// transcript — 1192 segments
0:00 <Untitled Chapter 1>
0:02 I think in general um my sense is that overall companies that have been
0:05 considering going public have now sort of opted to do that sort of route.
0:09 >> Okay. I I was asking because there's a great note in their S1 Jason you'll love
0:12 this. Uh they were talking about how they're using their AI powered virtual
0:16 agent chatbot and they say that our ability to control customer support
0:19 costs over time even as volume has gone up has contributed to an increase in
0:23 gross margin from 60% in fiscal 24 to 68 in fiscal 25 and even more. So
0:27 essentially they're automating things Jason static team size cost savings and
0:31 uh it's helping the company look more profitable. So kind of an AI story and a
0:35 fintech story a lot the the age of efficiency is upon us. Uh every unit in
0:41 every company um particularly in startups who are always resource
0:46 constrained they are the ones to first use these tools because they're resource
0:49 constrained. If they can save a they can turn a nickel into a dollar a startup's
0:53 going to do it. Now big companies like well we're spending $3 to get a dollar
0:56 in value and we can afford to do it because we're sitting on a bunch of
0:59 cash. There's no, you know, imperative. There's no existential dread about money at Apple
1:05 when you're sitting on hundreds of billions of dollars in cash. So yeah,
1:08 we'll get to it when we get to it. But at a startup, it's quite the opposite.
1:12 You're seeing that as well, I assume, lad. Is uh startups doing really
1:16 fascinating stuff with this technology already. >> Yeah, it's super exciting. I mean the
1:20 ramp on revenue for some of these companies is out of this world, right? I
1:25 mean between going from zero to a few hundred million in revenue in two three
1:28 years is something that I haven't seen in a very long time and there's multiple
1:33 companies doing that. So I think uh this is the big sea change and you see it in
1:36 enterprise adoption. You see it in all sorts of things. The company that um I
1:39 started with Jared and Eric and others that we mentioned earlier is basically
1:42 working on this for big enterprise. How can you help big enterprises really make
1:47 that adoption of AI and what applications should you provide for them
1:51 to be able to really thrive in this post AI world? So I think there's there's an
1:53 enormous amount of stuff to be done there. >> Yeah. And it's going to be really
1:59 interesting. The cynical take on this is a lot of people sampling AI solutions
2:03 and triing them and there'll be big churn issues and of course competition
2:09 issues and I guess the um optimist view of it which I'm primarily in that camp
2:14 is these things are providing massive value at very low price. Therefore, why
2:20 wouldn't you try it? It's like >> you know come get three hamburgers that
2:24 are grass-fed beef for 20 bucks. It's like how is that possible? Okay, I don't
2:27 need to know. I'm just going to order the burgers. Like, if you want to charge
2:31 me 20 bucks for something that saves me $2,000 a month. Okay.
2:37 >> And it's just wild to see how [gasps] mispriced some of these products are for
2:41 the value they're creating. >> This week in Startups is brought to you
2:46 by Squarespace. Turn your idea into a beautiful website. Go to
2:50 squarespace.com/twist for a free trial. When you're ready to
2:55 launch, use offer code twist to save 10% off your first purchase of a website or
2:59 domain. Sentry. Your team should be focused on shipping features, not
3:04 chasing down bugs. New users get three months free of the business plan, which
3:09 covers 150,000 errors. Go to centry.io/twist and use the code twist. And Pilot, focus
3:17 on your product. Let Pilot handle your bookkeeping. Pilot provides the most
3:22 reliable accounting, CFO, and tax services for startups and small
3:27 businesses. Head to pilot.com/twist All right, everybody. Welcome back to
3:37 This Week in Startups. I'm your host, Jason Calacanis. With me, my co-host,
3:42 Alex Wilhelm, and got a special guest today. Introduce our guest, Alex. Uh our
3:46 special guest today is the former founder of Mixer Labs and Color Health.
3:50 He helped incubate Brain and Co. but you probably know Allad Gil as an investor
3:54 per excelance. He has backed many many well-known names in the technology world
3:58 and Jason I believe commands the largest single AUM of any solo GP in the world.
4:02 All we're happy to have you. >> Oh, thanks for having me on board.
4:04 Our guest is iconic angel investor Elad Gil! What’s he working on…
4:05 >> Yeah. What are you working on these days before we get into all this crazy news?
4:10 What what's uh tickling your fancy? What what are you excited?
4:11 >> Yeah, it's kind of a mix of stuff. Yeah, it's a mix of stuff. Um you know I guess
4:15 three things. One is uh recently launched a company uh with Jared
4:18 Kushner, Eric Woo and a few other folks to bring AI to the world's biggest
4:22 enterprises. So really excited about that. Um second area uh just investing.
4:28 So um you know a lot of the companies I got involved with early Replexity,
4:33 Harvey, Decagon, a bridge etc. on the AI side are obviously all seeing you know
4:37 real growth acceleration etc. And then third, there's kind of some side
4:40 projects I've been working on around building large scale monuments,
4:43 translating books using AI, a variety of different things. And so there's also
4:46 more of the societal or philanthropic stuff that I think has been really fun.
4:49 >> Amazing. >> Kind of those three areas. Yeah. >> What tell me about this uh translating
4:54 Alexandria AI translates public domain books into all commonly spoken languages… Do people actually prefer AI translations?
4:54 books with AI. >> Yeah, we're calling it Alexandria. Um
4:58 the basic idea is that if you look at um every society, we we've benefited since
5:02 the days of Alexandria, the Egyptian library, uh from large-scale access to human
5:07 knowledge and books. And so we're taking a thousand of the most important books
5:11 that are off of copyright and we're translating them, working in concert
5:15 with OpenAI and Enthropic and 11 Labs and a few other folks on um creating a
5:20 book uh a version of the book in every language, audio books in every language.
5:23 So anybody around the world can stream it, they can download it, they can
5:25 interact with it. you've actually built a book reader. Um, and so, you know,
5:29 we'll we'll be doing a launch for that in the next few months. But, uh, that's
5:33 been incredibly rewarding. And we've actually done machine versus human eval
5:37 uh, where people check, do they like more what we've done with translation or
5:42 what human translators have done >> and um, people prefer the machine
5:47 translated works by far and so that's been a really interesting and an
5:49 unexpected. >> What do you attribute it to? >> What do you attribute it to? I think um
5:55 it is often more modern language, although you can obviously tune that
5:59 based on your prompts. >> Um and then I think often it's more
6:02 concise and often it's more consistent. You have better QA, especially if you're
6:06 thinking about a long work or corpus. You can enforce certain things that a
6:09 human translator may or may not be able to do. >> Fascinating. It is.
6:14 >> I want to know how long until you add a zero to that and then another zero. When
6:17 is it 10,000? When is it 100? And what's stopping you from going to like a
6:20 million? There's a lot of books out of copyright a lot. Yeah, I think um we
6:23 decided to start with the things that we thought were books that often should
6:27 exist in other languages that just have never been translated because if you
6:30 look at it, there's maybe two dozen languages that cover 80% of humanity.
