Really wide representation.
8:12
Canada, UK, Australia, Russia, China.
8:15
There's all sorts of people from all over coming together.
8:17
A couple people without photos have been able to stay anonymous, but most people we got their photo.
8:24
Shalto Douglas, good to see him on here. So, this should be fun.
8:28
Um, >> well, I'm sure we'll get a lot of feedback.
8:30
Oh, Dwar, uh, Shalto ranked 56, but three Dark Cesh appearances. >> There we go.
8:36
How are you, Were you waiting?
8:37
Were you waiting podcast appearances properly, Tyler?
8:39
Yeah, he should be at the top, but he wants to get >> putting up numbers like that for sure. >> Yeah, >> for sure.
8:46
>> Yeah, he's going to be coming for IA's spot. >> Yeah.
8:48
So, this is a lot of fun.
8:48
There's a uh there's an email sign up.
8:50
You can you can drop your email at the top to uh learn about the next drop, whatever we send out.
8:59
Um whatever we whenever we go live with stuff, we will let you know first.
9:04
>> And we're getting big into email soon, by the way, so stay tuned for that.
9:07
>> Very excited about that.
9:07
And uh yeah, hopefully this speeds up our trading card generation because you'll just be able to screenshot this and say this person just got traded.
9:15
Um because uh it's moving fast and furious and we can't spend that much time handcrafting each of those trade deal trading card images.
9:25
We got to be able >> Indeed. Indeed.
9:27
There's there's a lot of stuff going on.
9:29
Um well uh interesting related to Zach.
9:33
Did you know that one of the kids at the IMO that was competing against OpenAI and DeepMind was named Alexander Wang? >> Wow.
9:44
>> Great nominative determinism. Yeah. Future Alex. >> Future.
9:47
>> Uh and speaking of >> people are calling him the the >> the Alexander Wang. >> The next Alex Wang. >> The next in line. >> It does have an e.
9:54
It's not a it's not a lx e and d e.
9:57
It's d e r instead of dr.
10:01
Um, so slightly different, but still Alex Wang nonetheless.
10:03
Um, well, let me tell you about ramp. com.
10:06
Time is money save both easy to use corporate cards, bill payments, accounting, and a whole lot more all in one place.
10:10
And we just closed an account for ramp.
10:12
Uh, do you want to break down this?
10:15
So, Cody over at Jones Road Beauty, uh, the CEO over there, which is a massive uh, massive brand, hundreds of millions of revenue, fantastic business, >> and they're absolute dogs over there. Yep.
10:29
Uh he posted, "My team has an inerson demo with Ramp tomorrow.
10:31
If they bring us a TBPN hat, I've instructed them to sign on the spot. Let's go."
10:37
>> And uh Ramp posted a picture earlier today with the contract. >> They closed the deal. >> Love to see it. Cody, enjoy your hat.
10:45
We're glad that you're going to be wearing it.
10:46
And welcome to uh welcome to Team Ramp. >> Enjoy travel. Enjoy bill payments. Enjoy accounting.
10:52
>> Enjoy easy to use corporate cards.
10:53
>> Enjoy a whole lot more. Have fun with it.
10:55
So yes, uh the Wall Street Journal has a >> uh has a deep dive on the IMO and what happened between Google Deep Mind and OpenAI.
11:04
Of course, >> inverse machine, >> of course, Deep Mind and OpenAI won gold medals at the math Olympics, but these American teenagers still got higher scores, absolutely mogged by a child.
11:12
Uh will this be the last time humans outperform AI?
11:19
And I think the answer is clearly yes.
11:21
But it's still hilarious because, you know, there's so much attention on AI and it's so impressive and it's so great, but >> still got it.
11:29
But I mean, we'll go into it, but basically all the high schoolers like, "Yeah, they're going to get us next year, but like we still got them this year."
11:35
So anyway, from the >> This was the year.
11:37
This was the This will be one of the most meaningful wins, right?
11:41
Like the passing of the torch, >> maybe.
11:43
I mean, I think that what goes into a meaningful win is more of like the drama and storyline behind it.
11:49
just like, you know, uh what's the reference?
11:55
Computers beat humans at chess in 1997.
11:58
Computers beat Go, humans at Go in 2016.
12:02
Like there have been dramatic chess matches since 1997. >> Yeah.
12:07
>> Um it's just a uh like you just assume that computers are the best at it, but you move on and then you go back to like caring about what humans >> the players players and the personalities.
12:16
It is cool to think about um what what the energy requirements would have been for the high schoolers to win versus uh open AAI and Google.
12:24
You have to imagine >> the high schoolers like I had peanut butter and jelly sandwich and I won gold.
12:30
Google's like we we used the energy of Houston, Texas, the energy that Houston, Texas uses in a year in an hour.
12:42
>> I don't think it's quite that dramatic.
12:44
>> Not not that traumatic.
12:44
Of course, but it is probably more than a PB&J.
12:47
>> So, the Wall Street Journal says, "The smartest AI models have ever made just went to the most prestigious competition for young mathematicians and managed to achieve the kind of breakthrough that once seemed miraculous.
12:58
They still got beat by the world's brightest teenagers. Let's go. Let's go."
13:05
Uh, every year a few hundred elite high school students from all over the planet gather at the International Mathematical Olympiad. That's the IMO.
13:16
This year those brilliant minds were joined by Google DeepMind and other companies in the business of artificial intelligence.
13:21
They all had come for one of the ultimate tests of reasoning, logic, and creativity.
13:24
The famously grueling IMO exam is held over two days and gives students three increasingly difficult problems a day and more than four hours to solve them.
13:33
The questions can span algebra, geometry, number theory, combinotaurics, and you can forget about answering them if you're not a math wiz, Jordy.
13:42
You give your brain a workout just trying to understand them.
13:46
Because those problems are both complex and unconventional.
13:48
The annual math test has become a useful benchmark for measuring AI progress from one year to the next.
13:54
In this age of rapid development, the leading research labs dreamed of a day, their systems would be powerful enough to meet the standard for an IMO gold medal, which became the AI equivalent of a 4-minute mile.
14:09
But nobody knew when they would reach that milestone or if they ever would until now. Scott Woo knew.
14:13
He called it out on our show in April.
14:16
Um, also let me tell you about reream.
14:18
How we're streaming right now. We're using reream.
14:21
one live stream, 30 destinations, multiream, and reach your audience wherever they are.
14:25
If you're trying to stream, get on reream.
14:26
Uh, the unthinkable occurred earlier this month when an AI model from Google Deep Mind earned a gold medal score at IMO by perfectly solving five out of the six problems.
14:36
And we're going to go into what happened with the sixth.
14:37
In another dramatic twist, OpenAI also claimed gold despite not participating in the official event.
14:45
This is the they were running the marathon in the parking lot strategy. Absolute dogs.
14:48
get in the press without needing to deal with all the red tape inside the stadium or buy a ticket. >> Yeah. >> Yeah. >> Winners win. >> Yeah. I'm in the Olympics.
14:59
I'm I'm as f I'm Usain Bolt.
15:01
Usain Bolts in the parking lot. Don't worry about me. Don't ask questions. >> Yeah.
15:07
It really is uh the the way that they were able to get >> Yeah.
15:13
>> credit for basically on the same level of Google.
15:15
Uh >> arguably more >> on X.
15:18
>> on X. They went super viral at midnight on Friday like the moment the exam closed like they got a lot of the attention and it's clear that they actually did the job and they did it was just like >> you know certain you know mechanics of
15:34
like how they participated and like does that really matter or does the quality of the model matter really just the quality of the model so I think that uh it makes sense to give them just as much credit here or 99% as much credit. Uh
15:42
Uh the companies described their feats as giant leaps toward the future, even if they're not quite there yet.
15:48
In fact, the most remarkable part of this memorable event is that 26 students got higher scores on the IMO than the AI systems. 26. It's a lot.
15:57
Among them were four stars of the US team, uh including a man named uh Tiger Jong, a two-time gold medalist from California, and Alexander Wang. >> Tiger. >> Tiger. >> He's a Tiger Wang.
16:12
>> Get this man a term sheet right now.
16:16
future enterprise SAS builder. Yeah, >> for sure.
16:19
Founder >> easily deserving of 10 on 100 preede out the gates. Half in secondary. >> Yep. Secondary.
16:28
>> Get him a term sheet immediately.
16:28
Get him on Stripe Atlas immediately.
16:30
Get him a Figma installation so he can design his software.
16:34
He can think bigger, build faster.
16:36
Figma helps design and development teams build great products together. That's right.
16:39
Get started for free at figma. com.
16:40
Also, >> we built the metal list.
16:42
We created it with Figma make. Yep.
16:45
>> So you can go check it out.
16:45
You can see our work in pro or Tyler's work.
16:49
>> Uh you know in action our work.
16:53
>> Um we of course if there is any backlash we will be distancing ourselves from Tyler's work >> and Tyler will be quickly terminated.
17:00
>> We will not >> I will I will fall on the sword.
17:03
>> We will we will ride we we ride for Tyler. >> We do. We do >> until the end.
17:08
So, uh, Alexander Wang had a third straight back-to back gold medal. He's from New Jersey.
17:13
That makes him one of the most decorated young mathematicians of all time.
17:17
He's just got three IMO gold medals hanging around his neck looking good.
17:21
>> Do that picture with like biting.
17:22
>> Yeah, but he's at the top. He's at the top.
17:24
>> Yeah, >> we're we're at the bottom spraying the champagne being like, "Yeah, IMO >> America." >> Yeah, America's back.
17:30
Uh, he's a high school senior who can go for another imo gold next year.
17:34
So he could wind up being a four-time gold medalist. That would be crazy.
17:40
>> Um maybe you got to up that valuation at that point.
17:42
Uh but in a year he might be dealing with a different equation altogether.
17:45
And he actually gave an interview to the Wall Street Journal.
17:49
Alexander Wang says, not the scale AI founder, not the head of Meta Super Intelligence, but the IMO gold medalist says, "I think it's really likely that AI is going to be able to get a perfect score next year.
18:01
That would be insane progress. I'm 5050 on it.
18:03
So given those odds, will this be remembered as the last IMO when humans outperformed AI?
18:09
It might well be, says the leader of Google Deep Minds team.
18:14
So until very recently, what happened in Australia?
18:18
>> This really is is insane.
18:18
So So Tiger has three gold medals. >> Tiger has two.
18:22
>> He's going >> and Alex Wang has three. >> Alex Wang has three. >> Yes.
18:25
>> And going into But but is it Wang that can get a fourth?
18:29
>> Yes, he can get a fourth.
18:30
>> So he's going he's going into his fourth. >> Yep.
18:33
>> Yep. going for his fourth >> he's got a freshman year >> against and he's competing against the brightest AI researchers at Google and open AI and he's potentially the last man standing between you know total machine domination and uh >> this I mean I I can't wait for next year
18:51
we might have to go in person >> I think we have to live stream it for sure >> you're hackling the the deep mind >> full body paint for sure >> full body paint for Alex Wang >> yeah uh until very recently what happened in Australia would have sounded about as likely as Koala is doing calculus. But the in but the
19:06
But the in but the inconceivable began to feel almost inevitable last year when DeepMind's models built for math solved four problems and racked up 28 points for a silver medal just one point short of gold.
19:18
This year the IMO officially invited a select group of tech companies to their own competition, giving them the same problems as the students and having coordinators grade their solutions with the same rubric.
19:27
They were eager for the challenge.
19:28
AI models are trained on unfathomable amounts of information.
19:32
So if anything has been done before, the chances are they can figure out how to do it again.
19:37
But they struggle with problems they've never seen before.
19:39
As it happens, the IMO process is specifically designed to come up with those original and unconventional problems.
19:45
In addition to being novel, the problems also have to be interesting and beautiful, said IMO president Gregor Dolanar.
19:55
>> Let's hear for beautiful math. We love it.
19:57
the beautiful mind at work at this IMO uh championship.
20:00
If a problem under consideration is similar to quote any other problem published anywhere in the world, he said it gets tossed by the time students take the exam.
20:11
The list of a few hundred suggested problems has been whittleled down to just six.
20:18
Meanwhile, the Deep Mind team kept improving the AI system.
20:19
It would bring to IMO an unreleased version of Google's advanced reasoning model called Gemini Deepth Think and it was still making tweaks in the days leading up to the competition.
20:30
The effort was led by Thong Luong, a senior staff research scientist who narrowly missed getting into getting to the IMO in high school with Vietnam's team.
20:40
He finally made it to the IMO last year with Google. >> He never gave up.
20:43
>> Before he returned this year, Deep Mind executives asked about the possibility of gold.
20:48
He told them to expect bronze or silver again.
20:51
He overd his own expectations. >> He overd delivered. We love to see. >> He did.
20:56
Just like Vant overdelivers.
20:58
Automate compliance, manage risk, improve trust continuously.
20:59
Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation.
21:05
Whether you're pursuing your first framework or managing a complex program.
21:07
So, uh, the head of Google DeepMind's IMO team, he adjusted his expectation when DeepMind's model nailed all three problems on the first day.
21:16
And so they get past day one, nailed, nailed, nailed. Three for three. He's feeling good. He's going into day two. He's optimistic.
21:23
He says the simplicity, elegance, and sheer readability of those solutions astonished mathematicians.
21:28
The next day, as soon as he he and his colleagues realized their AI creation had crushed two more proofs, they also realized that would be enough for gold.
21:37
They celebrated the monumental accomplishment by doing one thing the other medalist couldn't.
21:43
They opened a bottle of whiskey >> like they're under 21. They're 18. >> They're 18. >> It's so funny.
21:52
Whi- whiskey is such an odd choice.
21:54
>> It is a funny choice, too.
21:54
But they're like, >> but they're but they're they're thinking what would what would a normal person do?
22:00
What would a normal person do to celebrate? Not champagne.
22:06
Grab a bottle of whiskey.
22:07
>> Whiskey is a funny choice, but I guess they like whiskey at Google DeepMind.
22:09
Uh so to keep focus on the students, the companies at IMO agreed not to release their results later uh until later this month.
22:18
But as soon as the Olymp's closing ceremony ended, one company declared that its AI model had struck gold and it wasn't Deep Mind. It was open AI.
22:28
>> Wait, so I don't understand how Open AI was was not a part of it officially, but isn't there like a pretty complex like grading process and they invited companies to participate?
22:38
So, I don't Yeah, I don't know.
22:41
Maybe we'll get into this, but I think um the the IMO questions go up as soon as the as soon as the uh competition ends.
22:48
So, you can just hit them immediately with your model.
22:52
And you could probably have your own IMO like u like tutors or former coaches or former graders grade the results and you could kind of just evaluate yourself on them.
23:07
And I think also >> I just thought that they graded based on like the actual process to get to the result as well, but maybe that's >> in that case it becomes a little less clear.
23:17
>> I think you're right because there are six questions and they're not binary because you can the six questions add up to 42.
23:24
And so there's there were five high schoolers that got a 42. Perfect score. Nothing missed.
23:30
But there was one that got 40.
23:33
So minus two points more than five out of six correct.
23:37
So got five out of six correct and then on the sixth one got almost all of it.
23:42
So got some partial credit.
23:44
But I still think you you could grade that and figure out um you know how how to score it.
23:49
So anyway, let's go into OpenAI.
23:51
The company wasn't part of the IMO event, but OpenAI gave its latest experimental reasoning model all six problems and enlisted former former medalists to grade the proofs.
24:00
Like Deep Minds, OpenAI's system flawlessly solved five and scored 35 out of 42 points to meet the gold standard.
24:11
After the OpenAI victory lap on social media, the embargo was lifted and DeepMind told the world about its own triumph and that its performance was certified by the IMO.
24:18
not long ago, it was hard to imagine AI rivals dueling for glory like this.
24:24
It is hilarious to think about if you've scrolled back to 2016.
24:28
Uh you know, uh Demis is with Google Deep Mind battling it out in Lis It all the the the Alph Go match and then like Sam Alman and the Open AAI team are just like playing Go like in the parking lot being like we also are super human at Go like and and people are like but you're not actually in this particular competition.
24:49
and they're like, "But we're still really good at >> how can you're not even in the stadium."
24:55
>> It's very funny, but I mean, they still demonstrated like fantastic performance, you know?
24:59
So, you got to hand it to him.
25:01
Uh, in 2021, a PhD student named Alexander Wei was part of a study that asked him to predict the state of AI math by July 2025. That is right now.
25:11
When he looked at the other forecasts, he thought they were much too optimistic.
25:14
As it turned out, they weren't nearly optimistic enough.
25:15
Now, he's living proof of how wrong he was.
25:19
Weey is the research scientist who led the IMO project for OpenAI.
25:22
So five years ago, four years ago, they ask him, "When do you think we're going to beat the IMO?"
25:29
He's like, "Definitely not in four years."
25:31
And then he's the one that does it in four years.
25:32
So even even the insiders are are bad at predicting how the future's playing out. >> Yeah.
25:39
>> Um >> the only thing more impressive than what the AI systems did was how they did it.
25:44
Google called its results a major advance, though not because DeepMind won gold instead of silver.
25:49
Last year, the model needed the problems to be translated into a computer programming language for math proofs.
25:55
This year, it operated entirely in natural language without any human intervention.
26:00
Deep Mind also crushed the exam within the IMO time limit of 4 and a half hours after taking several days of computation just a year ago.
26:09
You might find all of this completely terrifying and think of AIs as competition.
26:13
The humans behind the models see them as complimentary.
26:17
This could be perhaps this could be ha perhaps be a new calculator, said the head of deep mind that powers the next generation of mathematicians.
26:24
And if you want to generate the next generation of your code review process, get on graphite. dev.
26:30
Code review for the age of AI.
26:32
Graphite helps teams on GitHub ship higher quality software faster.
26:36
>> The problem of problem six, they got five out of six.
26:38
Let's talk about problem six.
26:41
>> Speaking of that, next generation, the IMO gold medalists have already been overshadowed by AI, so let's put them back in the spotlight.
26:47
Team USA is looking there, looking great.
26:50
Uh, so a 17-year-old student in Los Angeles on his way to MIT to study math and computer science.
26:56
Uh, Jeang, as a young boy, his family moved to the US from China and his parents gave him a choice choice of two American names.
27:06
They said, "You can pick your two two names. Two names. Here are your options. John.
27:09
John, what are you picking? Tiger or >> elephant? >> Elephant. >> He chose tiger. >> He chose tiger.
27:15
His career in competitive math began in second grade when he entered a contest called the math kangaroo.
27:23
>> Is the journal messing with us a little bit? >> This is so funny.
27:27
>> This guy's had a very animal themed life. It >> is.
27:30
It ended this month at the math Olympics next to a hotel in Australia with actual kangaroos.
27:34
very zoo themed math math battle.
27:37
Um so when he sat down at his desk with a pen and lots of scratch paper, Jeang or tiger as he is known spent the longest amount of time during the exam on problem six.
27:48
It was a problem in the notoriously tricky field of combinatorics, a branch of mathematics that deals with counting, arranging, and combining discrete objects.
27:57
And it was easily the hardest on this year's test.
27:59
The solution required the ingenuity, creativity, and intuition that humans can muster, but machines cannot. At least not yet.
28:06
I would actually be a little bit scared if the AI models could do stuff on problem six, he said.
28:12
And so we can read from problem six and see if if anyone in the audience can get this. Consider a 2000.
28:21
Can you >> just uh just >> just solve this?
28:23
>> Solve this real quick.
28:23
Consider a 2025x 2025 grid of unit squares.
28:26
Matilda wishes to place on the grid some rectangular tiles, possibly of different sizes, such that each side of every tile lies on a grid line and and every unit square is covered by at least one tile.
28:42
Determine the minimum number of tiles Matilda needs to place so that each row and each column of the grid has exactly one unit square that is not covered by any tile.
28:54
>> So yeah, you can just do this in your head if you're, you know, Scott Woo. >> Yeah.
28:59
>> Problem six, stumped Deep Mind and OpenAI's models.
29:02
Very interesting that they both failed on the same question.
29:06
Probably tells you a little bit about how they're designing these models.
29:08
Uh, but it wasn't just problematic for AI.
29:11
Of the 630 student contestants, 569 students received zero points for this particular question.
29:19
Uh the answer, by the way, is 2,112. >> Yeah.
29:23
And I see >> few few people few people got it.
29:27
>> Clearly clearly getting close. >> Yeah. Yeah.
29:30
Um only six received full credit of seven points.
29:33
Um, Jeang was proud of his partial solution that earned four points, which was four more than almost everyone else.
29:39
At this year's IMO, 72 contestants went home with gold.
29:42
But for some, a medal wasn't their only prize.
29:47
Jong was among those who left with another keepsake.
29:49
>> And I just got to say, thank you to Jane Street and XTX Markets for sponsoring the >> challenge. That's right.
29:56
really supporting young mathematicians and you know potentially introducing them to uh >> you know the incredible work that Jane Street does to you know bring liquidity to markets all >> markets absolutely and thank you for linear for sponsoring us.
30:10
Linear is a purpose-built tool for planning and building products.
30:14
Meet the system for modern software development.
30:15
Streamline issues, projects, and product road.
30:18
>> IMO people only get a a piece of paper and >> Imagine if they had game over.
30:23
>> Four and a half hours.
30:23
They could do it in 45 minutes. >> That's right. >> No doubt. >> Yeah.
30:26
I mean, so so again, I I guess to summarize, uh the the kind of >> uh the human accomplishment is incredible.
30:33
incredible. the controversy I think and and frustration from different people around opening eyes's approach a sort of rushed announcement that didn't seem to be very sanctioned drama >> uh Terrence Tao at the IMO called it rude and inappropriate
30:49
>> so he was not super happy and then the >> uh self-grading so Google DeepMind uh actually had their results graded by IMO coordinators they participated into the program that IMO created >> for technology companies and opening eyes proofs are public so you can go see them. Uh but they don't have the
31:06
Uh but they don't have the official IMO validation.
31:08
They just used um uh people that were uh experts and and had competed previously, graded previously, but uh not officially sanctioned. >> Yep.
31:21
Um so 72 contestants went home with gold.
31:25
We we should we should figure out obscure Olympic uh Olympic uh you know activities sports that we could do and and kind of like run our own competition in the parking lot when when the Olympics come to LA and try to get it try to get an Olympic gold >> Dr. D >> for sure. >> You know Dr. is coming to LA. >> Yeah.
31:47
>> They're going to have it at Sanonita. >> We'll be there.
31:49
>> It's going to be the greatest crossover event of capital allocators and technology journalists in history. >> Absolutely.
31:54
I think pieces that will be written that >> it's gonna be fantastic.
31:59
Uh there was >> no but it' be really I mean I don't know it'd be very very diff there has to be >> we have to do Dr.
32:06
We have to be dancing on a horse outside of out outside of Sanonita racetrack and be like we got gold we danced on the horse.
32:17
>> We we put together a team of interns to people inside the stadium. >> For sure. For sure. something there.
32:24
>> Uh, there was once a time when such precocious math students grew up to become professors or presidents.
32:28
The recently elected president of Romania was a two-time IMO gold medalist with perfect scores. Let's hear for Romania. >> Let's go.
32:38
>> While many still choose academia, others get recruited by algorithmic trading firms and hedge funds where their quantitative brains have been have never been so highly valued.
32:46
This year, the US team was supported by Jane Street and XTX Markets sponsored the whole event.
32:52
After all, they will soon be competing with each other and with the richest tech companies for their intellectual talents.
32:56
By then, AI might be destroying these gold medalists are going to love Jane Street.
33:01
>> You're going to love it.
33:01
It's going to be fun figuring out little puzzles in the market, trying to squeeze out.
33:06
>> Today, you're doing mathematics.
33:08
Tomorrow, you might be overthrowing a small nation state. >> No, it's uh great.
33:12
You you got an IMO gold medal.
33:15
Your job is to come to Jane Street and buy ads on the Dark Cash podcast.
33:20
You're you're you have a bright future.
33:23
>> You're the head of podcast marketing now.
33:26
>> IMO winning podcast marketing.
33:28
>> I just assume everyone who works at Jane Street has >> at least one >> regardless of whether or not you're actually writing any code. >> Yeah.
33:34
>> Um >> a former IMO gold medalist himself, Jung, is now an associate professor at Brown University and visiting researcher at DeepMind who worked on its gold medal model.
33:43
He doesn't believe this was humanity's last stand, though.
33:45
He thinks problems like problem six will flumix AI for at least another decade.
33:49
That is a crazy crazy timeline.
33:53
I feel like no one thinks that we're a decade out from a perfect score. That's wild.
33:57
Um anyway, um he walked away from perhaps the most significant math contest in history feeling bullish on all kinds of intelligence.
34:07
There are things AI will do very well, he said.
34:09
There are still going to be things that humans can do better.
34:12
And I guess I believe that broadly.
34:14
Uh it's very interesting.
34:14
I think the biggest takeaway is that there are there are levels to this.
34:18
are there are levels to this. There are always levels and in this it's that there there is a level where you want you know the IMO math to be solved like chess was solved in when deep uh when uh was it uh deep blue beat Gary Kasparov
34:37
in 1995 and so chess was considered like you know better like computers were better at chess than humans but then for a while >> there was something called centaur chess where a human plus a computer was better than a computer and it was also better than a human. And so the human would sit
34:52
And so the human would sit there and look at the recommendations from the computer that was crunching all the numbers and then the human would do the the soft skills stuff like read the the opponent kind of see if there were blunders and kind of act as >> they're sweating.
35:06
>> Yeah, they're sweating. They're >> sweating. Yeah. Yeah.
35:08
>> Sweating through Exactly.
35:08
So like >> we got them >> figuring that out.
35:11
But now chess is completely solved to the point where uh they're adding a human to the mix means nothing.
35:16
Um but chess had like a 10-year 20-year run of like centaur chess being the dominant the the actual peak of performance.
35:23
Um then the Lisa doll alpha go uh match showed that deep learning models were not just better at go than humans but could actually come up with novel strategies.
35:41
And that's the famous move 37 moment where Alph Go placed a piece on the go board that was completely unexpected.
35:49
Lisa it all stands up, walks away.
35:49
I think he goes and smokes a cigarette outside and he's like stressing and he's like that's weird. That makes no sense.
35:57
Like is the computer >> like broken?
35:58
Like is it messing with me?
36:01
And it turns out that that was a new strategy and it wound up winning that game.
36:04
And so I think that there are levels in the sense that like right now we're at five out of six.
36:07
We don't just have to get the AI models to six out of six.
36:11
I think we need to get them to six out of six.
36:13
And then also like, oh wow, they that like the the reasoning that got them there was novel. Yeah.
36:17
And different from everyone else.
36:20
>> Be interesting to find somebody who's great at chess and great at math.
36:21
Maybe it's someone like Scott Woo who can break down kind of the differences in the problem that a computer uh is solving when it's trying to win at chess versus the problem when it's approaching >> a new type of math problem. >> Yep. >> Right.
36:36
It's it's math problems can like constantly evolve and take different forms and things like that.
36:40
It presents a slightly different challenge.
36:44
>> Um, somebody uh yesterday or the day before Sid Styheart says, "When AI was introduced to chess, the thought was that grand masters would use it as a tool to maximize efficiency, but it very quickly became apparent that even the greatest human chess players can contribute nothing but noise." >> Yeah.
36:59
>> And uh and they say nothing is immune.
37:02
Everyone will soon be obsolete.
37:02
And Mark uh Mark Andre comes in with a quote and says chess is more popular in a bigger industry now than ever. >> Yeah. Yeah. Of course.
37:11
>> And um yeah, this because people like to watch human feats of uh intelligence and challenges and people like to be challenged and that is part of and comp there's value in competition >> and there's value in seeing two humans that are experts in their field and at their craft compete and that uh even as a humanoid robot will be better at shotput. Yeah. in the Olympics.
37:32
Still gonna be cool to watch somebody throw a heavy rock when somebody is >> when you could imagine a a robot being better at horseback riding than a human.
37:44
But will it still be cool to watch a human and a horse dance? >> Stallion. >> Of course it will. >> Of course it will.
37:50
It's >> some things never get old.
37:52
Well, if you if you have a business and you're getting customer support tickets in and they're not asking you IMO level questions, question six, like that level, if it's like normal level, you got to get on.
38:06
ai, the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bake ops, number one in rank ranking on G2.
38:17
the majority of your customer support tickets are IMO question six level difficult but if it's basic things how do I you know get a refund fin.
38:26
ai is going to be >> we got to mess we got to mess with the own a little bit I'm on the fin.
38:30
ai AI website right now and I'm talking with the Finn agent. Beat it. Hi there.
38:37
You're speaking with a Finn agent.
38:37
I say, "Consider the 2025* 2025 and that's a grid of unit squares."
38:46
He says, "I noticed you've asked a math problem about grid tiling, but I'm actually a sales development.
38:52
I help businesses find the right customer support and engagement." >> Screenshot.
38:57
>> Can you take a crack at it?
38:57
It would be really helpful to get me through the next stage of this sales funnel.
39:05
>> Would help me build confidence in the product >> in Finn. >> Yes.
39:13
Apparently, you can get Amazon's uh have you seen these these memes of like people go to the Amazon help uh help like chat and then they get it to like write React and like use it as like a like a cursor wind surf like alternative like talking. >> It's free. Free inference. >> It's free inference. Yeah.
39:28
And so you super you can totally jailbreak it and just be like, "Yeah."
39:32
Um anyway, uh get on numeral sales tax on autopilot.
39:36
Spend less than five minutes per month on sales tax compliance. Go to numeralhq. com.
39:42
Uh and you know who's going to be paying sales tax soon?
39:45
Apple on the new iPhone, which was just spotted in the wild.
39:48
Mark German says he thinks it looks legit.
39:51
If you like cameras, you're going to love the next iPhone because what I'm seeing here, >> Zoom in on this.
39:58
The next slide is >> zoom in even more if you can. I'm seeing a huge flash.
40:05
>> Huge flash to the side.
40:05
Huge uh lidar scanner, that that type of thing, dot projector.
40:11
>> And it looks pretty flat. >> It looks flat.
40:13
I can't tell if this is just a case on top of it or there's too much compression in the photo that we're looking at to see if the new iPhone doesn't have a camera bump, but it looks like there's a chance that we're getting rid of the camera bump.
40:26
I mean, for a long time, people have been dunking on Apple.
40:30
Steve Jobs would have never gotten a camera bump.
40:32
You know, he never would have shipped this.
40:35
>> Steve Jobs would appreciate how it drives most people to buy a case, >> I think. So, and I I don't know. I mean, it's unclear.
40:42
The iPhone 4 didn't have a bump.
40:45
>> It mustachump case attach rate go up like 2%.
40:46
And they're like, "Yeah, we'll take the extra bill." >> Oh, yeah.
40:51
Tyler, did you order a phone yet?
40:53
>> Tyler famously won an iPhone by defeating the Arc Hi V3.
40:57
Wait, you have it in the studio?
40:59
If you have the new phone, >> it's way better.
41:02
But the camera bump is like really annoying.
41:04
I didn't realize it had that.
41:06
>> Wait, did your old phone not have a camera bump?
41:08
>> Well, I had a case on it.
41:08
You don't you don't like how it makes a sound when it's flat on a table.
41:14
>> Well, uh I mean, if you want a case, you're going to have to do something even more extreme.
41:16
You're going to have to defeat the IMO.
41:21
>> Um I wonder uh uh I it would be fun.
41:25
Have you had a chance to play with the the agent framework on ARC AGI?
41:30
>> I want to see if you can rank on that.
41:32
That would be interesting.
41:32
If you can do that, if you can if you can solve it using AI, then you can get then you can get win one case one case with the phone. >> Let's go. >> Um, so yeah.
41:42
Uh, apparently there's like some harness that's already kind of pre-built and you can kind of just fine-tune and and you know swap things in and out and uh have have like a little bit more of a flexible approach.
41:54
I saw someone posting about they they beat the first level just kind it seemed like a kind of a brute force solution.
42:01
Um, but I think it'd be interesting to see like how far can you get in the games before it just becomes really really difficult to to to play through.
42:09
But um see see what models are out there and see what see what you can what uh what the experience is like building for it.
42:16
Um anyway, this uh this iPhone camera bump is interesting.
42:19
So um I I don't know if Steve Jobs would have shipped this.
42:25
There was always this weird trade-off because people say just make it flat, but there like the laws of physics do apply to cameras.
42:30
Like you need a certain focal length to get a certain quality of picture out of them.
42:37
And so Apple's made this tradeoff between the bump gives you better photos, which people really like.
42:42
You could go flatter, but then you get worse photos.
42:45
You could make the phone thicker to match the depth of the camera, but then it's big and chunky and heavy.
42:50
And people say that they want more battery life, but the real revealed preference is that like one day of battery life is fine.
43:00
People are fine being in the habit of charging their phones every day.
43:01
And we kind of saw this the same thing with like the the watches.
43:05
Like some people are like, "Oh, you you like if it would actually be more inconvenient to be like, yeah, I have to charge it every five days."
43:12
It's like, "Okay, so I charge it on Monday and then Friday and then Wednesday and then the next Saturday instead of just being in a routine and just doing every single day."
43:21
So, um, >> yeah, there's also like how fast does something charge, right?
43:24
Like if if something can get if you can add 20% battery life in >> they charge really fast, >> 30 minutes.
43:33
>> And there's chargers like everywhere.
43:34
Like you get in any car, there's a charger.
43:35
You any office, there's a charger.
43:37
Everything's so standardized on lightning or USBC now that like you can grab a charger anywhere.
43:42
So, and then there's also like those add-on battery packs.
43:45
If you're really crazy about that, you can modify it. So, I don't know.
43:47
I think we need a poly market on whether or not the um the next iPhone has a camera bump.
43:54
That would be interesting to track. I want to know.
43:56
I I feel like I don't know this picture.
44:03
>> I feel like this picture is a case.
44:03
I feel like it will have a camera bump. What do you think?
44:08
>> I'm going I'm going camera bump staying.
44:10
Staying >> bump bump stays.
44:12
>> Yeah, it makes sense that they would put a case on. It's like those cars.
44:13
I keep seeing the new BMWs driving up and down PCH and they're wrapped in that like confusing camouflage to to so that you can't really see the curvature if you take a picture and you can't really see the full silhouette of the car.
44:28
>> Yeah, >> it's kind of the same thing.
44:28
They should do that for the phone.
44:30
I mean, they kind of did with this case probably.
44:32
>> Yeah, >> but I don't know.
44:32
They moved the flashlight over.