6:33 >> Yeah. >> And uh you know for even some of the
6:37 greatest works ever produced or at least the consensus from you know academics is
6:40 these are the greatest books ever produced there just aren't copies in all
6:43 sorts of languages. You know 20 30 40% of humanity can't access them. Um so we
6:46 just thought let's start with the stuff that's kind of obvious. I think to your
6:50 point, you could do um we have the infrastructure built now. So you could
6:54 run an arbitrary book through and if you really wanted to, you could turn this
6:57 into a service where anybody should be able to upload any book and then
7:00 basically have this output in terms of um every language, every audio book. Uh
7:05 we we we've also been looking at building AI guides to the book so you
7:09 can start interacting with it, asking questions, etc. So I think it's almost
7:13 like this modern library of Alexandria, right? We want to bring humanity's
7:17 knowledge to the world in an accessible way. So >> it's been fun.
7:23 >> So you have you also gone down like the bit tensor and distributed computing as
7:29 related to AI and LLM's rabbit hole. >> Yeah, I've looked at it a bit. Yeah.
7:32 What's your conclusion on it? >> I don't have one, but I made a small
7:37 investment in a um like a hedge fund that's looking at all the nodes of Bit
7:41 Tensor every day trying to figure out the different projects, buying you know,
7:47 the the tokens, etc. And it feels to me like now that we're in the sort of Sachs
7:53 era of crypto regulation, I feel more comfortable spending a little time
7:57 there. And I've always felt like distributed computing and would be a
8:04 viable coexistor to cloud computing for certain jobs. Um, and we'll move from
8:08 that set at home >> kind of, you know, I I don't want to say
8:13 silliness, but hobbyist, you know, toy to something more meaningful. And
8:19 perhaps at some point even if we're building out too much infrastructure, it
8:23 could be like Bitcoin and that people will start putting up infrastructure and
8:27 know they'll make um a high return on it and you could get some you know I don't
8:34 know scale that was not previously considered realistic in the same way
8:39 Bitcoin and the number of >> you know uh people participating and the
8:44 number of nodes is would be hard to comprehend. and 10 or 15 years ago,
8:48 right? >> Mhm. Yeah, it's an interesting question because I think in the case of um
8:53 anything that has strong economic viability, compute tends to collapse
8:58 into centralization. And so, for example, um to your point on bitcoining,
9:02 I think uh it's like the six or seven largest mining pools are over 50% of
9:05 Bitcoin volume. >> Yeah. >> And so that's centralization in terms of
9:10 buying of hardware, in terms of energy and negotiation for that. And so anytime
9:13 you have something with a huge economic output, it seems to almost always
9:16 Why compute tends to centralize over time… (It’s because of economies of scale!)
9:17 centralize because of those economies of scale. >> Um, and then to your point, there's all
9:20 sorts of circumstances where you have more distribution either because it's
9:24 people who are just, you know, generating an extra bitcoin every once
9:27 in a while because they have some idle capacity, but on a volume basis that's
9:29 Squarespace - Use offer code TWIST to save 10% off your first purchase of a website or domain
9:31 pretty low. If you want your business to succeed online, it has to stand out. And
9:35 that's why you need a beautiful worldclass website. The good news is you
9:39 don't have to hire an expensive designer anymore. Nor do you need a bunch of
9:44 developers. No. Squarespace has all the tools you need to claim your domain and
9:47 start building your company. Hey, maybe you got a product to sell. Maybe you
9:50 want to show off some samples of your work. Maybe you've got a new service
9:54 you're providing. Well, Squarespace is the all-in-one tool you need for your
9:59 business to grow and flourish. And they offer beautiful templates. Maybe you
10:03 want to start an online course or you're scheduling appointments for people.
10:06 Maybe you're generating client invoices. All of that is built into the product.
10:10 Everything you create with Squarespace is pre-optimized to show up in search
10:14 engines as well. So, you don't need to hire some expensive SEO person. Nope.
10:18 They're building your metad descriptions. They're setting you up
10:21 with an autogenerated site map. And the SEO is done before you even get started.
10:26 Plus, there's a new AI powered feature. It's called Blueprint, which makes it
10:30 easier than ever before to customize your website and make it really pop. So
10:34 check out squarespace.com/twist for a free trial. And when you're ready to
10:38 launch, go to squarespace.com/twist to get 10% off your first website or domain
10:50 or it could be things like steady at home which is more philanthropic or very
10:54 early bitcoin mining or crypto mining where you know there isn't that same
10:58 liquidity in terms of mining of the token. So it's a really interesting
11:01 question of when do you transition from decentralized to centralized and
11:06 >> the charact and it's an open consolidation that occurs because it's
11:10 not like you have to ask a gatekeeper. It's not like you have to go to you know
11:14 somebody and say I would like to participate in this. Can we cut a deal?
11:17 You're not like negotiating with Oracle or Microsoft. It's like you can just
11:21 participate and >> if you have innovations that you can
11:26 bring to bear. You you come up with your own chipsets. You come up with your own
11:30 power strategies. You >> have a cousin who works at a nuclear
11:35 facility in China and they >> they happen to give you a little space
11:38 for your for your nodes. All of a sudden you're up to the races. So I'm I'm super
11:43 bullish on it. Also like >> um if you just think about all the
11:46 inference that's occurring, it's going to be unlimited demand for this stuff. I
11:52 think I don't know that the demand curve ever. >> Um yeah, I don't know if it's going to
11:59 abate anytime soon. It would What's your take on it? Like the overbuild versus
12:04 >> you know the dark fiber analogy. Do do you have a position on it or does all
12:09 this stuff do you have in paradoxic way to a conclusion? Yeah, I don't have a
12:13 strong point of view on it. I do think there's some really interesting shifts
12:16 that are happening. So to your point on the inference side, eventually more and
12:19 more inference will happen on device. And so to some extent the distributed
12:22 comput notion may just be everybody's iPhone in some sense >> in terms of at least the inference side.
12:26 Um in terms of training, the thing I've been looking at most is the degree to
12:30 which energy costs will impact where training centers are actually going to
12:32 get built. >> Yeah. And one thing that I think is really under discussed is the fact that
12:39 given how early Europe has governed itself relative to energy, you know,
12:43 shutting down nuclear power plants in Germany and um being dependent on
12:48 Russian oil, etc. Um, it basically means that training data centers are unlikely
12:49 So are we building TOO MANY datacenters? Will AI apps eventually run on your phone anyway?
12:52 to go there because it's going to be too expensive to run them just because of
12:55 energy, ignoring regulation, ignoring everything else. which means that
12:58 basically all the training centers should end up in the Gulf and in the US
13:03 and the Gulf in part because under the Trump um AI plan uh basically there's
13:09 some extension of some rights relative to uh being able to run training in UAE
13:14 and a few other places and those are going to be places with low energy
13:16 costs. So I think there'll be parts of the US, parts of the Gulf where all the
13:20 training is going to run. Inference is probably more globally distributed in
13:24 part because of uh GDPR. So Europe will force it from a regulatory perspective,
13:28 but they're going to lose out on tons of GDP spend and capex spend because people
13:33 just won't build training centers there. So I think that's a little bit
13:35 underdised right now. >> It's quite a paradox because also in the
13:39 Middle East, they're adopting solar and batteries at an extraordinary pace.