44:34
>> Well, one thing that we do know is that the company is going to be introducing a foldable iPhone at the end of next year.
44:41
It will be entering a product category that's already 7 years old, pioneered and dominated by its biggest hardware rival, Samsung.
44:47
And at this time, Apple won't be debuting a radic debuting a radically new interface or transformative hardware.
44:52
Instead, the device will offer similar design as Samsung's Galaxy Z Fold line and use many of the same core components, including foldable OLED screens sourced from Samsung Display. >> Yep. >> Oh, interesting.
45:05
So, it will be a Samsung screen.
45:08
Yeah, basically Samsung's done seven years of R&D on this and the supply chain's really robust.
45:13
Do you remember the first folding phones, >> the first folding phones, not the Motorola Razer V3, but like the first foldable screen phones like they wouldn't fold flat.
45:23
So like the hinge it would like when you close the phone >> and one I forget what it was where the the keyboard would flick out. That's the sidekick.
45:31
That's not that was an awesome era of consumer.
45:35
>> But when they folded, they wouldn't fold flat like this.
45:37
they would fold and be like this basically and so like you would be able to see through the hinge because like it wouldn't fold all the way closed.
45:44
Now they're at a point where they do fold flat and the hinges are extremely reliable so that you can use it for a year and if you get a little sand in there it's IP76, you know, you know, waterproof, all of that.
45:55
They've they've figured out all the industrial design and so >> yeah, so Samsung uh Mark German reviewed Samsung's latest Zfold 7 and called it the first foldable phone with true mainstream potential.
46:07
>> That device is a remarkable feat of engineering with a wider front screen and refined design that you can experience first that you have to experience firsthand to truly appreciate.
46:14
Already sales are outpacing the prior generation to a significant degree.
46:18
I'm told that that makes sense.
46:21
What's interesting is like it seems like they the the the ZFold 7 is the size of an iPhone when unfolded and it folds down to be half the size.
46:33
>> I think the real demand and the real interesting thing is it's the size of an iPhone when folded shut and then it unfolds to basically the size of an iPad. Exactly.
46:42
That's what I think people would be more interested in. But I don't know.
46:47
I I I'm I I I don't mind carrying like a bigger thing.
46:50
Um I always get the bigger phone and so I I guess uh I might be in the minority there.
46:56
Um >> so Mark says that means Apple's first foldable won't break any technological barriers or redefine the category.
47:03
Samsung has already taken care of much of the heavy lifting, but here's the twist. That may not matter.
47:06
Apple's unmatched ability to market premium hardware to consumers.
47:09
Vision Pro aside could make it a dominant player in the foldables market within months of launch.
47:14
There's a sizable group of iPhone loyalists, myself included, who have long wanted a foldable device, but weren't willing to switch to Android to get it.
47:21
That pent-up demand is real and Apple knows it.
47:23
I also just think, yeah, people used to just love the novelty of a new iPhone and it was like exciting if if you had it, if you're one of the first people to have it, people would wait in line.
47:33
You know, you don't see iPhone lines to the same degree anymore because it's just not that exciting. Yeah.
47:40
>> And I think a lot of people will get the foldable version of the iPhone even if they've never wanted a foldable just because it's new and it's from Apple and they use the device every single day and >> I imagine it will be very well done.
47:53
>> Imagine running Adio on a foldable phone just pulling out your CRM. That'd be so satisfying.
47:57
Adio customer relationship magic.
47:59
Adio is the AI native CRM that builds, scales, and grows your company to the next level.
48:03
>> Customer relationship magic.
48:05
>> So, uh, the new afoldable iPhone is expected to cost at least $2,000.
48:10
So, I mean, if this if it delivers at the level of an iPhone and just becomes the highest status thing to carry, everyone's just going to go all the top Apple fans are going to go from spending $1,000 a year, every couple years to 20,000.
48:24
And then also, I bet that a change in form factor like I think my I I used to upgrade every year, then I started upgrading every two years.
48:31
I think I had the last phone for like three years and I was kind of like, I don't really notice any difference.
48:35
I got the new phone, barely noticed any difference.
48:38
I think I got an extra button on the side.
48:39
Um, but even though it's only one year old, if there's a foldable and it's like this cool shiny new thing, even if I don't get that much value, I'm definitely upgrading.
48:46
I'm definitely upgrading.
48:48
Um, and so, you know, they just they they just pull forward a ton of revenue, which could be uh very good.
48:53
And there's an interesting thing about Apple's revenue mix shifting really towards like the services and then maybe this pulls some some stuff back.
49:01
This is also extremely popular as a form factor in China. Yeah.
49:03
And so >> Mark says, "In fairness to Apple, it foldable phone won't be a carbon copy of the Samsung.
49:09
As he reported months ago, the company is focusing on addressing a few of the foldable categories long-standing weaknesses.
49:14
The company aims to make the inner display crease less visible and dramatically improve the hinge mechanism.
49:19
And as part of the development of iOS 27, which formally kicks off soon, Apple will prioritize software features tailored specifically to this new form factor.
49:28
This will be fun for app developers who have been focused on, you know, optimizing uh, you know, conversion rate at every step of their funnel to to have some new surface area to be thinking about.
49:40
Um, and again, it talks about the popularity of foldables in China, which is interesting.
49:47
Apple uh reportedly shut down their first retail store in China ever.
49:52
They had never closed a store, but uh that points to the struggles they're having uh in China broadly.
49:57
uh as people just don't care as much about the status around the iPhone that in the way that they want it once did.
50:07
>> Whether you're bullish on Apple or bearish on Apple, you got to get on public.
50:10
com investing for those who take it seriously.
50:12
They got multiasset investing industryleading yields and they're trusted by millions.
50:15
Uh breaking news from DD Doss, friend of the show.
50:19
OpenAI is dropping GPT5 next week apparently.
50:22
Uh he says the source is an anonymous DM with some reasonable proof shared out of public interest.
50:27
Some of the details might be off.
50:29
He says, "A million token window, 100,000 output tokens, MCP support, parallel tool calls, dynamic short and long reasoning, uses code interpreter and other uh tools.
50:39
Code names are 03 alpha nectarine lobster >> lobster mini starfish Nano. >> Fantastic.
50:46
They're not beating the the naming challenges allegations, but uh I mean GPT5 is is clean." >> Yeah.
50:54
And so, uh, >> I'm going to call it Starfish Nano Lobster Mini.
50:57
Personally, >> Starfish, I like it.
50:59
Uh, so Poly Market has GPT5 released by July 31st, which would be in two days at 7%, but August 15th at 74% now.
51:08
So, it feels like it's coming.
51:12
Uh, and they have GPT5 released by December 31st at 97%.
51:19
Uh, Swixes, we're seeing almost daily releases of great open models.
51:21
Now, in completely unrelated news, GPT5 and Gemini 3 drop in the next few weeks.
51:28
Swix also believes that GPT5 is coming soon, and Gemini 3 is also coming soon.
51:33
Um, and it's interesting because we're in this era of like what does GPT5 even mean?
51:38
mean? because it used to mean a major version number bump was a order of magnitude in pre-training but then we tested that with what was supposed to be GPT5 the I think it was called Orion um and uh they did this massive training run they started testing it and they
51:58
found that it wasn't as useful as it had good like big model smell it had good vibes Dylan Patel said it made him laugh and it could do good green text and we have one of them here that read through, but uh it it wasn't as useful as 03 or 03 Pro and all the reasoning and all the tool stuff. So, we're in this weird
52:15
So, we're in this weird thing where GPT5 might come out and it doesn't necessarily mean that like, okay, it's the Stargate training run.
52:22
It's the 10X bigger pre-training run.
52:25
It's the most massive LLM training run of all time, all the data.
52:27
It's more like it's just the best most polished unification of all their different capabilities.
52:33
So they're putting together the best model with the best tools and the best systems and the best interface and that's what the version number bump means.
52:41
Now obviously dropping GPT5 will be like a massive moment and everyone will be like oh my god and expectations are extremely high.
52:49
>> Expectations are extremely high but really what people want is just something that's usable.
52:52
uh and it doesn't and so GPT4 point they they they changed what was supposed to be GPT5 apparently to GPT4.
52:58
5 and and kind of and then they they tucked it behind some UI so it's harder to get to now it's not the default it was slower but people did like talking to it and they felt like it had more personality and it was funnier and of course uh this comes from this uh Tyler Cowan green text be me be Tyler Cowan uh wake up at 5:23 a. m. Precisely.
53:19
Eat exactly seven blueberries, three Brazil nuts, and a/4 cup of Icelandic sky skier. Uh, productivity. jpeg. Write two blog posts.
53:29
One on obscure Romanian cinema.
53:31
The other on the economics of parking meters in Hyereabad. Check email.
53:35
Reply exclusively in lowercase. Very interesting. Thanks for this. Send. Perhaps send.
53:39
Attend breakfast meeting with unnamed intelligence officials.
53:43
Discuss Singaporean food stalls instead of geopolitics.
53:46
Food stalls instead of geopolitics. Lunchtime arrives.
53:48
Ask waiter, "What is your least popular dish?"
53:52
That actually is really funny.
53:54
Waiter, visibly confused, recommends some fermented herring special. Wonderful. I'll have that. Extra spicy. Genuinely enjoy it.
54:00
Back to back to office podcast guest waiting is a chess grandmaster who's also an amateur botonist and former DJ. Oh, so funny.
54:10
>> Feels extremely accurate. >> It's perfect. JPG.
54:11
Uh, finish interview by recommending three obscure books on medieval Hungarian tax policy. >> Guest nodding.
54:18
>> Guess nodding politely. Afternoon arrives. Brow J. Browse JSTO for fun.
54:22
Download papers titled an economic analysis of late Roman olive oil production and optimal seating arrangements in Mongol yurts. Dinner time.
54:30
Marginal revolution commenters awaiting restaurant recommendations.
54:33
The best fried chicken in America is actually in rural Nebraska. Chaos in the comments. Chuckle softly.
54:38
Bedtime is at exactly 9:47 p. m.
54:41
Dream in in Straussian subtext. MFW.
54:45
And so Nibil Keshi says, "Honestly, the first time an LLM made me laugh out loud." And I agree.
54:51
Uh >> it seems this is the way that it makes this is how this is its first like consistent uh humor form factor. >> Totally. Totally. >> It's been reliable. >> Yeah.
55:03
So Tyler Cowan's hitting bedtime at exactly 9:47.
55:06
When you're headed to bed, get an eight sleep. Get a Pod Five.
55:11
They have a 5-year warranty, 30 night risk-free trial, free returns, free shipping.
55:14
And we have our first guest of the show in the reream waiting room.
55:18
Let's bring in Ryan Peterson. How are you doing, Ryan? >> You look fantastic. Fantastic.
55:24
>> Yeah, whatever lighting you got going on looks super cinematic. Good job. >> Good stuff. >> Oh, really?
55:29
I'm in a hotel room in Las Vegas.
55:31
Uh >> it's the big window.
55:32
There's a big window to your left and I like it. It looks awesome.
55:36
>> It's the uh Arya Arya city center right over there. >> Very nice.
55:39
Completely blown out on the >> Are we getting Are we getting a flexport orb this year?
55:43
Oh yeah, >> people have been bragging. >> Oh, the sphere. >> Yeah, the sphere. The sphere. >> Man, I want to do it.
55:50
>> This might be the first uh the first gong ringing for a sale of a company. Is that what's going on? Break us down.
55:57
Break the news down for us.
55:57
Exactly what happened and is this a gongw worthy moment? >> And full backtory.
56:02
This is like six questions in one. Just cook. >> Take us on a journey.
56:08
Um, so we sold our convoy business to DAT. >> Okay.
56:15
>> Uh, and it the we're for some reason we're not supposed to talk about the numbers involved.
56:19
So you but I have they have been reported.
56:20
I have not uh I have not protested the numbers that have been reported.
56:24
So >> um it was very good return for us.
56:26
We bought this business out of bankruptcy 18 months ago from a bank. >> Yep.
56:32
>> Uh a little over 18 months ago.
56:32
And one thing that's been reported I feel like wrongly where I thought I'd come on and clear the record is like it's been reported is that we made this amazing trade like we bought this thing and then we sold it for way more than we bought it for.
56:47
>> Uh made a lot of money.
56:49
>> That is true but it was not a trade the way you buy like a stock and do nothing and you make a bunch of money like you're some kind of genius.
56:55
This was >> an act of entrepreneurship and business building.
56:59
We bought a a bankrupt we only bought the asset.
57:02
No employees, no customers, no transactions, no nothing.
57:07
Um, >> had to go recruit the former employees of Convoy to come back to work for us. >> Wow.
57:12
>> We had to recruit the carriers.
57:12
So, the way the Convoy business works is that at one at peak, this thing was worth $3. 8 billion, by the way.
57:18
Um, and you can go read what we bought it for. It was a lot lot less.
57:22
Uh, and >> but it again, it was nothing there.
57:24
We the team did an incredible job like real business building here to go get all those carriers back.
57:30
So the way it works, you have at at peak they had 400,000 drivers on a mobile app, truck drivers. >> Yeah.
57:37
>> Uh well, we brought it back to life and those drivers, many of them hadn't been paid by the former ownersh.
57:45
>> There was really like no transaction volume. Nothing happened.
57:46
We had convinced them to come back.
57:47
It was kind of solving the cold start problem of any marketplace.
57:51
Like how do you get loads in here to attract the drivers, the drivers to attract the loads?
57:55
Pretty hard problem when you start from zero.
57:59
Um, and Flexport is not a big truck broker.
58:02
>> So, we didn't have the demand to do that, to bring it back to life.
58:04
So, what we did, we completely pivoted the business model to be a neutral platform to say, okay, other truck brokers can come in here and post their loads and bring it and we'll have the drivers come back to serve those loads and and get the flywheel spinning.
58:18
So, that was the the change to the business model that we made.
58:22
And 18 months later, we had hit real liquidity kind of escape velocity.
58:27
We're uh probably on track to have like more than a hundred truck brokers by the end of the year, each posting thousands of some of them tens of thousands of loads to the platform, some over 100 thousand.
58:37
Uh and then the drivers would sign up.
58:40
You know, if you have demand, the drivers will come and they they all still had the mobile app. They liked Convoy.
58:44
They wanted to come back >> if there were loads.
58:46
Um so, but the moment we pivoted it away from serving being a truck brokerage, >> it was kind of also not a strategic fit for Flexport.
58:56
like our job is to serve importers and exporters and companies with cargo to move, not to serve other truck brokers with whom we sort of compete.
59:03
So, it was like this strategic mismatch was always an afterthought when I talk about our mission and what we're going to do and why we're doing it. >> Be like, "Oh, yeah."
59:11
And we also have this like business that serves truck brokerage, which is an incredible business and it deserves to be front and center. So, we sold it to DAT. DAT is the load.
59:20
It used to be called Dial Truck, which I love.
59:21
It tells you their heritage.
59:24
>> Um, been around forever.
59:24
Yeah, >> they're kind of the 800 pound gorilla in load boards.
59:29
>> What load board is like if you have you're a truck broker yet you can't call every truck driver in America to match your load >> y >> to pick it up.
59:37
And so if you don't have a contract or you don't have relationship and none of the carriers you work with, the people that operate trucks is able to service your load, you'll post it to >> DAT >> and then hope that somebody's kind of like Craigslist for freight.
59:49
Uh well, like Craigslist, you know, there's they've got um it's hard it's a hard problem to control like who's actually picking up the freight.
59:57
Is it the Eastern European mafia that just stole your stuff and disappeared?
1:00:00
Like some of that goes on.
1:00:02
Uh, and >> I just I just saw I just saw a a friend's company this morning had somebody show up at his warehouse, looked completely legit, like an official kind of um kind of a distributor kind of purchase order and just like got over a million dollars of product and then just disappeared. >> No way.
1:00:23
>> And I think that I'm sure I'm sure that happens like quite frequently.
1:00:24
It probably is through this kind of scam um where people and it's like >> now convoy built incredible machine learning based detection algorithms for this >> uh to go like do pattern matching and find it.
1:00:38
So we've actually um o over the 18 months that we operated the business doing hundreds of thousands of loads zero incidents of freight and tons of busting people.
1:00:47
We we even got a guy trying to commit freight fraud >> uh from the courthouse >> and it was a known actor who was like connected through these pattern matching algorithms.
1:00:58
And so they and we have GPS coordinates because you're doing it on the mobile app.
1:01:01
So they're like, "Oh, this guy's at a courthouse right now."
1:01:03
So we looked up his records and he was facing trial for freight fraud. >> No way. >> Freight fraud. >> Wow. >> That is crazy.
1:01:12
>> So yeah, it's it's really rampant and that I I think that's going to be one of the biggest wins from all this.
1:01:15
I think DAT is going to make a ton of money doing this.
1:01:18
Better the business model, making a percent of every transaction instead of currently they just sell subscription.
1:01:23
So >> DAT is part of Roper, by the way, public company.
1:01:26
Their stock didn't move on the news.
1:01:28
I think Wall Street didn't quite understand how big this can be for them. >> Sure.
1:01:31
>> Um not not investment advice, but but there's something there.
1:01:34
Uh it but the big and flex does great.
1:01:36
We we got uh put a lot of money on our balance sheet, help us >> on our mission.
1:01:41
the Convoy team is going to get to scale and instantly put test their algorithms at scale, right?
1:01:46
Test the matching like because DAT is the monster.
1:01:51
Like something like 40% of all the truckloads run on the DAT load board. >> Yeah. Wow.
1:01:55
>> And so piping them through convoys like let's see how that works.
1:01:57
It's going to be awesome.
1:01:59
But the industry in general, like yeah, I think it's a massive way to combat fraud.
1:02:04
uh truck brokers will really benefit because about 30% of the expense structure of a truck brokerage is of an FTL brokerage is um what they call carrier sales is these are the guys that are making phone calls to carriers to try to cover the load to say okay I got
1:02:20
this load I got the customer >> and let me find a trucker that'll move it for me >> um and with convoy that goes away you don't have to spend that money um or those people can work on harder loads and more interesting problems go do more sales So, um, huge win for the industry. I think probably the biggest beneficiary
1:02:35
I think probably the biggest beneficiary of this all.
1:02:37
So, yeah, pretty proud of how it went down.
1:02:40
>> How did Convoy go bankrupt originally?
1:02:43
It was I I remember kind of watching that unfold quickly and then you guys moved super fast on it from from what I remember. >> Yeah. Um, a lot of theories.
1:02:52
I'm sure that, you know, it' be interesting to get the Convoy founder Dan Lewis on your show at some point, see what his ideas are, what what he thinks, where he went wrong.
1:03:01
Um there's sort of the proximate cause which is you know every company goes bankrupt for the same reason you run out of money.
1:03:06
Um they they were in a um they were in a transaction um they were they had agreed to sell the business to a large competitor of theirs big public company.
1:03:18
I'll go unnamed I don't want to get sued but big company >> um who they should have they should have done and it was going to but because they had a deal to sell and then that big company unwound it.
1:03:32
He had not gotten proper y >> he hadn't gotten proper board consent and he had a couple activist investors who didn't want didn't like the deal. >> Interesting.
1:03:43
>> And backed out with and convoy had two weeks of cash left and they were supposed to close >> but like you never should have got yourself into that situation in the first place.
1:03:52
>> Um >> so especially you know you kind of allow your competitor to do that for to you like don't let a win for the Yeah.
1:03:59
we can we can buy you for a lot of money or >> we can just put you out of business and then we're just like you know more strong.
1:04:06
strong. So how did this particular >> so your your your big >> I think that's that all of that is sort of proximate causes upstream of that go back a few years my my if I if I ran the business it would have been way more salesheavy I think they were really
1:04:25
product and tech driven like we're we're doing hundreds of thousands of loads through convoy >> only 2% of them >> a human has to touch at all the tech is dialed >> so like >> but You were able to bring your sales force to bear pretty quickly to ramp up the business. >> And just in general, like it's kind of
1:04:40
>> And just in general, like it's kind of like a a it's a Midwest kind of steak dinner industry. >> Totally. Yep.
1:04:47
>> And they're like West Coast product people. >> Yeah. Yeah. Yeah. No. No. That makes sense.
1:04:51
How did >> I needed a little more steak dinner action?
1:04:55
>> How did this particular steak?
1:04:58
It's all about this stake dinner capital of the world.
1:05:00
>> How did this particular deal come together?
1:05:01
Is this like there's an investment bank involved?
1:05:03
Is this like founder to founder deal?
1:05:04
like uh like walk me through a little bit of the anatomy of like how you sell what is essentially like a subsidiary.
1:05:09
I've never been in that situation.
1:05:11
I think a lot of founders haven't been in that situation. >> Yeah.
1:05:14
Um well great businesses are bought and not sold.
1:05:16
So totally >> they approached us.
1:05:19
>> Um >> I think they they started to realize we were getting escape velocity on the marketplace and that at scale that would be quite threatening to them. >> Sure.
1:05:28
>> Uh because you don't need to go to the subscription and uh load board if you load is already covered. Sure.
1:05:32
uh via our platform and our business model >> now that their business model I think is better uh at scale when it works because you're taking a percent of every transaction instead of just a monthly fee. >> Um >> that's cool.
1:05:46
>> Like by my math, >> very amateur math, I think this could add like5 billion dollars in EBIT. >> Wow. >> To the business. Like this is crazy. Yeah.
1:05:57
>> So, it's very valuable.
1:05:57
I'm sure that's off, but um but like order of magnitude is right.
1:06:01
Um and so I think they I don't know how they they decided but there's a relationship.
1:06:09
Uh D is a big player uh in the industry obviously and our team convoy team was just connected to them and they were like hey we got this inbound and they're they're pretty interested in selling the business. What do you think Ryan?
1:06:20
I'm like well I have to pay a lot money for it because I really like it and but the more we looked at it the more it was like dude it's worth more to them.
1:06:29
>> I mean I think that's the core in any M&A transaction.
1:06:30
It needs to be worth more to the acquirer than it is to the >> yeah totally >> seller.
1:06:35
Uh so that there's room there's grounds there for a negotiated price a settlement and we found a good one.
1:06:40
So >> well congratulations on the deal.
1:06:42
Let's ring the >> amazing amazing work from the team. Super impressive.
1:06:47
>> We need to warm this up. >> Yeah.
1:06:48
We've been told we got to warm up the gong.
1:06:50
We're going to break it if we hit it too hard. >> There you go.
1:06:54
[Music] >> People are very people are very opinionated about about the gong hits.
1:07:00
We we really do our best here.
1:07:02
Super super I mean uh given given, you know, everything that that Flexport has navigated through the last few years.
1:07:10
It's pretty incredible to just casually as kind of like a little little side project buy, you know, just turn it around and sell it fantastic in such a short period of time. So nice work.
1:07:23
>> You're a testament to the Convoy team, frankly.
1:07:24
frankly. like u and I hope they all that we we feel like we've put them off in a in a good place here u financially but like the potential is all what can they do for DAT >> and how does DAT make the best of this right um because man if you can pipe all
1:07:42
that volume through this platform you have the potential to generate something really really massive >> it's exciting well thank you so much for stopping by thanks for taking some time out of your day in Las Vegas good luck with the rest your steaks out there and enjoy the steak. >> I will be having a steak dinner tonight.
1:07:56
>> I will be having a steak dinner tonight.
1:07:57
Actually, Top Golf with the customers.
1:07:59
So, >> that's Top Golf steak. You love steak.
1:08:03
That's what it's all about.
1:08:04
>> That's the American dream right there. Have a good one. Good stuff, Ryan. Great.
1:08:07
>> We'll talk to you later.
1:08:07
Uh Dan Primac has a report says Flexport sells Convoy assets to DAT for $250 million.
1:08:12
So, while plenty of people can't comment on it, uh Dan Primac has the scoop and we will as he often does.
1:08:20
We have Ty Haney in the reream waiting room. Let's bring her in. How you doing, Tom?
1:08:25
>> Speaking of comeback stories, what's happening?
1:08:28
>> Welcome to the stream. Good to meet you. How you doing? >> Yes, I'm doing great.
1:08:31
I'm actually reading a tweet from Emily Sunberg.
1:08:32
I know she's one of your best friends. Mine, too. >> We are good friends. >> Amazing. Are you in New York? >> I'm in SF. >> Okay, cool. Uh, give us >> Yeah. Where? Yeah.
1:08:43
So, SF, where uh I mean, I want to get into everything, but uh where is that where you're going to be basing >> the the company out of going forward?
1:08:53
How how are you how are you thinking about that?
1:08:55
>> Uh PYB my rewards platform.
1:08:55
We actually work out of the X Twitter office on Market Street. >> Cool.
1:09:02
>> So it's >> Wait, the X like the X office? The X office?
1:09:06
>> The former office X office or the current X office?
1:09:10
>> The the last X office, former Twitter office.
1:09:16
>> There's like old Twitter merch uh which we should probably put places, sell places, and then there's some X branding.
1:09:21
And then now there's QIB. So >> cool.
1:09:24
>> It's worth we will market by all the old Twitter merch. >> Yes. Yes.
1:09:27
I saw there was like a huge bird sculpture that was going up when Elon first bought it.
1:09:31
He was thinking about auctioning this stuff off and it was like $60,000 for this massive bird logo.
1:09:36
Like I don't want this bird logo anymore.
1:09:39
There's a bunch of >> whatever you got. Well, congratulations.
1:09:43
Why don't you give some quick quick background on yourself for the few people that might be living under a rock >> that do not know and then we'll get into the the news from this week.
1:09:52
>> Yeah, I'll give you the quick walk up.
1:09:52
I started a company in the active horse space called Outdoor Voices at 23 out of school in New York City.
1:09:59
>> It really became a darling in the direct direct to consumer world.
1:10:01
I ran it for eight years to 100 million. I left five years ago.
1:10:06
>> Uh there was kind of an ego battle at the board level and then between me as the CEO and this person.
1:10:10
Um I left uh and that was a sad point of time.
1:10:14
Um 90% of my experience building that company was awesome, 10% was hard.
1:10:19
That said it was a masterclass in learnings.
1:10:21
Anyway, since then I started a tech platform, a community rewards platform called Try Your Best.
1:10:28
>> We work with the top consumer brands to essentially create a more fun and effective loyalty program.
1:10:32
All that to say, I announced yesterday that I rejoined my original company, Outdoor Voices. Um >> amazing.
1:10:40
And so pretty cool to see there was a very warm response to original OG OVers and here I am.
1:10:46
I'm I'm running TYB and I'm running OV as well. >> Amazing.
1:10:52
>> The the the video was amazing.
1:10:52
Uh maybe we'll play it after you jump off, but uh my wife said she almost teared up watching it.
1:10:58
It just was like it's such a such a cool story and uh yeah, I think everybody's excited to have you back in the driver's seat.
1:11:05
So what is that what is that going to look like going forward?
1:11:08
How are you going to like what what what what do you think uh what do you think the takeaways from the last go are and how are you kind of what you know where do you want to take the brand going forward?
1:11:20
Is it back to back to its roots back to what you know people loved originally >> more retail or e-commerce shifting the mix?
1:11:28
I'd be interested to hear about that.
1:11:30
>> Sorry, can you say that again?
1:11:31
>> I I'd be interested to hear about the the the a lot of brands uh you know they they start online then they go offline.
1:11:38
There's owned retail, there's channel partners.
1:11:39
Uh, how do you see that mix changing going forward?
1:11:41
How did it change while you were there or over the last decade? >> Yeah, 100%.
1:11:45
The northstar vision for OB is to build the number one recreation brand.
1:11:49
Uh, our call to action or kind of motto that people have a crazy emotional connection to is doing things.
1:11:54
You guys might have seen the Doing Things hats like trotting trotting around the west westside highway.
1:12:00
>> Um, >> that was kind of like a souvenir from the brand that people would wear, but you could only get them by participating with the brand.
1:12:05
All that to say, like we started in 2013, so we were very much part of this direct to consumer model that I'd say in a lot of ways has not proved to be all that successful.
1:12:14
Um, and so similar to others, we were able to raise a lot of capital.
1:12:18
That meant I was diluted pretty significantly as the founder.
1:12:21
Um, and ultimately we spent too much money trying to acquire people, you know, online that were expensive and didn't last.
1:12:28
didn't last. Y >> we also kind of had this grassroots um what would I call it like community uh of 10,000 or so or so people that were really our super fans and as we clicked into the data like anyone who came through this community activation was
1:12:44
ultimately four times more valuable over time and so that's essentially what prompted me to start TYB to allow brands to directly connect with their super fans and grow their value over time own that relationship own that data um and then now with OV me it's been cool. I've
1:12:58
I've been working on it for a year. Um >> Oh wow.
1:13:03
>> I I rejoined obviously as the founder as an owner with new owners um and a partner, but the last year has been just almost like a cherry on top.
1:13:10
It's been a creative outlet.
1:13:12
I obviously spent eight years like obsessing over every detail of OV >> um to to kind of like play again in this space and think about from a product perspective like the recreational lifestyle and what the core products are here.
1:13:25
It's just honestly been fun.
1:13:25
Um, and so no change in terms of the vision, but we're off to the races.
1:13:31
>> I feel like there's such an opportunity for almost like a turnaround fund, venture capital fund, private equity fund to specifically target uh companies that where the founders left and we're going to set them up to bring them back.
1:13:42
I think that there's so much more value unlocked from that as opposed to just like, okay, let's trade this asset to someone who's going to squeeze even like the last few pennies out, just kind of wind it down and bleed it out as opposed to let's there's actually something here, but we need a founder back in the back in the seat and we need to bring them back and set them up.
1:13:59
The cool thing too is is it seems like every apparel brand that went ultraheavy into online only, not online only, but made that the priority struggled all the way from like Nike, you see their their like DTOC efforts.
1:14:12
They basically capitulated there and realized like, okay, people like going to stores, like they like experiencing person, they want to be on all these different channels.
1:14:21
And so I think it was just a massive lesson for everyone in the industry.
1:14:26
I'm curious how you see the like kind of broader athleisure category evolving.
1:14:32
It feels like there's kind of distinct categories around >> Nike is still for athletes, but Nike kind of missed your insight with OV was like, okay, what is like active wear for people that don't consider them like don't aspire to be a professional basketball player or people that aren't like weight training.
1:14:49
Uh, and then we've seen, you know, the rise of aloe targeting kind of the, uh, Pilates industrial complex, um, and kind of the lifestyle around that.
1:14:59
So, I'm curious kind of where you see kind of the broader market going.
1:15:04
>> I'm curious, what do you guys wear active wear wise? >> I wear cotton. >> Yeah.
1:15:10
>> Like I wear like cotton like tank tops and and shorts.
1:15:14
>> I I actually don't know. >> John doesn't know.
1:15:15
John's one of the John's in the category. My wife shops for me.
1:15:18
So >> yeah, John, she's like, I think I need new shorts.
1:15:21
And then shorts appear from his lovely wife.
1:15:23
But I back in the day, I I would wear I remember I was the first person in my friend group in college to discover outdoor voices.
1:15:32
>> I was early to Lululemon at a time when, you know, men were just criticized for wearing >> uh what people saw as like a women's brand, and I was like, "The product's just great. Try it."
1:15:42
And then people would try it, and it was great. And then I got on OV. Okay.
1:15:44
Um, and uh, but but yeah, I think uh, >> yeah, for me I I think I'm excited about like the return of like like natural fibers in in sportsware.
1:15:57
Yeah, >> I'm excited about that.
1:15:59
But, uh, it's it's a broad category.
1:16:03
>> Yeah, I I asked that because for this fiveyear gap, like similarly, I would only wear vintage.
1:16:08
I feel like every brand every brand in the active space says so much about you.
1:16:12
like Alo wasn't for me, etc.
1:16:15
So, >> uh I also think of this space almost like the like artist space where there's like different flavors of pop stars and each of them have like equal amounts of fandom or or cultiness and they all can coexist.
1:16:28
One thing I'm excited to kind of really blast away from is just this oversaturation of like compression top and bottom and that like being the look.
1:16:38
So, our collection launches next week and like we've really infused more styling options and to your point like that includes like cotton popppllin striped button-downs that you wear to like protect from the sun and like put over your Pilates pants or cool >> uh cotton cotton cashmere cropped cardigans.
1:16:56
Similarly, like I'm going to the grocery store in that.
1:16:58
Uh yeah, >> I think it's like time to infuse more style beyond just like stretchy, you know, bottoms and tops.
1:17:05
So that's that's certainly what you'll see from us starting next week.
1:17:10
>> How are you integrating uh TYB uh I'm assuming that's going to be a big part of the launch. >> It launched today.
1:17:15
So OV's community launched on TYB today uh in a very full circle moment.
1:17:20
We're off to the races there.
1:17:22
Our designers are in there.
1:17:22
We're essentially like getting all of this amazing feedback in terms of what people have missed from a product standpoint, what hero products they want to bring back.
1:17:32
Like it's shocking to me how much people care about out their voices 5 years later.
1:17:37
Like holy [ __ ] But but uh OB is like perfectly suited for this toolkit now.
1:17:42
It's a tool that I wish I had as a brand builder 5 years ago.
1:17:45
>> How do you uh does uh what what comes to mind uh when you hear the phrase grow at all growth at all costs?
1:17:51
It feels like one of the challenges with building an enduring lifestyle brand, apparel brand, whatever you want to call it, is is the DTOC era was like, "Okay, here's $50 million.
1:18:01
Turn this into as much revenue as you can as fast as possible."
1:18:07
>> And when you look at the great brands throughout history, very few of them had founders that were that that where growth at all costs was like, >> "We know it's a marathon.
1:18:15
We know it's a marathon, but what if it was a sprint?" >> Yeah. Exactly. Yeah.
1:18:20
But it's like how it's more like, you know, I I can see the path to outdoor voices getting to a billion dollars in revenue in in 10 years, but if you tried to get there in three years, like it probably make the job a lot harder. >> Yeah. Becoming takes time. >> Yeah.
1:18:36
You'd have to have too many shortcuts. >> Totally. >> 100%.
1:18:38
I mean, for me, like success looks like longevity, and I've felt that like pain of not having longevity here firsthand.
1:18:45
So if you think of obviously like Patagonia and Ivon Chinard like I met with him somewhat recently and like that's the model for success.
1:18:51
I think I think in terms of this like lifestyle like physical product brand at the same time like I went from software with a W to software W in terms of my businesses.
1:19:05
So I understand how tech companies like deserve the capital and require the speed just by nature of they don't have physical products or widgets that they're trying to you know get right from a planning perspective and then sell like so it just makes a lot more sense to me why funding around software companies like is the way it is and the speed that's required and then possible there I've very much just felt firsthand.