13:44 These are people who it seems to me the people who understand energy are not
13:49 being particularly precious about where the energy comes from. In Texas, we have
13:52 the largest >> installed base of solar. And I can tell
13:57 you living in Texas, they're not super precious about, >> you know, where the energy comes from.
14:01 Is it the cheapest way to do it? Great. We can make a business out of that. We
14:04 got a grid. Let's compete. And that's what they're doing in the Middle East
14:08 where there's a lot of sun. There's a lot of sun and there's a lot of capital
14:12 >> to buy those solar panels. And you're right, they just their energy costs are
14:17 going to be so dimminimous. Um and they also have a populace in that region that
14:22 has been educated in the west over the last 30 or 40 years very quietly at the
14:26 top institutions who have now boomeranged back >> to Qatar to UAE to Saudi uh you know and
14:34 then of course you have Israel which is you know just a juggernaut per capita in
14:39 terms of startups. So it's I'm spending more time in the region. I'll probably
14:44 spend two weeks, three weeks, a year there >> um looking at startups and working with
14:48 different investors. Are you spending time in the region? >> A little bit. Yeah, I'm an investor in
14:53 um a few companies, some of whom have been bridged across Israel
14:56 traditionally. So that'd be things like trip action, it's not called Navan, etc.
14:59 But then also more recently a company called Eon, which is doing data infra,
15:03 one called Kella on the defense side, etc. So I have a a handful of
15:07 investments in Israel and then um obviously there's just really
15:09 fascinating things happening across the entire region. Yeah, it's going to be
15:12 really interesting when you can fly direct from, you know, Tel Aviv to Riad
15:21 to DOA to, you know, this is kind of mind-blowing in our lifetime that this
15:24 is happening. >> So, Allad, you mentioned Nvant, which is
15:27 going public right now. It's currently in a little bit of a limbo with the
15:30 government because the government shut down and such, but uh what can you tell
15:33 us about the company's, you know, road to IPO? >> Yeah, I can't really say anything. I
15:37 mean, um, you know, usually IPOs have quiet periods. I think um in general
15:41 >> I was hoping that didn't apply when the government was shut down but apparently
15:43 [laughter] not. >> In general um my understanding of what's
15:48 happening in a meta situation is just uh because of the government shutdown the
15:53 SEC for a short period was not reviewing or fully reviewing S1's and then they
15:57 actually issued a rule that would allow companies to still go public um if they
16:00 did certain filings and certain disclosures and then within a certain
16:04 time period they could go out. And so I think in general um my sense is that
16:07 overall companies that have been considering going public have now sort
16:11 of opted to do that sort of route. >> Okay. I I was asking because there's a
16:15 great note in their S1. Jason, you'll love this. Uh they were talking about
16:17 how they're using their AI powered virtual agent chatbot. And they say that
16:21 our ability to control customer support costs over time even as volume has gone
16:24 up has contributed to an increase in gross margin from 60% in fiscal 24 to 68
16:29 in fiscal 25 and even more. So essentially they're automating things.
16:34 Jason static team size cost savings and uh it's helping the company look more
16:37 profitable. So kind of an AI story and a fintech story a lot.
16:39 Jason says “The Age of Efficiency is upon us.”
16:41 >> The the age of efficiency is upon us. Uh every unit in every company um
16:46 particularly in startups who are always resource constrained they are the ones
16:51 to first use these tools because they're resource constrained. If they can save a
16:54 they can turn a nickel into a dollar a startup's going to do it. Now big
16:58 companies like, well, we're spending $3 to get a dollar in value and we can
17:01 afford to do it because we're sitting on a bunch of cash. There's no,
17:05 you know, imperative. There's no existential dread about money at Apple
17:08 when you're sitting on hundreds of billions of dollars in cash. So, yeah,
17:12 we'll get to it when we get to it. But at a startup, it's quite the opposite.
17:15 You're seeing that as well, I assume, lad. Is uh startups doing really
17:19 fascinating stuff with this technology already. >> Yeah, it's super exciting. And the ramp
17:24 on revenue for some of these companies is out of this world, right? I mean,
17:29 between going from zero to a few hundred million in revenue in two three years is
17:32 something that I haven't seen in a very long time. And there's multiple
17:36 companies doing that. So I think uh this is the big sea change and you see it in
17:39 enterprise adoption. You see it in all sorts of things. The company that um I
17:42 started with Jared and Eric and others that we mentioned earlier is basically
17:46 working on this for big enterprise. How can you help big enterprises really make
17:50 that adoption of AI and what applications should you provide for them
17:54 to be able to really thrive in this post AAAI world? So, I think there's there's
17:57 an enormous amount of stuff to be done there. >> Yeah. And it's going to be really
18:02 interesting. The cynical take on this is a lot of people sampling AI solutions
18:07 and triing them and there'll be big churn issues and of course competition
18:13 issues. And I guess the um optimist view of it which I'm primarily in that camp
18:18 is these things are providing massive value at very low price. Therefore, why
18:23 wouldn't you try it? It's like, >> you know, come get three hamburgers that
18:28 are grass-fed beef for 20 bucks. It's like, how is that possible? Okay, I
18:31 don't need to know. I'm just going to order the burgers. Like, if you want to
18:34 charge me 20 bucks for something that saves me $2,000 a month. Okay.
18:41 >> And it's just wild to see how [sighs] mispriced some of these products are for
18:45 the value they're creating. >> Yeah, 100%. And some of them are uh
18:48 burning a lot of money along the way to get there. So, as uh to your point, the
18:53 pricing versus the value differential in one or two well-known cases are also
18:57 just leading to odd economics. But I think most of them are very solid in
19:00 terms of what they're providing the cash the potential cash flow to support the
19:03 business etc. >> So a lot on that point uh we're not seeing a lot of gross margin negative
19:08 businesses then because I've heard bits and pieces about people saying that some
19:10 companies are so enterprised that they're actually gross margin negative
19:14 which is so uneconomic as to make me worry but sounds like that's not the
19:18 case too often. No, I think there's one or two very high-profile versions of
19:22 that. And one could argue in those cases, they're basically trading
19:24 When companies trade inference for market share
19:25 inference for market share. And so you could view it as a marketing funnel.
19:27 Sentry - New users get 3 months free of the Business plan (covers 150k errors). Go to and use code TWIST
19:29 >> I've been working with and investing in founders for so long, I've got
19:33 supernatural startup powers. For example, I know your team is spending
19:36 too much time on debugging. Your engineers are chasing down bugs at 1:00
19:41 a.m., 8:00 a.m., and every time in between instead of building amazing new
19:46 features or pushing fresh code. Just admit it. you'll feel better. The
19:49 reality is from time to time your code is going to break. We all know that. But
19:53 it doesn't have to be the end of the world. Now there's Sentry. They're the
19:57 one-stop application monitoring platform that's not just going to tell you, "Hey,
20:00 something's broke," but what specifically happened, where, and most
20:04 importantly, why. Plus, they've added Sentry Logs. So now, when something goes
20:09 wrong, you get a complete report. No more random jobs that just don't run. No
20:13 more tab hopping, just actionable information with tons of context. And
20:17 that means faster fixes. And we want to give you three months free of Century's
20:22 team plan. That's right, f just for being a loyal Twist listener. That
20:26 covers 150,000 errors. Just go to century.io and use the code twist.