1:19:28
So, I'd say like winning to me, yeah, is around longevity, endurance, and no pun intended, it's an active word brand.
1:19:35
>> Can you take me on a tour of some of the good community, maybe good and less good community building stuff.
1:19:39
We were talking about this uh Emily Sunberg highlighted a candle company that put a bunch of influencers on a private jet and the private jet maybe didn't even take off.
1:19:48
And uh and it feels like that is a unique unexplored space of like take the super fans, whether they're influencers or not, and just do something way above and beyond for them.
1:19:58
Even Red Bull went back and did this with like the supercross community.
1:20:00
It was just a bunch of guys driving motorcycles on dirt ramps and they were like, "Let's turn this into a real sport."
1:20:06
That sport came back to, you know, love them forever. They did this with F1.
1:20:10
There's companies that have done this really well, but then it can go less well.
1:20:12
So, like I I understand that like, you know, surveys are valuable, but like what how else can you integrate uh or or activate a community?
1:20:21
what I'm like what we're seeing really matter on the user side is status and identity as essentially like signaling my proof of fan and like so giving people the ability to >> level up and like prove their fandom and then have something like as a bragging right that they can share on social around that >> is ultimately like what they care about beyond anything.
1:20:41
Um, and so like that proof of fan idea, I think like loyalty becomes kind of the new identity or status and like your ability to to signal that like really matters.
1:20:52
>> Um, what else would I tell you there?
1:20:54
>> Wait, so the proof of fan, does that need to be like qualitative or quantitative?
1:20:56
quantitative? Because there's a world where it's like, uh, okay, like you're in like the Kohl's cash world where like you buy $1,000 and you get 50 bucks back versus this like more qualitative world where it's like, look, you came to some
1:21:10
event and you bought some stuff and like you were, you know, in you were we saw you in our replies on social a lot and like we didn't really try and put like a send five tweets and you get a free hat, but like how how quantified does it need to be? because I feel like it can feel a
1:21:23
because I feel like it can feel a little sterile if it becomes an algorithm and then people are gaming and then you have bots getting the hats and then reselling them and stuff.
1:21:31
>> How do you think about that?
1:21:33
>> I think it's it's definitely all about social.
1:21:35
So like within my profile I'm essentially collecting things from, you know, events I've gone to.
1:21:41
So Rare Beauty is a good example. They do.
1:21:45
They're they're excellent at essentially like creating these little collectibles or souvenirs that that their fans can collect as they launch new product or host an event.
1:21:54
And essentially what 2IB allows you to do is like build your consumer profile, but it's very visual first.
1:22:00
And so you're earning these things that then essentially are like your top shelf.
1:22:04
Like here's what I'm obsessed with.
1:22:06
I've bought it seven times.
1:22:07
Here's a Selena Gomez event I went to.
1:22:10
It sits next to kind of, you know, my streaks or for lack of a better term in terms of uh product purchases.
1:22:16
And all of this like essentially builds your identity in a really social first way.
1:22:21
Um and so we see people share like this wallet or pro profile content just reflecting kind of their engagement with brands in a run really fun kind of like social gameled way.
1:22:31
social gameled way. The other thing just to your guys's point I think there is certainly like oversaturation of brand events is like the way I think about building fandom it's kind of four-part like one articulate the purpose or
1:22:44
mission very clearly of the brand like why do people care what's the reason for being two build rituals for activation and obviously we're seeing like run clubs take off but like the key for me there is like something that you as a brand can do consistently not just one and done. >> Yeah. >> Yeah.
1:22:59
>> The third is connection.
1:22:59
And so like allowing brands to like or bands that like obsess over Nike or whatever to connect with one one another and like that flywheel of engagement really takes off.
1:23:09
And then the fourth is incentivization.
1:23:10
And I think for too long fandoms been like clicking like on Instagram but like ultimately that's not giving me anything for my loyalty over time.
1:23:19
>> So there's a lot to play with as you think of kind of like the combo of those things.
1:23:22
But I agree like brand events there's too many like I don't care. >> Yeah, makes sense.
1:23:27
What's the How are you thinking about product release schedules coming back to OV?
1:23:30
You've been working on this first collection sounds like for a year, but are you thinking >> drops, collections, capsules, a mix of everything?
1:23:42
>> It's just a consistent drum beat of of excellent newness.
1:23:45
And I think we've always like done a good job of having a kind of stunty or unexpected perspective on in the space.
1:23:52
And so we have a another collection dropping uh in September that's like a very specific activity like not going to be relatable to all but like something you certainly haven't seen Nike do before.
1:24:02
And I remember like early innings outdoor voices uh on our site we merchandised by a category of dog walking.
1:24:10
Sounds simple, no non-issue, but like Reddit went nuts.
1:24:14
They were so mad like dog walking is not a sport and that was perfect.
1:24:17
I'm like that's so it's so ownable.
1:24:19
Dog walking is how dare you make a pocket for >> get in the Olympics.
1:24:23
I want to see Olympic level dog walking for sure. For sure. That sounds amazing.
1:24:28
Thank you so much for stopping by. We got to jump.
1:24:30
Super excited to see you. Congratulations.
1:24:33
>> We got to ring the gong for a founder back in the seat. >> Founder mode. >> Founder mode. Did you have the sound? Founder mode.
1:24:39
Thank you so much for hopping on. We'll talk to you soon. >> Great to see you. Cheers.
1:24:43
>> I'm excited for Outdoor Voices.
1:24:43
Maybe they'll do a billboard.
1:24:45
Maybe they'll get on adqu. com.
1:24:46
Out of home advertising made easy and measurable.
1:24:48
Say goodbye to the headaches of out of home advertising only.
1:24:51
Adqubines technology out of home expertise and data to enable efficient seamless ad buying across the globe.
1:24:55
And next up we have AGM coming in probably talk about ads too because the timeline was in terms we were talking about AI uh AI ads but I want to talk about a whole bunch of stuff.
1:25:07
Uh kick us off with an introduction. How you doing? >> I did good.
1:25:11
Um I'm uh talking I dressed it up for you guys by the way.
1:25:14
I've got fantastic on as Jord mentioned.
1:25:16
I've got the Patagonia fleece.
1:25:17
I look like I trade bonds at Caner Fitzgerald. >> Yes, you do.
1:25:21
>> You guys trading uh trading tariff rebates today. >> That's amazing. >> Cool, man.
1:25:29
So, yeah, I'm >> What's new in your world? >> What's that? >> Yeah. Yeah.
1:25:32
Just what's new in your world?
1:25:34
>> So, bit of a product announcement today.
1:25:37
Um we a little bit related to a guest you had two weeks ago.
1:25:39
You had Brian and Jesse on there from Coinbase. Yep.
1:25:41
I'm here to announce uh we're launching ads or sponsored offers as we like to call them inside the new app that Coinbase launched uh which is a new social app on chain.
1:25:52
Um it replaces the previous wallet.
1:25:54
Uh it's really quite cool and interesting and um >> yeah, I mean it's it's kind of the fulfillment of the spindle vision.
1:26:00
So we recently got acquired by by Coinbase. >> Yeah.
1:26:04
>> And um you know I'm I I was a little late to crypto to be honest.
1:26:06
That's how I bought some Bitcoin in 2013.
1:26:08
But um >> you know circle 2022 when board apes were worth like a million dollars and FDX hadn't happened yet and crypto was pumping.
1:26:16
I mean it's pumping now but it's pumping even more.
1:26:18
>> You know what crypto needs ads. >> Well no no no no no.
1:26:21
That was not the first thing.
1:26:23
The first thing is I talked to I I talked to the head of marketing.
1:26:25
I won't name the names of like billion dollar crypto companies and I asked okay so like what's your CAC? What's your LTP LTV?
1:26:31
What does your retention look like?
1:26:33
>> And the guy looks at me and says what's LTV?
1:26:34
I I I don't know what you're talking about. >> Yeah.
1:26:36
And I I I basically had a personal existential crisis that I couldn't believe that people in crypto didn't know their LTVs.
1:26:41
And so we actually >> What do you mean the number the number goes up? That's all that matters.
1:26:47
>> Well, no, that's that's that's part of the problem in crypto.
1:26:48
Often numbers go if money is free, then cats don't matter is kind of the problem.
1:26:52
But um but money doesn't always stay free forever.
1:26:55
So, um, we first built what's called an attribution system, which I know sounds super wonky and may not be totally familiar, uh, to to all your listeners, but basically, whether you realize it or not, every time you click or install an app or monetize on your app, an event is getting fired into an attribution system that's trying to figure out where you came from and how much you're worth.
1:27:12
And this is like a billion-dollar business.
1:27:15
It's incredibly disorganized.
1:27:15
There's literally billions of events firing a day.
1:27:18
I worked inside one of these big attribution companies.
1:27:19
And I just remember like, what if we just had one database?
1:27:23
What if everyone just agreed what happened in the world?
1:27:26
And then when I really got into blockchain and Ethereum and ecosystem, I realized, oh, look, this is the consensus layer of truth that we actually need and we also have user identity and we also have a payment layer.
1:27:36
The blockchain is going to be the biggest marketing database in human history, right? Yeah.
1:27:40
>> And so the question is, how do we build it?
1:27:41
How do we take the best learnings from web two world and then build it into the onchain world in a way that by the way respects user privacy and all the rest of it, but also kind of functions in in in a much better way than than Web 2 ever did. >> Yeah.
1:27:51
I remember hearing the original pitch for spindle and being like uh there was a lot of like froth and stuff and I was like this feels like enterprise ass for crypto.
1:27:59
This feels like it's like valuable regardless of what happens with everything else.
1:28:02
All the craziness you don't need to you know feel about like oh what's the value of art it's like so subjective but it's like it was like you and like that other company chain analysis I was like no matter what happens in crypto chain analysis is going to be a good company.
1:28:14
Uh so congrats on on the uh on the deal with Coinbase and coming into the team.
1:28:19
uh bunch bunch of questions about how this actually manifests.
1:28:21
I imagine that the first advertisers will be web 3 companies.
1:28:26
Is that the correct way to think about it or is this kind of untapped low CAC for outdoor voices or something? >> Yeah. Yeah.
1:28:33
And by the way, you you totally put your finger on John.
1:28:34
My my goal in life is providing actionable analytics and enterprise SAS in the morning. >> And I I love it. I I'm not joking.
1:28:40
I genuinely love it cuz it's like I the value is very traceable and not built on any sort of like okay there's like some perpetual motion machine over here that we're going to find out about.
1:28:50
It's like no like this is just like what people want. >> You wake up at 2 a. m. You wake up at 2 a. m.
1:28:56
What's your first thought? Enterprise workflows. >> Let's go. I love >> work. >> Um >> that's legit. Uh so yeah. Yeah.
1:29:02
First users, I imagine you're onboarding folks right now. Uh who who loves this? Who wants this? >> Yeah. Yeah.
1:29:11
I mean it's it's heavily it's heavily biased towards DeFi option.
1:29:15
When we started Spindle actually I thought gaming and web3 gaming was going to be like a thing which who knows it it might still be but as everyone knows gaming was one of the biggest drivers of like mobile ads attribution and like that whole that whole cycle from 2010 2015 was basically mobile casual games and zinga and the rest of it and I assume that would be the earliest adopters.
1:29:31
Web3 gaming I I understand is still the perpetual technology of the future.
1:29:34
future. uh you know it's it takes a while to make these AAA games in Q4 maybe it'll change but actually it was it was DeFi apps actually because of course they actually I mean there stable coins have PMF and DeFi does have PMF and there's protocols like some of our advertisers Morpho I can call out a few other ones that have billions of dollars of TVL on chain in which people are
1:29:51
actually earning and you know they they they basically function like a lending and and borrowing institution and so yeah they're they're the biggest ones and what's interesting is that getting back to LTVs right like the MAUs in
1:30:02
crypto might be relatively small say to the WhatsApps of the But the LTVs, like the actual individual user value, LTV or RPO, however you look at it, are can often be significantly higher, right? Like like or where where have where
1:30:11
Like like or where where have where where have companies like Morfu advertised to date?
1:30:16
Are there like if they're if they're DeFi app?
1:30:19
Are they running can they even run ads on on Meta?
1:30:23
Is that is that like >> Yeah, I mean that's it's Morpha, by the way, because if I don't if I don't correct you, I'm going to get an email from the the CEO in a second.
1:30:29
Um but um yeah, the answer is nowhere, Jordy.
1:30:32
And and that's the problem, right?
1:30:34
Like we built all this amazing measurement technology like comparable to the best in web 2, but it's like the growth guys like this is all great, but like what do I do with this?
1:30:41
Like where's the other leg of the flywheel where I go and like target the users and figure out and provide like a compelling user experience. It didn't exist.
1:30:47
And so Spindle's big obsession and part of the reason why we joined Coinbase is, you know, consumer crypto where you have a wallet connected user who actually interacts with the chain is is an important step and you're seeing more and more of that.
1:30:59
Just to site a random company that I'm not involved with at all, Stripe bought Privy, right?
1:31:01
And why did it buy Privy?
1:31:04
Because Privy is a wallet provisioner and it gets people on chain in a very seamless way.
1:31:07
And then boom, suddenly a user's on chain.
1:31:09
They don't even realize it, but they are.
1:31:10
And so we've really kind of exclusively focused on really compelling crypto consumer experiences where that where that wallet is present and where the user can engage with things on chain, whether it be content coins, NFTts, lending and borrowing, sending money to their family overseas, whatever it might be. >> Yeah.
1:31:26
How how will uh ad like what what role will play will ads play in the overall base ecosystem?
1:31:31
I imagine some of that money might flow like where does it flow to?
1:31:39
>> Um well it's a good question.
1:31:39
By the way, one thing I have to mention and we've got like a blog post dropping on it.
1:31:44
Everyone hates the word ads.
1:31:44
Like it gives everyone the if in crypto, but just just as a quick little comment, everything in crypto is an ad.
1:31:49
Like literally you scroll through every wallet and every publisher, every mint, every swap offer, all of it, whatever you're looking at usually gets a referral kickback as a function of the value of the transaction.
1:32:00
So like literally everything is an ad.
1:32:04
>> Well, and and here's a funny story.
1:32:04
So back in 2021, I uh my my team at Party Round, we made >> we realized if you send someone an NFT, it will just sit in their wallet that other people look at and track.
1:32:14
And if they want to get it out of their wallet, they have to figure out a way to burn it.
1:32:19
Send it send it off into the ether. So I was like, great.
1:32:22
So I can spend five, whatever it was, $5 to send an ad that's just permanently sitting in their wallet.
1:32:27
And so we sent like NFT ads out to a bunch of different people.
1:32:30
And I was like, oh, I can go to Dylan Fields wallet and see an ad here. That's >> really great. Really great actor.
1:32:36
Really, really, really white hat work.
1:32:39
Jordy, >> to be clear, Jordy, I hate to break it to you.
1:32:42
Most wallets figured this out and they don't surface uh the junky NFTs anymore. was I was I was early. I was early.
1:32:47
I >> I'm the reason I'm the bad actor that caused that change.
1:32:51
>> No, it was there was much worse actors that were like if you click one of these NFTs, it'll like take your all your money. Yeah.
1:32:57
This was this was just >> um Can you can you >> can I question?
1:33:01
I'm going to answer your question though.
1:33:02
Where does the revenue go?
1:33:03
So part of the spindle vision was always that like the publisher could be anybody, right?
1:33:08
So like an influencer who like pumps a thing or endorses a thing and has an audience, you know, affiliate marketers, consumer apps, anybody can be kind of a publisher.
1:33:16
And and one of the unique things like like our ambition is not just to like equal Facebook ads is actually to be better and a little bit more user conscious than than web two ads.
1:33:26
And one way we're better is that we can pay people that otherwise wouldn't kind of get paid, right?
1:33:30
If if you're a I mean on YouTube you do get a rev share with ads but broadly speaking and then with Elon Bucks you get a little bit of the Twitter thing but broadly speaking if you're a creator and you have an audience you have to monetize in some other way and it's really just your top of funnel.
1:33:42
>> Sure in the case in the case of BAP that's that's not going to be true.
1:33:44
Um anybody who can drive like real user actions that advertisers or builders or developers want can actually share in that revenue.
1:33:53
And again the the the blockchain just makes it much easier.
1:33:55
I mean, there's there's this wonky thing called multi-touch attribution, which is always the dream of every web two marketer.
1:34:00
What that means is if someone converts to Pavont and by the way, John, you're going to fly right into an ad after this, I'm sure.
1:34:05
On because they see because they see TBPN, but then they go Google it on Google, guess who gets all the credit? Not you, John. Google, right?
1:34:13
And it to it totally sucks because attribution is broken.
1:34:15
In the crypto world, it could be different.
1:34:16
You can make a cut of that, too. >> Yeah. Yeah.
1:34:19
And you see this online where like let's just say like a new protein bar is going viral and it takes a while for a health influencer to actually set up a deal and set up an affiliate code, but like you could potentially do that on the base app like like at on in an ad hoc way so that as soon as you repost you're you're you're basically tied into the affiliate program. >> Yeah.
1:34:41
There's something there's something amazing too about um there's something amazing.