20:32 That's s n t r y.io and make sure you use that code twist so they know we sent
20:41 But I think in most cases, most of the companies I know are running at, you
20:44 know, quite healthy margins. and in some cases software plus one thing I've been
20:48 investing behind over the last two years kind of quietly is these AIdriven
20:52 rollouts. So can you fund the purchase of services businesses roll them up and
20:57 then increase margin dramatically using AI because effectively you're making the
21:00 people on those teams dramatically more productive and that's been a thesis
21:03 that's really been proving itself out quite strongly. There's two companies
21:06 I'm involved with in particular where we're seeing very strong lift
21:11 >> um and very strong execution relative to these really big opportunities because
21:13 services >> can you say the names of them or are they stealthy?
21:16 >> I can I can mention one of them. The other one's quite stealthy. Um the one
21:19 that is non- stealthy is called long lake >> and or long lake management. So lln and
21:25 um basically >> clever >> uh yeah very good. Um one of the
21:31 founders Alex is uh an exceptional PE person. uh Zach Frankl who's more
21:35 technical helped get it up and running and they um uh basically have a really
21:40 great team of both engineers but also private equity and operations people and
21:46 if you think about it the um the key for any sort of AI driven transformation is
21:50 you need if you're doing a rollup you need a PE person if you're not you just
21:54 need you know uh the sort of edict to do it then you need somebody to
21:57 Why one of the main challenges of adopting AI is buy-in and convincing teams to use it.
21:57 operationally implement it a lot of the issues with adopting AI is not the AI a
22:00 lot of it's actually pretty easy to build given that AP API that we have
22:03 from Open AI and anthropic and others for Google. Um, it's usually the change
22:08 management. It's convincing the team to adopt it. It's aligning incentives. It's
22:12 saying, "Hey, we understand you really like this tool, but we're switching."
22:16 You know, it's all that kind of stuff. >> Real, right? The resistance.
22:19 >> The resistance is very real. Yeah. In some cases, maybe it's going to threaten
22:22 that person's job or perceptually do so. So, >> um, often when you're doing rollups for
22:27 AI, you need the PE person to be able to do good purchases. You need the ops
22:30 person to really help revamp the org. And then you need the engineers to
22:33 implement the AI and the workflow. And so very few of the rollup companies that
22:38 I've seen have this magical convergence of those three skill sets into one team.
22:42 >> Yeah, Athena isn't a rollup, but they're, you know, people think of them
22:46 as, and I'm the first investor in that company, as well as was the first
22:49 investor in Thumbtac, Jonathan's other company. Um, people think of it as like assistants,
22:57 right? uh virtual assistants uh in a different time zone at a different
23:00 price. But it's really an AI company because every time they can automate a
23:04 task across 10,000 assistants, if they can help one assistant schedule 5%
23:09 faster, right down the line, if they can help them book hotels faster, right down
23:14 the line, etc. Um and we have an investment in a company called Howie uh
23:17 that Austin Peter Smith is doing that is just doing one thing, scheduling better
23:21 than anybody. So there's a lot of people taking swings at this. I guess the
23:25 company this week of note, the main character energy goes to Amazon. Alex,
23:30 maybe you could tee up the breaking news from the last 48 hours uh about Amazon
23:36 and jobs and employment. >> Yeah. So, the New York Times
23:40 >> Yeah. New York Times got its hands on some internal documents from Amazon, the
23:44 cloud and e-commerce giant. These are from the 2024 time frame, so they're a
23:47 little bit dated, but essentially the company's investments in automation at
23:51 the warehouse level in particular are going to help it hire 160,000 fewer
23:57 people through 2027. And the times right that could save them up to about 30
24:00 cents per package that they pick and then deliver. 30 cents is not that much,
24:03 but if you multiply times billions of packages, it adds up quite quickly. and
24:07 all the work they've done over the last 10 to 15 years to build out robots and
24:11 increase automation in their warehouses will save them about 600,000 hires
24:16 through 2033 and the ultimate goal is to get to about three quarters automation
24:20 about 75%. Um, one of the major stories you talked about through the COVID era
24:23 was how many people Amazon hired to build out its warehouses as e-commerce
24:27 took off during that period of economic disruption. Now we're seeing the other
24:31 side of that and a lot of long-term bets by the company come good. Um, great news
24:34 for Amazon shareholders. all of us here who have 401ks. Pretty tough, I think,
24:38 Jason, looking ahead for the people who currently uh feed their families on $19
24:43 an hour Amazon warehouse jobs. >> What's your take, Lad? Uh seems
24:47 conservative to me, but what are your thoughts on automation, robotics?
24:52 >> Yeah, purpose-built humanoid. >> Yeah, it's interesting. It feels like um
24:55 for a long time, robotics was one of those areas that was always going to
24:59 happen and didn't quite, right? If you looked at the installed base of robots,
25:02 it was in the hundreds of thousands of new robots per year largely for
25:06 automotive and very specific industrial applications and it seems like now we're
25:09 going through a sea change and I think um you know it's kind of funny actually
25:12 if you ask what's a robot your dishwasher to some extent is a robot
25:15 right like a lot of things that we kind of take for granted now are robots um in
25:19 the modern era the things that are really emerging now are obviously some
25:21 of the things that were just mentioned relative to Amazon but also even
25:26 self-driving cars and Whimo and uh Tesla and what's happening there those are
25:28 robots right those are cars that drive themselves. There's a lot of startups
25:33 right now uh Pi and the bot company and a few others who are all building
25:38 nextgen sort of uh foundation models for robots. And so it seems like within the
25:41 next couple years we'll see a lot of these things substantiate. And there's
25:45 sort of two questions that come to mind for me. One is what proportion of the
25:47 Elad’s robotics questions: (1) What % of winners will be incumbents?
25:48 winners will be incumbents versus startups. So for example, if you look at
25:53 self-driving, there was two dozen self-driving companies, but the winners
25:56 arguably are Whimo from Google and then Tesla, right? two incumbents actually
26:00 effectively have so far at least once self driving there's cool companies like
26:03 applied intuition and others who are providing pieces of this but you know
26:07 car OS and other things but I think um that's sort of clear winners at least so
26:10 far those two and robots it seems like Tesla will be one >> right through their optimist work and
26:15 then the question is who else is it a startup is it another incumbent is an
26:19 incumbent who buy the startup um is it at home robots which a lot of people are
26:23 working on uh which I I I view would be kind of awkward there's like a robot in
26:26 your living room and you walk in and you know it's like mopping or something. I'm
26:31 joking with that. Um or it's a >> um or is it an industrial application
26:34 more B2B which is sort of the Tesla Optimus sort of view and so I think
26:39 there's this big question of um you know who wins and why and then what
26:43 are the first application areas on top of that. >> Yeah, it was pretty obvious to me. I saw
26:48 Andy Jasse speak at a private event I don't know a couple months ago and um
26:54 you know was pretty bullish on AI and then a day later he did this public
26:58 missive um where he wrote hey AI is going to transform everything and we could see
27:06 different team sizes that was like a >> to me when a CEO that sophisticated
27:11 doesn't write that memo without a reason he was telegraphing internally
27:16 externally with partners Hey, you know, this is going to result in less jobs.