1:34:45
something amazing. every time uh we get our Elon bucks I'm like wow I'm getting paid to post I was doing that for free and it's a weird thing but at the same time like these apps have no value if the users don't post right like we every user is like coming together to create
1:35:01
value whether you're just providing attention or you're providing content and um yeah most of the value kind of will acrue to the people that are creating the content and that's and that's probably right but uh but yeah just kind of taking another look at these systems is is worth doing. >> And that's how it works on Bass App, by
1:35:18
>> And that's how it works on Bass App, by the way, to plug the app side of it rather than the outside.
1:35:21
Users get users get paid to post.
1:35:23
If you're like a mega poster on base, you're going to see your wallet balance just go up.
1:35:26
And again, it all just happens effortlessly.
1:35:28
You're not setting up a Stripe account.
1:35:30
You just like check your balance and oh, there's like 100 bucks there and it wasn't there yesterday, right?
1:35:34
Like that's >> Yeah, that's the other Yeah, that's the other like you know do you know uh making the the feedback loop faster where you know Elon bucks or what what they hit every every couple weeks >> um you never really know but if you can effectively like stream like getting closer to kind of like streaming payments based on how much attention you're capturing that's pretty cool.
1:35:55
Can you bridge the uh outdoor outdoor voices kind of case study that we just ran through to uh to like how how crypto companies are thinking about CAC to LTV.
1:36:10
Uh and it doesn't need to be outdoor voices specifically.
1:36:12
There's just a you know we all live through the DTOC boom and the Facebook ad boom where you know 99% of VC dollars are going to Facebook ads that whole meme.
1:36:21
And there was this time where companies didn't understand their LTV from really when they crazy scaled their advertising budgets on digital platforms.
1:36:29
But pretty quickly DTOC companies at least figured out the analytics.
1:36:34
What does that look like today in crypto?
1:36:36
Are people thinking about it more?
1:36:39
Is there are there still like case studies of crypto companies that like raise too much VC?
1:36:42
Because I feel like a lot of crypto companies, they do a lot of incentives to onboard users, but they do it with like their own tokens that they created.
1:36:50
So there's maybe less bankruptcy risk, but walk me through how that works in in web 3. >> Yeah. Yeah. Yeah.
1:36:56
I mean, so crypto marketers have gotten a lot smarter about LTV and CAC since 2022.
1:37:00
Most of our advertisers actually understand that very well. >> Sure.
1:37:05
>> Um and I think one of the one of the innovations that crypto introduces is the business model.
1:37:08
So the the DTOC thing that you're talking about >> typically operate in a world in which you you paid what's called CPA or CPI, cost per action or cost per install.
1:37:17
In other words, you paid Google or Facebook literally if the user installed the app.
1:37:22
You you didn't pay if the ad showed up in front of them.
1:37:23
That means that the ad system actually takes on some of the risk of the conversion and advertisers love it, right?
1:37:28
But you're still juggling the LTV, what the user is actually worth versus what it costs to acquire.
1:37:33
One of the business models we offer that you're not going to find on Web 2, and again, we're trying to build a better ad system than Facebook is what we call CPV.
1:37:39
It's ad tag, so it's got to be a three-letter acronym.
1:37:41
And what CPV basically means is cost per value.
1:37:43
And what what that means is um you can actually literally pay like your bid is not like I want to pay $30 for a user.
1:37:51
It's I want to engage in a 40% rev share with the publisher on the lifetime value of the user.
1:37:56
And that's like a baked in two and a half or whatever it is ROI.
1:38:00
And and that again you can't you can't do that in Facebook.
1:38:01
I I can't tell Facebook, look, bro, Zuck, I'll I'll split I'll go have these with you on the LTV going forward as long as you acquire me a user and they retain.
1:38:09
There there's no way to do that right in the web world.
1:38:12
And you can do that actually inside BAS apps very naturally because again you have you've got a consensus truth like I know how much that user is worth.
1:38:19
I know how much they stuck around and by the way I have payment >> real because it's because you can see it on chain. Right. >> Exactly.
1:38:26
There's mutual transparency on both sides.
1:38:28
By the way, the publishers can look down for >> Yeah.
1:38:30
Because Zuck doesn't want to Zuck doesn't want to do that deal because he's like, "Well, I'm going to have to like get I'm going to have to keep hitting you up every month for years to figure out what you know if the user stayed subscribed or whatever and it's just too much of a hassle."
1:38:41
So, you do the work of figuring out the cactile TV kind of analysis and if it's not a good deal, just stop running ads basically or or figure out different creative and and make it work. >> Exactly.
1:38:53
Well, in this world, right, the publisher, the new base app can actually look downfunnel via the blockchain and say, "Okay, I I I know how many users I showed ads to and I know how much value I drove, right?
1:39:03
So, you can't you can't, you know, stiff me.
1:39:05
You've got to give me what you said you would, which is 30% or whatever it is of the LTV."
1:39:08
And that's again that's the advantage of having a consensus sort of reality layer and there's no way to do this in web 2.
1:39:14
It's basically >> right now it feels like online advertising is dominated by search engine ads whether that's in Amazon or Google and then social media ads on X and uh the meta platforms and Tik Tok.
1:39:29
Uh what does the ad ecosystem look like in the future?
1:39:32
How big is the crypto ad slice versus the AI LLM ad slice?
1:39:35
How do you see these two things playing out?
1:39:40
Because I I I feel like there's bull cases for both of those and but there's only so many ad dollars flowing around, right?
1:39:47
Even though it's like a trillion or several.
1:39:49
>> Well, in this case, I mean, the exciting thing about this new product is there's a lot of advertisers that can't advertise elsewhere that can now get >> sure >> AGM's ad product and and have these kind of um splits.
1:40:01
But yeah, we we were quoting um >> Yeah, we read your post yesterday.
1:40:05
I would bet my entire net worth that one, we will have ads and AI.
1:40:08
Two, if they will take the if they take the form of highly relevant offers, users will welcome them.
1:40:14
They'll have sky-high conversion rate and web shopping will die.
1:40:17
And three, they will be necessary to pay for the computers. Love the take. Love the parlay. Uh forgot to like it.
1:40:24
I'll leave a like on it now even though a thousand people have already liked it.
1:40:26
Um but but yeah, give me give me your take on the evolution of the ad ecosystem more broadly. >> Yeah. Yeah.
1:40:33
So to be clear, I don't think it's a zero sum between crypto and AI.
1:40:36
I think it's different sets of budgets and at some point they're going to intersect.
1:40:39
I mean people are going to start using crypto inside I mean for microp payments and AI agents, crypto is the natural way to do it and Coinbase is building stuff there as well.
1:40:45
I mean just to switch the topic a little bit to the AI side.
1:40:48
I mean I'm obviously an AI mega user.
1:40:50
I barely use Google anymore.
1:40:52
I probably fire up chatbt over 100 times a day. Yep.
1:40:54
>> And it's what's interesting is that again ads dollars actually aren't limited.
1:41:00
Human attention is limited because we only live so many hours a day.
1:41:01
But the amount of intent I can actually express inside an AI is fascinating.
1:41:05
I think >> so much about what we do is so skumorphic to like the pre- internet world.
1:41:11
I don't have to remember the before times.
1:41:12
Like why do I go to an e-commerce site and it's organized by type of thing?
1:41:16
Like when I go to Instacart, why is it organized by food type?
1:41:19
I want to make bofin or whatever it is. Here's the seven things.
1:41:21
Just buy me those seven things.
1:41:22
I don't want to go to seven aisles.
1:41:23
And yet somehow the internet websites or even PDFs still reflect like the pre- internet world.
1:41:27
I should just be able to tell the AI, get me the recipe for this complicated thing I want to make and then just like go buy the things or I want to travel to Paris and not pay more than $3,000 and stay in this place and I should see a big green go button and I just buy and transact and it's done, right?
1:41:41
The business of having to go to three different websites.
1:41:44
It's just I mean Kayak I I like it, but it drives me crazy.
1:41:46
The fact that we're clicking little radio buttons and drop downs, it's ridiculous.
1:41:50
We've been doing that for 15 years.
1:41:50
I don't understand how we haven't proceeded beyond that UI.
1:41:52
And I think AI is going to be the absolute gate to that to that intent and that desire.
1:41:56
And what's going to be interesting is e-commerce sites that are, you know, typically buying ads are going to have to surface offers.
1:42:04
And what does an AI ad mean, right?
1:42:06
There's nothing to see >> strictly speaking.
1:42:08
But when the AI scrapes your website and sees the product metadata and sees a price, you may want to service given the personal like, oh, this user hasn't bought from me before. Here's a 10% off coupon.
1:42:16
And that shows up in the results that AI returns, which by the way way defays the expensive AI cost to go to. 3 in my tweet, right?
1:42:23
And so you're going to have to have some way to actually channel intent and actually get paid for that intent.
1:42:28
Um, and there's also issues around I mean in cloud. >> Yeah.
1:42:32
>> Yeah. what what are the what are the kind of like potential traps or you know the people the Mark Cuban's I think primary concern is that that the the uh the the results are just going to be they'll be like profit maximizing conversations and our our take was like
1:42:47
this has been solved in search it's been solved in the influencer space like ads should be disclosed >> uh and if you just have disclosure it's probably uh fine but um but I think there's >> yeah I find it funny that a guy who fortune selling ads suddenly finds ads ads problematic. I'll just say that off
1:43:04
I'll just say that off off the bat.
1:43:05
Um I >> Yeah, he's like ads he he said something and we're we're going to have him on the show to talk about it, but he said something to the effect of like ads are about manipulation and it's like okay well do you want to ban them all then?
1:43:18
Should we just have >> So I I I take the under on that and in fact I think people actually like Instagram ads for example as a form of discovery.
1:43:25
I think the clickthrough rates, the the revealed preferences of how users actually react, the click-through rate is probably higher on some of that paid content than is on the organic content. Yeah.
1:43:33
And I think users, if it's a good ad, and if it's a crappy ad, I agree it sucks.
1:43:36
But if it's a good ad and gives you a discount and helps you discover a product you didn't discover before, users will like it in the only way that matters, which is what is their engagement and conversion rate.
1:43:45
>> And dozens of founders have tried the it's it's this but paid, no ads, and it rarely works if ever. >> Yeah.
1:43:50
>> Yeah. I mean, I think I think the interesting thing is, you know, like it it makes like I like paying to not have ads on X because the ads were never that relevant to me and it's nice and I use the app a lot so it's nice but the fact
1:44:04
that Meta doesn't have an adree at scale adfree product says a lot when >> people use that app I'm sure more on a I don't know more more on average than >> the only people that would upgrade would be people whose uh whose arpoo is higher than the price. I feel like it's like if
1:44:20
I feel like it's like if they're getting if they're getting $1,000 worth of ads against me, they're going to be like, "It's $20 a month."
1:44:28
And I'm going to be like, "Yeah, I'll buy that every day." And that's bad for them.
1:44:30
>> And then the people who are like, "Oh, I'm only driving, you know, $50 of ARPO a year."
1:44:34
Well, they're not going to upgrade.
1:44:36
And so they're like, it's going to actually wind up with lower revenue.
1:44:39
I would imagine that's the calculus. >> I don't know. >> Yeah.
1:44:42
I mean, look, ads pay for the internet, right?
1:44:43
I mean, as much as you might dislike them, it pays for a large swap of the internet.
1:44:46
And I think we've we've done the experiment for 20 years and users aren't going to pay $200 a year for for Facebook, which is roughly their aroos, right?
1:44:53
Like they're just they're just not right. >> Yep.
1:44:55
Uh last thing, I want to ring the gong for a big product launch, but I also want to get a wrist check.
1:44:58
What's on the wrist today? >> It's the great.
1:45:01
So I I posted this earlier today.
1:45:03
Um I've got the I've got to This is somewhat ironic. The War Rolex I got on.
1:45:08
I've got the Rolex Submariner that I bought with my first YC.
1:45:11
Let's go full full dig here.
1:45:11
I bought this thing with cash when I sold my first YC company.
1:45:15
And then on my other side of the wrist, a little brief product pug.
1:45:19
>> I became a Whoop American to me so I can actually see exactly how bad my how bad my sleep is.
1:45:24
Uh so yeah, that's what's >> amazing.
1:45:28
Well, congratulations on the product launch.
1:45:29
We'll ring the dog for you. >> Hit it. >> There we go. >> Good stuff.
1:45:35
Uh great great having you on finally long overdue. >> Yeah. Yeah. Yeah. Yeah. Congrats.
1:45:38
You come back on and and just chop it up with us for a little bit.
1:45:41
We could talk about so much other stuff if we love >> some other things. Yeah. >> Yeah. Uh, but congratulations.
1:45:44
Good luck with the rest of the launch. We'll talk to you soon. >> Have a good day. >> Bye.
1:45:50
>> Uh, I'll be right back, John.
1:45:50
I'll let you kick off the next one.
1:45:53
>> I will be kicking off a bezel ad.
1:45:53
If you want to get a Rolex Submariner, just like AGM had on his wrist, you can go to getbzzel. com.
1:46:00
Your bezel concierge is available now to source you any watch on the planet. Seriously, any watch.
1:46:04
And we will bring in our next guest, Ian Webster. Welcome to the stream. How are you doing, Ian? Good to meet you.
1:46:13
>> Good to meet you as well. I'm doing well. >> Thanks so much. Thanks for joining.
1:46:17
Would you mind kicking us off with an introduction on yourself, your company? What are you doing? What's the news? >> Yeah.
1:46:22
So, um the the company that we're building is called Prompu.
1:46:25
Um we are building security tools for Genai applications.
1:46:31
So, if you have a Gen AI application, you probably want to secure it, make sure it behaves well um in in in production.
1:46:37
And that's that's what we're doing.
1:46:40
Um the big news today I guess is that we have raised a series A. >> Congratulations. Give us the numbers.
1:46:48
>> Yeah, we we raised 18. 4. Um >> congratulations. >> Yeah.
1:46:56
>> Uh who did you raise it from? >> Insight Partners.
1:46:58
Uh and A6 also participated. >> Yeah.
1:47:01
Two kind of smaller smaller firms, you know, they really had to scrap the money together to get this deal done. >> Yeah.
1:47:06
I mean, um, I'm I'm very lucky. >> Yeah, it's fantastic.
1:47:10
Well, congratulations on the round.
1:47:11
Uh, talk to me more about security in generative AI.
1:47:16
That could mean so many things.
1:47:18
We've talked about LLM psychosis, someone, prompts deep.
1:47:21
We've talked about leaking user data from one instance to another.
1:47:26
Uh, somebody like hallucinations and just kind of like not not performing.
1:47:31
I'm trying to, you know, use an LLM to convert raw text to JSON and it's just kind of messing up.
1:47:36
Like what are the key areas where you would define a security that you're going after solving or are you trying to do everything all at once?
1:47:45
>> So there are kind of two sides to the coin.
1:47:47
Um the first is uh the foundation model security or safety.
1:47:51
Um >> basically stuff like is the LM going to say something racist that reflects badly on on my company?
1:47:58
Like is it going to drive people into a psychosis?
1:48:00
like all that type of stuff, >> brand risk basically. >> Yeah.
1:48:04
But I I think the this the part that's actually more interesting is at the application layer.
1:48:09
So um you know the like like once you put a model into the hands of a developer, they put an application on top of it, they connect it with a database and PII, they connect it with an with an API and now you've got a rag, you've got an agent, etc.
1:48:24
>> Um that's when there are many ways to shoot yourself in the foot.
1:48:26
shoot yourself in the foot. Um so the the biggest concerns that we see um I mean there there's definitely an aspect on on that uh foundation model side but the the big problem that scares a lot of large companies that are working with
1:48:40
LMS are things like PII leaks tool misuse through agents um and also just thing softer issues like like LM recommending competitors or um sure you know like giving you medical advice when >> they're just supposed to be like an e-commerce chatbot, stuff like that. >> Financial advice, too, would be a risk.
1:48:59
>> Financial advice, too, would be a risk. >> Yeah.
1:49:00
I mean, we were talking about Amazon's uh Amazon has like a chatbot that you can ask about the product that you're searching for or whatever, but then people were getting it to write like React modules and getting it to write JavaScript for them.
1:49:11
And it's like that's not a big deal, but it's just like clearly not the intended purpose of that particular chatbot.
1:49:16
What are what are the key what are the key reasons why you think over the long term companies will want to outsource a lot of this function versus build their own kind of systems to protect against some of these edge cases.
1:49:29
I'm sure there's a bunch of good reasons otherwise you wouldn't be building this and raising all this money. >> Yeah.
1:49:35
So I mean it's it's it's pretty hard to do.
1:49:38
Um so like I I encountered this problem firsthand because I was leading um Genai engineering and and and products at Discord.
1:49:46
we were launching these these apps to hundreds of millions of people and like to to do this right you really need to be able to train models that can behave adversarily because your GPT or your claude is not really going to be a good redte teamer or attacker off the shelf um and uh just kind of test uh extremely extensively like over tens or even hundreds of thousands of test cases to try to find all those rough edges.
1:50:13
Like the the thing about AI is that the attack surface is all of human language and and then some.
1:50:20
So it's just it's a really difficult problem >> and the way that many like large enterprises are are dealing with this right now is um they are doing it manually and just covering a very small fraction of of the risk areas.
1:50:34
Um >> so you know that's why we think that there's big opportunity in the space. >> Yeah.