27:22 And um I I was on the Bull Work podcast, a political podcast, and I said, you
27:28 know, um I I don't see a world in, you know, the next 10 years where a human
27:33 being touches a package at an Amazon factory or to your doorstep.
27:37 >> Here's here's the quick clip. >> Jason, do you want to just run the clip?
27:40 >> Yeah, it's a little self-indulgent, but sure. Let's uh
27:42 >> It's this is literally your podcast. You're allowed to be self-indulgent.
27:44 Don't worry. I don't want to be too self-indulgent here. Quoting
27:48 >> we'll stop you. Anyways, here's Jason quoting himself
27:50 Jason called the Amazon news last month and we have the receipts!
27:52 >> is where things could get really nutty. Um, and we're already starting and
27:55 that's where like the doomsday scenarios, Terminator ones kind of come
27:59 out of that group. Um, >> so what happens if you know all drivers
28:06 go away? Uh well, every self-driving car is four full-time jobs, and every uh
28:11 humanoid robot in a factory is five jobs, maybe six, because those jobs you
28:15 can you can work for maybe seven hours as a just physical limitation. As
28:18 humans, we're driving, you can work a 12-h hour shift. >> And those are those are going to be
28:24 here, folks. And before 2030, you're going to see Amazon, which has massively
28:28 invested in this, replace all factory workers and all drivers. The idea that
28:32 when you order something from Amazon, a human would touch it at any point in
28:37 that supply chain is insane. It will be 100% uh you know, robotic, which means all of
28:43 those workers are going away. Every Amazon worker, all those jobs, you UPS
28:49 gone, FedEx gone, every all of those are going to be gone. And those companies
28:52 will be more profitable. And when you order something, it's going to come
28:56 faster and cheaper and better. And your Uber will be half as much. But somebody
29:00 needs to retrain these people. And that's actually a very val What do you
29:04 think about job displacement, Alad? Um, obviously I have strong feelings that
29:09 this displacement is coming. Um, I don't know if anybody disagrees with that. I
29:12 guess we're all just going to debate timelines. Three years, 5 years, 7
29:17 years, 10 years. Do you think we're going to have societal issues with uh job
29:23 displacement? We're already seeing computer science graduates having a hard
29:28 time recently getting jobs. Uh, and I think most people are attributing it to
29:32 AI making the senior people more um, productive. So, what's your take on job
29:36 Pilot - Visit and get $1,200 off your first year
29:36 displacement? I'm curious. Probably the best part of being a founder is watching
29:41 a project that started as just an idea blossom into a full-fledged startup. The
29:46 worst part, it's bookkeeping. It's boring. It's detail- oriented. And it
29:50 takes time away from building your product. and hey, growing your team.
29:54 It's time for you to let Pilot take care of all this bookkeeping so you can get
29:58 back to running your company. Pilot's the largest accounting firm and it was
30:02 built just for startups. They understand the technology industry and the
30:05 landscape. They know the stakes and they're already trusted by amazing
30:09 companies we love like OpenAI, Scale, Air Table, and beyond. And you know
30:13 what? Pilot is about so much more than just getting your taxes done on time.
30:18 Pilot has a smart dashboard that puts all the essential metrics your startup
30:22 is tracking at your fingertip. And they have helpful adviserss who can walk you
30:26 through crucial financial decisions you're going to need to make in the
30:29 coming years. Plus, when it's time to raise your next round, Pilot COO and CFO
30:35 services are there to help. So, it's time to focus on your company and let
30:39 Pilot handle the books. And Twist listeners get $1,200 off their first
30:44 year. just go to pilot.com/twist >> Yeah, I mean I think any technology wave
30:55 uh definitely has uh job dispeling that comes with it. And so then the question
30:58 is what do you do from a retraining perspective? Um how do you think about
31:02 social safety nets? And then also how well you governed. If the you know if we
31:08 were governed remarkably well I'd say well we'll figure it out and we'll if
31:11 we're governed really badly then it's probably a different story in terms of
31:14 the degree to which you end up with uh you know different forms of unrest or
31:23 or other things you provide. So I think it's a pretty complex thing. Um but I do
31:29 think that it's it's pretty clear in certain areas that you're starting to
31:33 see shifts in terms of um labor particularly around you know customer
31:37 uh support may be one of the earliest places that I think ends up getting
31:40 impacted to some of the points made earlier. >> Yeah I think point Alex
31:45 >> oh just Salesforce said they're saving 100 million a year now on automated
31:49 customer support a lot just to back up your point it's a place where people are
31:51 finding enormous cost savings already which of course impacts jobs. Yeah. And
31:55 if you have like a temp or outsourced labor team or you've partnered with
31:59 somebody about it as a company, you're more likely to willing be willing to cut
32:01 that back than if it was your own employees, right? And so often when you
32:04 have these outsource or delegated systems, >> uh you're more willing to shut them off.
32:09 And so that could be some of the uh savings without actually changing
32:13 official employee headcount, but it it changes the virtual set of people that
32:16 you've hired. And I think that's >> that's why we're not seeing it in some
32:19 cases as much, right? It's kind of hidden because >> such a good point. Yeah. You outsourced
32:24 it already. So you're just ending a contract. You're not >> doing a plant shutdown. So in the United
32:30 States you have that >> it's not available warrant notice.
32:34 >> The warrant notice like warrant notice like you don't have to warn anybody.
32:36 It's like yeah I just got rid of a vendor. And I wonder if that is a pass
32:39 through >> where you still have to do a warn notice. Putting it aside. Um,
32:45 I will say I just had this realization as you were talking. I now as a default
32:53 do not go to, you know, the manufacturers's website to get an
32:55 answer. >> I go to an LLM. Why? The LLM has indexed all the Reddit threads, Kora threads,
33:03 Twitter threads, >> you know, uh, message boards of people
33:08 solving the problem on their own >> before the company even knows there's an
33:15 issue. and you so I think most organizations probably don't even know
33:20 what their true customer support >> uh request. >> Yeah. Yeah. And there's companies like
33:25 Decagon obviously helping with that and then um I think there's two other
33:28 implications of that or implications of all this. One is there are certain
33:31 countries that should be impacted sooner rather than later from a employment
33:33 perspective that we're not actually thinking about. Right. Because there are
33:36 certain types of jobs outsourced to specific regional hubs >> instead
33:42 etc. Sure. And then um I think the second thing and this is a point that
33:45 Nikesh Aurora at the CF Palto Networks has uh brought up with me in the past um
33:49 which I thought was a really good insight that you know he said this quite
33:53 publicly that they're using uh AI tools and foundation models and everything
33:55 else to iterate on product because customer support in some cases not all
34:00 just means your product isn't working good enough or the UI is hidden or
34:03 something else is going on. And so to some extent by iterating on product
34:08 faster, you're going to cascade into support as well from that perspective,
34:13 right? It's essentially if you have bots fixing every problem for you
34:17 automatically, your support load should go down, >> right?
34:19 >> And so there's all these really interesting sort of second or third
34:23 order things that that you know are really interesting to speculate on in
34:27 terms of the future world. It's really interesting to see the
34:30 number of people employed at these companies stay the same as their
34:34 earnings go up. >> Exactly. And this is why. Yeah. >> We've never seen that in our lifetime.
34:38 And this is >> if you were going to think about where to park money in the world.
34:44 >> I was hearing somebody who's like super liberal podcaster who hates America and
34:49 Trump, derangement syndrome, the whole thing. And he's like, I'm taking all of
34:53 my money out of the United States. >> Not you, Alex. I don't think you're
34:56 >> I was gonna say, dude, I'm sitting right here, not you.
34:59 >> You don't have TDS. >> Just got doxed. >> Yeah, exactly. [laughter] You don't have
35:03 TDS. You just have a love of democracy. It's slightly different.
35:06 >> And and I'm a capitalist. >> Yeah. Um and he said, I'm just putting
35:10 it I think United States is on the decline. I got to put my money in Europe
35:15 and every other thing. And I was like, >> a narco capitalism would argue the
35:21 opposite. [laughter] In fact, people are going to deploy this
35:25 technology aggressively >> uh in the most vibrant capital market on
35:31 the planet, the United States. >> Um, we could see Amazon, to my point
35:37 earlier, like what does Amazon look like with no delivery drivers? What does it
35:43 look like without wi with, you know, delivery times cut in half and running
35:48 24 hours a day? >> What does that even look like? Yeah, eventually it's just Jeff Bezos
35:54 and a bunch of bots though. >> I mean, back to the one person billion
35:59 dollar startup. Like maybe it's a, you know, 10 person trillion dollar startup
36:03 at some point if it's if it's that efficient. >> Um, here in Texas, they're delivering
36:11 stuff from Amazon warehouses so fast >> that I'm like, I don't need it that
36:14 fast. They literally like, would you like this from 4:00 a.m. to 8:00 a.m. or
36:17 8:00 a.m. >> to, you know, 10:00 a.m.? And I'm like, um,
36:25 >> it's 10 p.m., sir. I don't need it. I'm ordering a cable. It's 10 p.m.
36:28 >> tomorrow. Yeah, it's okay. At >> your leisure. [laughter]
36:31 >> Yeah. >> And like literally the drones are coming. Like if you live on a ranch or
36:37 something like that. I mean, the drones are going to be the ultimate.
36:40 >> Um, >> how crazy is it that >> Larry and Sergey bought every robotics
36:49 company for five years? what 1015 years ago and then they sold them all because
36:51 they couldn't figure out an application for it. I mean, >> yeah, they were so far ahead on so many
36:55 things. And so, I think a lot of creditors do in terms of being really uh
37:00 advanced on the curve. And, you know, there's the old saying that um I mean,
37:04 there there's there's two different things I think that apply here, but one
37:08 of them is just that um uh you know, a the features here is just not
37:12 distributed, then b um you know, often it's less a matter of um of uh uh who
37:18 but when. >> And this is a good example of a when, right? They they were just too early
37:24 >> and I think they've been remarkably early on all sorts of things. They
37:26 experimented with delivery. They experimented with uh internet using
37:30 weather balloons. They experimented with satellites. They exper you know because
37:33 eventually I think they sold their satellite company to Planet Labs for
37:35 somebody, right? But they had satellites. They had everything.
37:40 >> They were just too early on so much. So >> kudos to them, you know, for trying.
37:44 >> And you know, it's like a venture portfolio. You only have to hit on one.
37:47 And it looks like Whimo will be that one of the other bets. The Loom project was
37:54 too close to Earth. Turns out low Earth orbit better. You look at Google Glass.
37:59 >> Um that might actually be landing soon in the form of these Ray-B bands that no
38:03 everybody can't shut it about shut up about for meta. >> So they could have kept going with that.
38:07 I also think their Chromebooks and the Chrome operating system is eventually
38:11 where >> um AI lands. Like if you think about what will be the starting point for
38:20 every AI journey, where do you think that will occur? Do you think it's voice
38:26 and wearables, phones, dumb terminals? >> What do a new operating system? Where do
38:30 you think the endgame is for mass consumption, mass interaction of [sighs]
38:34 AI? >> Yeah, I think it depends on how good the companies are who are currently
38:38 providing all our hardware because the default state is they should win. Now
38:42 the question is are they going to win which I think is a separate question and
38:46 one could argue that there's been uh very little ondevice innovation in AI
38:52 and then the question is is that because they're waiting to see what works is it
38:57 because they're um you know uh investing and it takes time to build client
39:01 applications is it a lack of imagination and so I think there's this broader
39:05 question of why has so little AI been built >> into these devices so far and in some
39:09 cases um you know you look at something like theory it was never you know the
39:14 best voicebased system etc and so there's a there's a broader question of
39:18 um you know why is that and in some cases it's to protect users so for
39:22 example Apple has a very strong data privacy policy which means that they
39:25 don't have a lot of data that they normally would have for training that
39:27 Google has >> and so it's been purposeful but now in the AI era should that still apply or
39:32 not or how should you think about it as a company I think those are very
39:36 interesting probably ongoing questions in terms of where does that AI feature
39:39 come from the default should be the devices you use today or the devices
39:42 that should be the launching point for AI. Whether that actually happens, I
39:45 don't know. I mean, what do you think? >> Yeah, I've been thinking about Apple a
39:51 lot. Um, they are very good at making hardware. One of the greatest things
39:55 they ever did was get off of other people's chipsets and just make their
39:59 own, right? And that's been kind of the revolution. And they can make devices
40:03 that are higher margin, more powerful, leaprog everybody, low battery
40:06 consumption, solve so many problems for them. these laptops they build are
40:11 unnecessarily powerful and unnecessarily long battery lives because they've just optimized so
40:17 well. So, they actually have an advantage in that. And then >> there will be a time when people have
40:25 their LLMs um come up in um you know lawsuits and discovery and Sam Alman will be more
40:32 than happy to turn them over. He said like don't put your stuff in here folks
40:36 like we will >> you know have to hand them over you know
40:39 I think that's good that he was honest about that. So encrypted on your device
40:47 uh with an Apple phone, Apple Mini, Apple, you know, M17, whatever they get
40:51 to, you could actually run the LLM locally. >> All the, you know, context window could
40:58 be run locally on all your items and it could be running in the background
41:01 indexing your photos >> without ever putting them in the cloud
41:06 like Google and having your photos, you know, train Google's LLM, which I don't
41:09 think you really want to do. and Zuckerberg will do that in a heartbeat
41:13 obviously. So I I think they have a potential there. Yeah. >> Yeah. It's a really interesting question
41:17 because if you look at search as an analog and I don't think it's the
41:21 correct analog, right? Um but if you look at as an analog, basically Google
41:26 ended up paying Apple tens of billions of dollars a year to use it search
41:29 instead of developing its own. >> And it was all margin, right? You didn't
41:31 have to hire all the engineers. You didn't have to implement all this stuff.