1:50:41
>> Yeah. And if and if somebody engages with prompt fu, I'm assuming you start running like a do you start automatically like trying to jailbreak the LLMs getting like are you kind of running these sort of like repetitive uh
1:50:54
attempts in order to like find problems so that you can stop them or or is the product itself kind of uh you you have it things figured out enough that you can sort of that you let them kind of like and their communities try to jailbreak it and then react to it. Yeah. So the the way to think about it Yeah.
1:51:11
So the the way to think about it is that we've trained models that behave as adversarial users like misbehaving users in your application.
1:51:19
Um and we we have we have agents built on top of these uncensored models that um target specific risk areas like PI leaks or toxicity or like talk about competitors or you know what whatever 70 plus different areas.
1:51:34
Um, so the the way that we work is we have these attackers generate use cases um or like attack objectives and then um we feed those objectives into a bunch of different uh what I would describe as like searcher optimization techniques to kind of poke and prod um the entire attack surface of of the application.
1:51:57
So what what this amounts to is we wind up having um you know thousands and thousands of conversations where like the attacker is just trying to wiggle its way around some of the guard rails or some of the safeguards that are put in place.
1:52:11
>> Is there a big uh difference right now in the perception of security or ability to deliver on a secure application whether you're using a closed source or open- source model.
1:52:21
open- source model. I can imagine like if I'm like oh I'm using an open source model I have the ability to fine-tune it maybe I feel more confident or maybe hey I'm using a closed source model I'm paying a fortune to open athropic like they're going to do more work to secure this thing than I am like what's the
1:52:38
perception around like which which paradigm leads to better security >> I would say right now the perception at the at the top of market at least like we we work with some of the world's largest companies Um, >> there's definitely a strong preference for closed source right now, but I don't really think it's the security that's driving it necessarily. I think it's
1:53:00
I think it's more just >> performance per dollar and or just overall intelligence, even if it's more expensive.
1:53:06
It's just like I want the best. >> Yeah. >> Um, I would say so.
1:53:08
Um I think like it the the the the thing is for for security even on the closed source side um the the incentives really differ based on whether you're open AI or a company building on top of open AI.
1:53:22
Like OpenAI has has plenty of geniuses who are who are working on um you know making sure that it doesn't say racist things or whatever.
1:53:31
But like no one's going to stop you from shooting yourself in the foot by by hooking it up incorrectly to to like a database of PII. Yep.
1:53:41
>> So that's >> yeah or open AI open AAI is like you know they're not releas giving you this API and telling you we're making sure that it will never the model will never recommend a competitor like you have to kind of build a layer on top of that.
1:53:54
>> Has there been a big case study of like LLM security going gone wrong that people like to point to in the industry yet?
1:54:01
Like we were just talking about the T app getting hacked and that did not have any generative AI features that I'm aware of.
1:54:08
It was >> potentially just made with a lot of >> maybe it was vibe coded, but it seemed like it was just a misconfigured Google Firebase bucket.
1:54:14
They just changed the the access rights on an old database that they were using and it seems like something that could have happened 10 years ago. It happened today.
1:54:23
It's not a uniquely new phenomenon, but I'm wondering if there's any case studies that you point to is to to your clients like, hey, the value of working with us is that we can avoid that happening.
1:54:36
Yeah, I mean there there have been a handful of things that that have been pretty public.
1:54:38
I think it it really depends on like the industry and also the geography.
1:54:42
Um there was there was a mail carrier in in the UK that that like had had a lot of issues with with a chatbot that basically had no guardrails.
1:54:52
I think there was a case where um a a Canadian airline, I think Air Canada, had a chatbot that that like committed to a refund that was out of policy. >> That's right. I remember that one. Yep.
1:55:04
And then of course you have Sydney and Bing and Microsoft Tay and there's there's >> the fact that the like biggest companies and labs are struggling with this at that scale with that amount of resources.
1:55:15
Can you imagine some random company with that rolling this out to to users?
1:55:20
They just don't have the the ability to stress test these things the same degree. >> Yeah.
1:55:26
So my my not so hot take on this is that um like this this is really the blocker to for for big companies to use Gen AI.
1:55:33
Um >> like if if AI doesn't live up to the hype, it's because companies are are too scared to go public with their with their AI because of incidents like this.
1:55:44
um like we we encounter like Fortune50 companies that have hundreds of internal AI use cases but just haven't pulled the trigger on on like pushing these prototypes out there really because of this issue.
1:55:58
So that's why I think this is like a a big bottleneck in this in this wave of AI right now. >> Yeah. Yeah.
1:56:03
It's way better to just take all of your employees, give them chat GPT Pro and say, "Hey, you're still responsible for everything that you all your work product, but you can use these LLMs as a tool to speed up your workflow, but ultimately, you know, even if Cloud Code is writing a lot of the software, you got to review the death and you got to review the email that gets sent to the important client.
1:56:26
So that, you know, we go from 99% accuracy, 1% hallucination to zero. That's your job now."
1:56:33
and you're going to do you're going to hopefully be faster, hopefully be better, but there's still a human in the loop in these things. Fascinating.
1:56:39
Uh well, thank you so much for joining.
1:56:40
Uh congratulations on the new round. Uh great to see it.
1:56:43
Thank you so much for stopping by. We'll talk to you soon. >> Appreciate it. Take care. >> Have a great one. Bye.
1:56:47
Up next, we have >> uh a someone who proved the uh oh, cinematic viral videos don't work anymore. Yes, they do. They still got it. He beat the odds.
1:56:59
Uh very excited to have Dr.
1:57:03
Let's bring him >> coming in from off Deal. >> How you doing? >> What's going on?
1:57:08
>> Hey J, how's it going? Good to see you both. >> It's going great. Great to see you.
1:57:13
>> Enjoyed the Financial Times article.
1:57:13
We love the Financial Times. >> We got the delivered.
1:57:16
We get it delivered in print every day.
1:57:18
We have, you know, tons of respect for the Wall Street Journal and the Financial Times.
1:57:23
Uh it may make sense given your uh given your company and your target market.
1:57:25
Uh but talk to us about the company and the target market. What are you building?
1:57:31
Uh yeah, so we're building an AI investment bank.
1:57:33
To my knowledge, it's the first one of its kind.
1:57:37
>> Probably the easiest way to think about it is uh think of what what would Goldman Sachs look like if it was started in 2025 and not like 1865 or something like that. >> Yeah. >> Right.
1:57:47
So, uh, we have our own software, our own data, um, our own infrastructure, our own bankers, um, and we just ripped sells for small businesses that are ignored by traditional banks. >> Let's go. >> That's awesome. Yeah.
1:58:01
When we we met, uh, just over a phone call a couple months back and, uh, my first question was, why not, you know, build the software and just make it available, the Harvey approach. >> Yeah.
1:58:11
The the sort of more uh, SAS approach.
1:58:14
And I think it'd be awesome if if you kind of broke that down for for our audience today. >> Yeah.
1:58:20
And I'd be lying to you if I said that we didn't consider uh building software for banks.
1:58:24
And and I do think there's some great products out there like Rogo and some others that that are doing good work.
1:58:31
>> Uh but when we were thinking about this uh from first principles, we saw how bureaucratic large banks are and how long it takes to enact any meaningful change.
1:58:41
change. Uh I personally spoke to you know dozens dozens of bankers when I was asking them what is the most requested sort of like problem that you want us to solve they would say uh call transcription and I would say well zoom exists and they said well we using Cisco Webex and I would say well Cisco Webex
1:58:59
has zoom trans or call transcription and it would say oh we don't have that allowed and that you know I'm being a little bit facicious but that is verbatim what the number one requested feature was >> and I was just kept thinking to myself if I was running my own bank I would completely different. >> And uh we kept kind of noodling on this,
1:59:13
>> And uh we kept kind of noodling on this, noodling on this and then we uh eventually came across an article from Sarah Tel from Benchmark that famously said >> um sell work not SAS.
1:59:22
And uh after some uh deliberation, my co-founder and I decided to go all in on a crazy idea of being your building our own investment bank from scratch.
1:59:31
And so here we are um you know just over a year uh since we started and uh you know closing all kinds of deals. >> Yeah.
1:59:40
break down what what your initial deals have looked like, what what kind of businesses you you're targeting.
1:59:43
I know it's generally companies that are valuable but but not big enough kind of ticket sizes to get uh the traditional investment banks on board. >> Yeah.
1:59:55
So even though we say that we're what Goldman Sachs would have looked like in 2025, we're not actually trying to compete with Goldman Sachs.
2:00:02
we're we're much more interested in going after uh the market that is currently unserved by uh legacy incumbents because the unit economics don't make sense.
2:00:11
So there's millions and millions of small businesses in America.
2:00:14
Uh and hundreds of thousands of them change hands every year.
2:00:18
Uh for vast majority of them, they end up doing it on their own.
2:00:21
Uh like less than one in five small businesses under $10 million as you duh actually have a self-ight advisor, which is kind of crazy if you think about it, right?
2:00:30
It's the most important sale of their life.
2:00:32
It's an asset that's worth over 80% of their net worth.
2:00:34
And you know, these business owners, plumbers, roofers, etc.
2:00:37
, they're just doing it on their own.
2:00:40
>> I've personally experienced this actually all the way back in 2019 when I tried buying a small business.
2:00:45
>> Uh, and it was a complete disaster.
2:00:45
I was talking to brokers or I was talking to owner operators who didn't really know how deals work.
2:00:51
It was really hard to get a deal done.
2:00:53
Um so when we saw you know this opportunity we got really excited because AI allows us to have vastly better unit economics and therefore go after the segment that's currently uneconomical for the hula hookis or their lazards or ever corores of the world. >> Yeah.
2:01:11
So when I think of Goldman Sachs, I think this monster bank that's not just an investment bank.
2:01:15
They also have wealth management and you know not just sellside advisory but you know uh you know uh like sellside coverage and they're doing you know buy sell hold reports and then they're also doing IPO stuff.
2:01:29
Um but we're talking here specifically narrow it down to M&A then take it down into the mid-market and then take it down even further.
2:01:35
And there's been companies that have done kind of like the it's a buy and sell a company platform at the super subscale.
2:01:44
You can I mean Justin Khan famously like sold a calendar app on eBay.
2:01:47
He sold a company on eBay for like a I think a $100,000 a couple hundred thousand.
2:01:52
Um, but walk me through the actual uh the actual approach because when I think of a traditional M&A advisory firm, it's it's find the companies, build the relationships, then build the materials, then value the company, then be kind of a back channel while everyone's talking to actually get the deal done.
2:02:10
A lot of that feels automatable, but what what in the flow do you feel like is the lowest hanging fruit?
2:02:18
How does it all piece together?
2:02:21
Yeah, maybe it's helpful to think through a journey of a off deal customer, please.
2:02:25
Um, when we think about this, so I'll speak about a deal that we recently closed.
2:02:29
It was a, uh, a monastery school in in Arizona.
2:02:32
Um, they actually tried selling it on their own, uh, but the private equity firm told them um, the business was too small for them from a cash flow perspective.
2:02:41
Uh, so they somehow found us on Google.
2:02:43
We have this cool AI tool that people can check out that allows you to put in your company website and we preview some of the potential buyers for your business because we built a proprietary data layer over millions and millions of businesses.
2:02:57
We have our own recommendation engine so we can actually pinpoint which companies in the United States could be potential buyers.
2:03:02
So they got in touch with us and we decided to take them on.
2:03:07
Um what we did is our banker uh used our software to identify every single monastery school in the United States, find their contact information, who the owner is, etc.
2:03:17
Then we reached out to them.
2:03:19
There's about,00 of them that we've talked to.
2:03:21
Um 72 of them signed NDA.
2:03:24
Um you know, that means all of those people got to check out the SIM that we put together, the financials, etc.
2:03:31
Uh about 20 of them actually met with management um in person for meetings.
2:03:36
and then uh uh four submitted final offers and then the the offer that got it over the line was 40% higher than the first one that came in and it was an old cash deal. >> Congratulations. That's amazing.
2:03:49
And the cool thing about this is that um even though most of what we do does involve private equity buyers or private equity back strategics, in this case, we were able to bring a subscale asset to market and activate companies into being buyers when they weren't actively looking for a business to acquire, right?
2:04:06
Because the investment banker on our team just reached out to Monastery Schools and they said, "Hey, are you interested in expanding expanding your footprint?
2:04:13
We're actually working with a school that looks a lot like yours, etc."
2:04:17
These kinds of things are only possible with AI because the the the bio universe is just so so fragmented. Yeah. >> Right.
2:04:23
And so we find our own software for each part of the workflow. >> Sure.
2:04:28
>> How do you avoid the uh I I feel like VCs get these emails a lot where they'll get an inbound email from uh an investment bank and it's like, "Hey, I have somebody interested in buying your >> a ton in e-commerce."
2:04:39
When all this e-commerce rollups is happening, it doesn't matter how much money you'd raise.
2:04:44
because they'd be like, "We'd love to buy your business for 200k."
2:04:46
>> And I think that's a that's a challenge for you guys.
2:04:48
You eventually want to get to the point where you're selling thousands of companies a year.
2:04:51
You want to make sure you're not sending millions of emails that people are looking at being like, "Okay, this is clearly not nobody a person did not approve sending this email.
2:05:01
It was part of a kind of an automated system." >> Yeah.
2:05:04
So, uh this is a literally a daily point of discussion like what does this look like when we're doing hundreds and hundreds of concurrent deals?
2:05:11
Uh the the nice thing is that we're not spraying and praying indiscriminately, right?
2:05:15
We we we actually pinpoint uh buyers and we're very thoughtful about who we reach out to.
2:05:20
So today the volume is not as high as you might think uh from a pure outreach perspective.
2:05:25
But the nice thing is that there's a network effect here, right?
2:05:28
Because every time we interact with a buyer, they join our platform and next time they just get an instant notification when the deal matching our investment mandate shows up.
2:05:35
And and actually right now we have about 30 businesses for sale, right?
2:05:38
ranging from, you know, million to almost eight billion dollars in Utah.
2:05:42
And there is dozens of private equity firms that are participating in multiple of these sell sites, right?
2:05:49
Um it might be different teams, right?
2:05:51
You might have like an industrial team versus a, you know, a healthcare team, but but these firms get onboarded onto the platform and then from then on they're repeat customers because they're going to be accessing deals all day every day. That's their job, right?
2:06:04
So that's kind of the long-term play here is to build this uh you know ecosystem of of buyers and we track by the way all buyer behavior right so we know who's bidding on what how much are they submitting how do they like rank to structure deals etc.
2:06:16
So there's a lot of firstparty data we're collecting.
2:06:19
It's going to be tremendously helpful moving forward. >> Very cool. >> What's the goal?
2:06:23
What's the goal for the next 12 months?
2:06:24
What's the deal the deal volume?
2:06:28
>> So the the nice thing is that the messaging has been resonating really well.
2:06:32
Turns out that, you know, an average business owner in America has already heard from like a 100 private equity firms banging on their door because everyone's doing rollups and the LBC's even getting in for roll-ups, etc.
2:06:42
That's been actually great marketing for us because an average private equity firm will look at about a hundred opportunities before pulling the trigger on one.
2:06:50
And so it's a pretty bad user experience for for the sellers who, you know, end up engaging with one of these guys maybe twice, three times and then they get ghosted at the 11th hour.
2:06:58
So when we come in and we say, "Hey, we're able to uh flip the script and put you in front of a hundred buyers so only one of them is the lucky one to buy your business."
2:07:08
That's been really working really well for us.
2:07:10
So, in terms of our pipeline, uh like I said, we already have 30 businesses for sale today.
2:07:14
Um we have over a billion dollars worth of businesses um you know, in the near-term pipeline sort of over the next uh 9 to 12 months that we expect to launch.
2:07:23
And the ramp has been just exponential both from the inbounds, right, and the referrals we've been getting, but as well as our outbound motion has been very effective.
2:07:32
Um our internal metrics sort of for for end of 2027 is to hit $und00 million in fees uh which will require based on current level of automation about uh 15 or 20 bankers, right?
2:07:44
So each banker is able to do about 10 cell sites concurrently. >> Wow. It's amazing. >> Wow. Uh and what's the news? Raise money.
2:07:52
>> Well, the news is that uh we got 12 million bucks in funding. >> Congratulations. >> Great stuff.
2:08:01
Thank you so much for hopping on.
2:08:02
>> Awesome having you on the show.
2:08:02
I'm sure you'll be back on soon.
2:08:05
>> Yes, we got to hit them up.
2:08:05
I have this interesting idea.
2:08:07
I was thinking might blow people's minds.
2:08:09
I'm thinking about buying an HVAC business.
2:08:13
>> I don't think anyone's thought about this roll up. >> You're selling three. Okay, I'll go to you. Perfect. Might be the first one.
2:08:21
>> Your next uh I got one feature request.
2:08:23
I think you should you should add Tesla robots to the platform that can go have steak dinners with potential clients because that's the one >> we're training a model that can play golf with people right now. >> Yeah, that's key.
2:08:33
That will be the final unlock.
2:08:34
That's the final investment bank.
2:08:38
>> If you can eat a steak dinner and you can golf, >> I'd like to see AI do that. >> Goldman Sachs cooked. >> Absolutely cooked. Thank you so much. >> Awesome. >> Have a good one.
2:08:46
We'll talk >> having you on Ory. Cheers.
2:08:47
And we have our last guest of the day hopping on the stream, Eric.
2:08:52
>> Went viral uh yesterday for creating a robot that can uh that that looks like a lamp.