41:34 You literally got paid money to just distribute somebody else's search. M
41:38 >> effectively it seems like that's the path that's been taken so far in LLMs
41:42 around foundation models and the question is is that the right approach
41:45 or wrong approach in this era because you're not just searching for things
41:48 >> it's not just a way to interface with the web it's the way to interface with
41:52 everything you're shifting the the the set the framework >> but from a deal perspective that deal
41:57 worked so well that I wonder if that's part of the framework that they're
41:59 thinking it I have no inside knowledge of I'm just saying if I if I had gone
42:03 through that historical arc maybe that's how I'd be thinking about things. Um,
42:07 and it may or may not apply to this different world of like uh AI doing sort
42:11 of all these various tasks for you and agents suddenly spawning up like in the
42:14 agentic world that's a very bad strategy potentially. >> I'm so uh I'm so bullish on Apple making
42:21 a big AI purchase. I just think they should just take out >> whatever LLM doesn't break, you know,
42:28 stay in the top three. They should just take out Perplexity is a great user
42:34 interface. They really know how to build >> slick interfaces uh the comment browser
42:38 and >> my lord just uh since the antitrust case >> think about how much traffic they could
42:47 spill over to duck.go Brave or Perplexity or anybody in that kind of
42:50 zone [clears throat] of excellence. It would be extraordinary for those
42:52 companies. >> And um >> yeah. >> Yeah. I led um Perplexity's first round.
43:00 >> Wow. And Arvin is one of the most uh impressive founders I've worked with in
43:03 a really long time. Like just a complete he's >> incredible at hiring, incredible at
43:08 execution, highly strategic, >> morphed into a very commercial person. I
43:11 mean, when I met him, um he was at OpenAI as a researcher, right? Uh and I
43:16 actually met him because I think he reached out to me on LinkedIn
43:20 and said, um, hey, I was, you know, I'm working at this lab at Berkeley and I
43:24 joined OpenAI like eight months ago and and this is when nobody was doing
43:27 generative AI stuff, right? I think I founded him in uh 22 sometime and um so we started
43:34 meeting up weekly and just brainstorming and you know he obviously came up with
43:38 perplexity and that's what he wanted to do and what he wanted to pursue and all
43:41 this stuff. Uh but I've just been impressed by how he's um really just
43:46 ramped so so quickly on so many areas. So it's it's been a very impressive
43:49 journey to watch. >> Yeah. Are companies sniffing around
43:53 Perplexity trying to snap it up a lot? >> No, I don't know. I'm not on the board
43:55 there or anything. So >> Okay. And now look at >> trying to get this breaking news. I like
43:59 it, Alex. >> Well, no, I'm just I mean Jason's been talking about his secrets. [laughter]
44:06 >> The different the difference. >> What's the worst thing that happened in
44:09 your childhood? Let's talk, you know, like >> no one cares about my secrets a lot.
44:12 That's that's that's the difference. My secrets are news. No, but I'm more just
44:15 asking the general point like how much interest is there from the incumbent
44:19 technology companies towards AI startups? because we've seen some
44:22 acquisitions, but Jason's keep seeing these interesting possible combinations
44:26 and I don't see as many of those >> uh in the market as we'd expect. So, I'm
44:28 just kind of curious about the the temperature or the appetite for that
44:32 kind of dealm. >> Yeah, I think it's because um you know
44:39 the uh the FTC has continued to be reasonably anti-large tech acquisitions
44:43 and that was started under the Biden administration um with Lena Khan and
44:46 others and I think that's just been continued into this administration. So I
44:51 think it has um as much to do with kind of a purview of how the government may
44:55 intervene. And that's not just the US government, it's also the EU, it's the
44:58 UK, right? The Figma deal was scuppered by the UK. It wasn't the US that did it.
45:02 And it almost feels like there's this um regulator Ron Robin where these
45:07 different groups are taking turns in terms of who's going to scupper which
45:09 thing so it isn't the same party every time. >> Yeah. They must be on a group chat.
45:13 They're like, "We're busy with stopping the totally inconsequential Figma Adobe
45:18 deal. Why don't you guys find an inconsequential robotics purchase by
45:23 Amazon for a two billion dollar company to block? And it's like,
45:26 >> how does the UK have that much power though, Jason? Like, the UK is this
45:30 little bankrupt off the coast of Europe, which is a natural economic block.
45:33 Jason says Adobe and Figma should abandon the UK entirely.
45:33 >> I I said this publicly many times. They should just ban you from using Adobe and
45:38 Figma products in the UK for ever [laughter] and just punish them if for blocking it.
45:43 They should just say like, "Okay, if you're going to block it, we're not
45:45 going to sell to your citizens. they can go use some open- source project. Hey,
45:50 um, speaking of interfaces, we have this sesame round which I'd like to go to,
45:55 Time for a Polymarket: The sharps say 80% chance Tesla beats their quarterly earnings
45:55 but before we do, I think it'd be good to look at a poly market. Will Tesla
46:01 beat their quarterly earnings? So, I chose this Jason just to let everyone
46:04 know because we're getting into earnings season and you know I'm a big dweeb of
46:07 the public markets and I thought it's interesting that as we see sports
46:10 betting and prediction markets come together. It's important to keep in mind
46:14 that also people use these for what we used to use the stock market for which
46:18 was a surprise to me because why would you bother to use a different vehicle
46:22 for that but people seem really into it. >> Yeah. 80% chance they're going to beat
46:25 their quarterly earnings. I think you could really uh do well on these
46:30 prediction markets by looking at the historical data. Almost every company
46:35 tries to sandbag these earnings, right? A lot. They're always trying to
46:38 underestimate them so they can beat them. So beat is almost always the
46:44 default case. But this is an interesting if you were a large shareholder I guess
46:49 that you could use these prediction markets in some way to hedge you know
46:54 just like sometimes people collar stocks or something like that or get a little
46:57 extra juice. How do you think about putting aside just this one specific one
47:01 which you're welcome to comment on. >> Uh what do you think about these
47:06 prediction markets becoming just so um quickly integrated into our lives? I was
47:13 using one to gamble on uh the Emmys and made a little bit of money watching the
47:16 Emmys and then I did I was watching a boxing match. I won again. So obviously
47:20 >> these things are amazing since I've won two times in a row. If I lose then I'll
47:24 have a different opinion but for now my conf Yeah. My confirmation bias is these
47:28 things are awesome. >> They're fantastic. Yeah. Um yeah, you
47:31 know, it's interesting because this is one of those concepts that's existed for
47:35 years and there was crypto projects to do it and other and finally it's working
47:38 between Poly Market and Kelshi. >> Um and really if I look at it, it feels
47:43 like there's two or three big use cases, right? There's basically sports and
47:45 sports betting. >> Y >> and they took what was a really
47:49 complicated interface, not uh they did three things, right? One is suddenly
47:51 it's available in every state when it wasn't before because sports betting was
47:55 regulated on a state-of-state basis. They simplified the UI dramatically,
47:58 right? It used to be what's the spread and the point thing and the this and all
48:01 this complex stuff versus yeah >> who wins yes no >> so they simplified the UI it's like
48:07 swiping left right um and then there's kind of like um uh you know that that
48:11 global liquidity pool where you're suddenly reaching a much larger mass of
48:15 people and so you have way more crowd wisdom and that's sports is a huge area.