2:08:59
So, let's bring him in from the reream waiting room. >> There he is. >> Good to meet you. Welcome to the stream. How you doing? >> Great. Great. Thanks for having me on.
2:09:08
>> Uh break down the video. What are you building?
2:09:09
Uh how much of that is CGI?
2:09:12
How much of that is real?
2:09:12
Where are we in the techn technology roll out of this type of robotics project? >> Yeah. Yeah.
2:09:19
So, uh the video you saw probably yesterday, it was a it was rendered.
2:09:24
>> Um it's not AI generated unfortunately, but it was rendered. >> Sure.
2:09:28
>> Um so, we have a prototype working. >> Yep.
2:09:30
>> This one right here that you see in the background. >> Okay.
2:09:32
>> Um so, we we we are building robotic lamps.
2:09:35
Uh they're meant to sort of disappear into your home when you >> Oh, look at it working over there. >> Yeah, it's working. Wow. That's working away.
2:09:40
We're we're getting the laundry folding going.
2:09:44
>> Yeah, it's a super cool uh like I don't know.
2:09:47
I mean, part of why it went so viral is like people have been talking about laundry folding robots for years and then we've seen uh companies take it very seriously, but never with a form factor that could just melt into your home.
2:09:59
I don't know how protectable that is, but it just felt like from a user experience, from a from a perspective of like actually integrating this into your life, it felt much more real to me.
2:10:07
Um, how much of where did the idea come from and and what are the benefits of like building it into the bed frame basically?
2:10:16
>> Well, I mean to be clear, so it's not it's not actually built into the bed frame at all.
2:10:20
It's a single floor lamp that you can technically place anywhere.
2:10:23
>> Oh, you can put it anywhere. Okay.
2:10:25
>> So, it's like by your couch, by your table, by the bed.
2:10:27
>> Uh, it's just through most people that we spoke to, the bed happens to be where you do your laundry.
2:10:30
People just dump it there before they go to bed. >> Yeah.
2:10:33
Um, but yeah, like uh we're familiar with like, you know, some of the previous attempts at laundry folding.
2:10:39
And I think the main thing that has always held it back is the form factor.
2:10:44
>> Um, you know, like there's usually they come like sort of with these kind of like pre-installed rollers that you kind of have to feed in manually.
2:10:50
It's like semi-autonomous.
2:10:51
It only works for specific type of clothing.
2:10:55
>> Um, what we going have here, you know, mimics a little bit kind of like the human arms.
2:10:58
So we can adapt to different types of clothing sizes, uh, different types of articles and styles.
2:11:02
Um, and you know, with the recent breakthroughs in AI, we're finally able to generalize across different clothing.
2:11:11
>> How much of those recent breakthroughs in AI can you actually capitalize on?
2:11:15
Because I, you know, like GPT 4. 5 is amazing.
2:11:18
GPT4, you know, 03 Pro is amazing, but like it doesn't seem like it would translate to laundry folding just yet.
2:11:25
what exactly are you a beneficiary of and then what work do you have to do to kind of take it across the last mile? >> Yeah.
2:11:34
So I mean there's there's been a lot of tremendous work done in the vision aspect uh for robots to now really be able to understand >> uh what is seen in front of itself right and then also in the last year within robotics community there's also a lot of foundation models for uh vision language action models that are also available.
2:11:51
>> It's the BLMs that are very popular. Right. >> Exactly.
2:11:53
But yeah, yeah, but we call them VAS because there's an action component to to these models.
2:11:57
So they're they're they're doing pretty well.
2:11:59
They can generalize the clothing items that, you know, they haven't seen.
2:12:01
know, they haven't seen. um they're able to plan in sort of like the joint angle space of these robotic arms uh to to be able to fold them um into neat piles uh you know whatever the >> So do you have to build a uh I I imagine
2:12:17
that like like recognizing that I'm looking at a t-shirt is somewhat commoditized like you can kind of get that out of the box for free with some of the foundation models but then there probably no off-the-shelf model that understands exactly how your particular set of joints work. So is that where
2:12:30
So is that where you're doing fine-tuning or post- training?
2:12:34
Like how exactly are you translating into the specific robotic machinery that you've cho chosen to assemble in a particular way? >> Yeah. Yeah.
2:12:43
So I mean it comes down to like I think you touched up on there uh it comes down to the data collection uh process. Okay.
2:12:50
>> So we right now we manually telly operate these arms that you see behind me >> sense collect the data >> and yeah we fold the laundry ourselves. Yeah.
2:12:56
Depending on what you want to fold it can be pretty fast.
2:12:58
Um then we post train um fine-tune fine-tune the the existing models essentially.
2:13:05
>> Teleoperation is it someone using like gloves basically?
2:13:07
How are they actually doing the teleoperated folding?
2:13:09
I imagine it' be very hard to do it on an Xbox controller.
2:13:13
>> There is two two methods there.
2:13:16
>> Can we go full screen on this video? Yeah. >> Hold up.
2:13:19
>> Oh, you got some gloves on. Oh god. Whoa.
2:13:22
>> This is Ready Player One stuff. I love it. >> Okay. Oh wow. >> That's awesome. Oh yeah. Yeah. VR. Yeah.
2:13:29
with the >> yeah you you hold it yourself and you sort of do the task. >> Okay.
2:13:33
And then the inverse kinematics happens within the robot to decide where how it actually gets to that particular place in the 3D space. Yeah, that makes sense. >> That's cool.
2:13:41
How do you how do you you know where where do you expect obviously your number one competition is just people doing their own laundry.
2:13:46
I think it's smart to build a robot that has some intrinsic value just like constantly, you know, just even a a reading lamp, however, however somebody would look at it.
2:13:57
But uh are you expecting to have most of your competition coming from humanoids?
2:14:00
How do you think about that category broadly?
2:14:05
Because obviously a big pro a lot of people like the first demo they'll or the the promise you know a humanoid robot manufacturer might make is that oh it's going to do your laundry.
2:14:14
>> First promise is dancing. It's always dancing.
2:14:16
But then yes the second promise is laundry. >> That's right. >> Yeah. Yeah.
2:14:21
I mean, I think human noise are, you know, they're they're going to be there and I'm sure that there's going to be a, you know, market segment of the market that that would welcome those things into their homes.
2:14:28
I think uh where we are sort of the differentiator is that we are really sort of going after uh people that simply just don't want to share their living space with like robotic humans.
2:14:42
>> Um and they they want to be in full control of their their own place.
2:14:43
And what that means is that you know in your home you have furniture, you have appliances, right?
2:14:48
and they sort of serve a single purpose.
2:14:49
They have a place that they belong.
2:14:51
They don't do anything when when you don't want them to and they only activate when you want them to.
2:14:57
So, you're you're very much sort of in control of your space.
2:14:59
And that's that's sort of where we come in >> um to try to also I mean just the cost standpoint if you were going to buy Yeah.
2:15:06
If you're going to buy a humanoid and let's say it costs $20,000 and you really wanted to do laundry and then you can come out with an option that's like >> it feels like it's a fifth of the cost because it's a fifth as many machine like parts, right? >> It's a lot. >> Yeah. Yeah.
2:15:18
It's got to be a lot less.
2:15:19
So, I I would imagine that you're, you know, closer to Roomba territory than Tesla Optimus territory.
2:15:23
And that just pulls you forward a couple.
2:15:26
>> Final final boss for robotics folding uh socks feels feels pretty difficult.
2:15:32
Maybe you could like line them up and sit them down on top, but you need to kind of human's last job will just be like, you know, folding them. >> Maybe. Maybe.
2:15:41
>> Have you have you cracked that problem yet?
2:15:44
>> Uh we're working on it.
2:15:44
Uh I think we can fold most articles of clothing right now.
2:15:47
The the main challenge we're going on right now is anything that's inverted.
2:15:50
Uh that might require you to sort of like push the stock out the other way. >> Yeah, that's hard.
2:15:55
>> That's a bit more challenging without fingers. So, >> yeah.
2:15:58
What's the What's the price target?
2:15:59
You you got $50 pre-orders right now, fully refundable, but I imagine that the final price will be some multiple of that. >> Yeah. Yeah.
2:16:08
So, right now we're looking at trying to get the price to under uh $2,000. >> Wow. >> Uh per per unit.
2:16:14
>> Yeah, that'd be definitely doable. Would you need two?
2:16:16
I know in the demo video there were two arms working together.
2:16:20
That feels like folding laundry with one hand would be uh kind of an IMO level challenge for me at least.
2:16:24
Uh do you need two of these things? >> Yeah. Yeah.
2:16:29
I mean it's doable with one, but you're just going to have to wait a little bit longer.
2:16:32
So if you're watching a movie or something like that, hopefully it's done by done the movie.
2:16:37
>> Uh but I think the ideal setup is is probably to have two.
2:16:40
>> Um >> but again, it depends, you know, whether your space can afford it, you know.
2:16:43
Um if if it's the couch, if it's a single chair, if a smaller table, you might only have room uh for one, right?
2:16:48
So um >> is there a world where you sell the robot with a teleoperation package?
2:16:54
So there's someone somewhere else who's folding my laundry remotely and I still don't have to fold my laundry and I pay $2,000 for the install and then $10 an hour for someone to tell operate it or something like that. >> Yeah.
2:17:12
Well, I think I mean our goal I think is to really try to get to ship with autonomous voting out the box because one thing that we're really cautious about is like the privacy aspect. Sure.
2:17:21
>> Um we we are trying to avoid having to have people >> that you don't know basically now be given a set of arms and a pair of eyes into your into your private do some like blurring of the >> I think I I I don't see consumers being down to to know >> you could say that about whales though.
2:17:39
Those are teleoper operated and they have cameras inside like but But it's also like a Google product and Aaron Aaron's smart but not not the same level of brand but good luck. >> Yeah.
2:17:50
I guess like how how are you thinking of mitigating people's concerns of like putting robot arms right above their beds and like the you know sort of like black mirror scenario where the robots start.
2:17:59
We saw a lot of that in the comments a lot of that in the quote tweets.
2:18:03
You know it happens when you set the timeline on fire. >> Yeah. Yeah.
2:18:08
And I mean I think I think it's key that for these arms that you know when they're not activated they have to be and function like a regular lamp.
2:18:15
So that that means like a hardware switch that turns them off.
2:18:16
The lamp hoods naturally sort of covers up the grippers and the cameras.
2:18:20
So they act as sort of a mechanical blockage already as is.
2:18:26
>> Um and you know I think you know do when when we do things like that and then we communicate clearly uh to our customers on how to actually operate these things.
2:18:35
Um that's when we can really set the expectations pretty clear.
2:18:37
Um it shouldn't work when there's people on the bed.
2:18:41
Um they should only f laundry when there's laundry on the bed.
2:18:45
>> Um and yeah, >> it's very Yeah.
2:18:47
The other thing I I Yeah, be curious.
2:18:49
It's very possible that people don't actually need to put these by their beds in the long run.
2:18:54
You could put it somewhere else in the room as well if you don't want to have the >> the laundry room. Yeah, basically. >> There you go. Lots of options. Yeah.
2:19:03
>> Well, I'm excited to see how things >> Good luck with the next phase. Congrats on the launch. We'll talk to you soon. Have a good one. >> Good stuff.
2:19:11
>> And if you're looking for top tier cleaning and 24/7 concier service, go to wander, find your happy place, book a wander with inspiring views, hotel great amenities, dreamy beds.
2:19:19
In the future, there'll be a robotic lamp that folds your laundry in every wander. Who knows? It's possible.
2:19:26
It's a vacation home, but better, folks.
2:19:28
And in the new in the world of uh robots that could potentially fold your laundry, Luke Metro has been traded from Android to skilled AI.
2:19:38
Modern AI is confined to the digital world.
2:19:41
At skilled AI, we are building AGI for the real world unconstrained by robot type or task. A single omnibody brain.
2:19:46
Today we are sharing our journey.
2:19:49
And um Luke Metro says some personnel news. I work on this now.
2:19:55
I'm super excited about the future of robot AGI. Well, no, he messed up.
2:19:57
He said some personal news.
2:19:59
This feels more like professional news or personel. >> Personnel news.
2:20:04
Yeah, this is personnel news.
2:20:06
This is not personal news. >> Yes, this is cool.
2:20:08
Uh, big vote of confidence for skilled AI.
2:20:10
I'm sure there's a hundred companies in the Gundo alone that would have wanted to pick up >> the legend Luke Metro. >> Metro.
2:20:18
Absolutely >> number one on the timeline and uh a great engineer as well.
2:20:22
We we will close it out with this post from Alexis O'hane because we got to get to New York soon.
2:20:29
Let's go to Tyler Cosgrove though.
2:20:29
He has a little bit of an update for us. Let's hear it. What?
2:20:33
>> This is about something else. >> Okay, tell me.
2:20:34
>> Um I have some unfortunate news.
2:20:36
>> What happened >> in Bloomberg?
2:20:36
Luke Farador got hit with a hit piece. >> Wait, what? >> Yeah. >> No way. Why?
2:20:42
>> Might have to cover this tomorrow.
2:20:43
>> Okay, we'll cover it tomorrow.
2:20:44
>> We're going to be in New York. We'll protest.
2:20:45
>> We'll get to the bottom of it.
2:20:45
We might have to go to Washington DC and defend our boy Luke Farador. One of the best ever. Do it.
2:20:50
A scrolls enjoyer being attacked. unfortable.
2:20:53
They the I'm I pulled up the image.
2:20:56
He looks incredibly sick.
2:20:56
So, >> it might it might be one of those things that just builds his aura. You never know.
2:21:04
>> I think I think they're they're Yeah, he's a farming business week. >> Let's see.
2:21:08
Well, we'll dig into that tomorrow.
2:21:09
But we will close out with this post about Figma because we're going to New York to hang out with the Figma crew.
2:21:14
Uh, Alexis Ohania says, "Sure, I've seated dozens of billions of dollars, billiondoll companies, but I absolutely have my embarrassing miss list.
2:21:22
Here are my Figma notes from 2016. We passed."
2:21:25
Uh, congrats Dylan Field and the entire Figma team.
2:21:27
Y'all absolutely made something people love.
2:21:29
And his notes say, "Figma, why now?"
2:21:31
WebGL co-founder was uh Elf on the first was on Elf, the first people to make WebGL demos.
2:21:38
Avary had a great team, good software, but tech wasn't there. Lack of focus.
2:21:41
I actually knew someone who worked at Avary.
2:21:42
It was kind of a a Photoshop competitor for a while and then they served a uh an SDK that you could basically bake Instagram filters into your app.
2:21:51
If you wanted to have that as a feature in your app, you could pay Aviary to to offer that service to you.
2:21:57
Um they launched two months ago.
2:21:57
Uh four out of seven users of seven days in the week.
2:22:02
How many people log on four days a week? 600 now a thousand.
2:22:04
three uh three three days per seven users or 3,000 weekly active users at 9,000 80,000 signups.
2:22:14
>> This was at a time when Sketch was completely dominant.
2:22:17
>> Yes, that's that's right.
2:22:17
So mostly word of mouth, no paid content definitely drives a lot.
2:22:21
Interested in learning better why someone signs up and invites collaborators primarily designers using this but sees engineers, PMs and marketing people who are in the orbit being absolutely correct. >> Absolutely correct.
2:22:31
A decade early but Dylan knew that in the long term engineers would be using it, PMs would be using Figma and marketing people would be using Figma. >> This was 2016. >> 2016. Uh >> it's crazy.
2:22:42
It's crazy this if if you had >> this this a a team with this competency >> Yeah.
2:22:51
>> and this kind of early traction. >> Yeah. >> Today. >> Yeah.
2:22:55
>> You'd probably get a series A done. >> 100%.
2:22:57
>> And this was a this was a seed pass. >> Crazy. Crazy.
2:23:00
So, not clear about pricing strategy yet, but excited to get started.
2:23:03
The pricing strategy is fantastic.
2:23:05
Obviously, >> they want to have people paying. >> Yeah.
2:23:07
Users say they'll pay right now. Uh 40% of users are US.
2:23:10
60% international estimates are $25 per user per team per month.
2:23:15
That's >> ended up also being I think pretty much on point. >> Yeah. Amazing.
2:23:19
Uh we will close out with one last post from Joe Weisenthal.
2:23:21
He says, "I'm still flabbergasted that people talk about pursuing their white whale in a proud manner.
2:23:26
Setting aside the risks the book may also makes it clear this that's immoral."
2:23:31
Uh well, I have a white whale.
2:23:33
My white whale it would be doing a show in person, a podcast in person >> in New York with Joe Weisenthal. >> Yeah. I think it's possible. It might destroy me.
2:23:44
>> It might destroy me, but I'm willing to risk it all to >> catch my white whale, which is a an inerson show with Joe.
2:23:51
He's been on our show multiple times.
2:23:53
I've been on his show, but we've never done it in person. Is it possible? Anything's possible. Anything's possible. We will catch the white.
2:24:00
>> Folks, we New York City.
2:24:03
>> We're going to hop on a plane.
2:24:03
And our production team is like, "Hang up. Hang up the phone.
2:24:07
>> They really want us to go off. Going off, >> you know.
2:24:09
Let's keep going a little bit longer.
2:24:11
He's actually trying to unplug all the cameras.
2:24:14
They're going to shut us off.
2:24:15
>> But we will have a regular show tomorrow.
2:24:17
We're figuring out the exact.