48:20 Um, politics when there's the political season is obviously a big area and one
48:24 could argue the stock market is just a prediction market, right? It's the same
48:28 thing. And so I always wonder where does this go? Can you start buying stocks or
48:35 um shares that are basically uh uh collateralized via crypto, you know,
48:38 there's also >> basically tokenized shares um >> uh through these markets at some point
48:45 like do these things converge? Are they completely separate? in which case a lot
48:48 of the financial speculation will continue to be the actual markets and
48:52 this becomes more sports and politics. So I've been kind of wondering like
48:56 where does this whole world go and um what sort of financial instruments
48:59 emerge either on top of this or does it go the other way right the New York
49:03 Stock Exchange announced that it was investing through ICE I think it was $2
49:07 billion in poly market do eventually the stock markets effectively become these
49:11 prediction markets because they are in some sense already. Yeah, I think if you
49:16 want to own the underlying asset, you buy the stock and you uh and if you want
49:23 to speculate uh and get extra juice on more acute, specific timebased moments
49:30 >> options, it's option trading. >> It's like you're doing it in a simple
49:34 way. Yeah. Simplifies options way down. Yeah. And that I think is the real
49:37 innovation here is you've collapsed these really complex speculative things,
49:44 sports betting, options, etc. >> And you've turned them into a yes. No.
49:48 >> Yeah. This is um when you're in the professional gambling circles
49:53 >> um and you're like the bottom rung of it as I am and I watch people try to do
49:58 side bets, what they call string bets. They make them very complicated which
50:03 for the people who are the sharps who are constructing these between the poker
50:07 sessions where they drop them out. It's like those option traders you know
50:10 they're coming up with some really interesting synthetic thing they want to
50:15 do like Michael Bur and the big short or whatever and then there's somebody like
50:19 Goldman trying to make it more complex to try to get an edge and they're all
50:22 just trying to get some little edge on it and you're right the simplification
50:25 of it. >> Yeah. you know, will OpenAI raise money at a billion dollar valuation in the
50:34 next two years. Well, I can't own >> um OpenAI shares. It's too complicated
50:42 you know, it's like, okay, it's a 60% chance. Great. I can make,
50:45 >> you know, whatever percent more on my money, I'll take it.
50:48 >> Yeah. It's a it's a transition from blogs to tweets, but from a financial
50:52 speculation perspective. >> Perfect analogy. Perfect analogy.
50:56 shorter form, fewer calories, you do more of them and they have more people
51:01 do it, easier to everybody can do it, etc. Yeah. >> Okay, let's talk about the Sesame
51:02 What is Sesame? They just emerged from stealth, they raised $250M, and they’re working on AI wearables.
51:06 raising $250 million cuz I don't know what Sesame is. >> Yes, I thought you might not. So, this
51:12 is a company that came out of Stealth back in February. It's done a couple
51:17 things. One, it's built voice first AI, and its first two models that it put out
51:20 earlier this year, Jason, were popular enough that they racked up 5 million
51:24 minutes. But what's interesting about this company and why I wanted to bring
51:27 it up on the show today is they're also working on hardware. They're building
51:30 glasses for you. And I have um an image here from their oh let's call it their
51:35 lab or their office. Uh it doesn't really show us much. But the idea here
51:38 is back to the point about Google Glass being early is that Sesame wants to give
51:42 you some hardware that lets you chat with your personal AI agent or assistant
51:45 or friend or whatever you want to call it on the go. And they're taking the
51:48 hardware approach to it. And I think that explains the amount of capital
51:51 here. Just as background, how did they manage that? Well, they're founded by um
51:56 Brendan Aribe, the co-founder of Oculus and also the former CTO of Ubiquiti 6,
52:00 which was an AR startup. So, they have a lot of that AR um DNA in their veins.
52:08 >> Interesting. So, they worked at Ubiquiti, so they understand AR. Uh $250
52:16 million. And exactly how is the device going to work? Is there a demo of this
52:20 device out? Have they made a video of it? >> Very, very little. the Verge got to try
52:25 it and talked about it a little bit and that's where that image came and I do
52:28 have another one from the same post but it's pretty pretty nent. In fact, this
52:32 company is only putting out the beta of their iOS app now. They're actually
52:36 taking applications for the beta of their iOS app. So, it's very much under
52:40 wraps if you will. But the idea is and Sequoia wrote about this in a piece
52:44 explaining their investment Jason is that it's not just another AI voice chat
52:50 like your chat GPT instance for example. It's supposed to be more of like a
52:53 collaborator with you and with a focus on how it handles speech to make it seem
52:58 natural between you and um maybe your AI. One thing I'm sure you've noticed is
53:02 that if you chat with GPT5 and you make any sound while it's talking, it just
53:06 stops to give you space, which makes it almost impossible to converse. So, I'm
53:10 hoping that what they're doing here is making more natural communication with
53:14 AI, more personalized AI, and then making it functional on the go without
53:17 having to opt in to the Zuckerberg um I would say, you know, spyware empire.
53:21 Jason has concerns about AI wearables that are always recording… Does Elad share these concerns?
53:22 What are your thoughts on wearables, Aladd? And this idea of like it's
53:27 persistently monitoring what you're doing in the world in what could only be
53:32 described as the most creepiest >> Truman Show concept ever, but the
53:36 benefits. >> Seriously, I didn't like it. [laughter] >> I I love The Truman Show in a dystopian
53:43 way. Like >> that's cool. Yeah. I think um Have you guys ever read this book called
53:47 Accelerondo? >> I have not. >> It's like a book. It's like a it's it's
53:53 it's a book with wonderful concepts. I think the book itself is a little bit
53:55 roughly written, but the the basic idea is it's the um >> you know f it's it's like the decades
54:02 leading up into the singularity where you have this technology discontinu
54:07 discontinuation and you know eventually at the point where every 10 years you
54:09 accomplish more than all of human progress then eventually every year you
54:12 accomplish more than everything ever before that etc. and things keep
54:15 accelerating and in it they have this notion of glasses which are effectively a form of external
54:21 memory. They're an agent. They act on your behalf. This book was written maybe
54:24 20 25 years ago. >> Yeah. 200. >> Very sort of looking. Yeah. And um
54:27 >> yeah, >> one of the really cool questions that it
54:31 raises is if so much of your memory and all these agents and all these things
54:35 you do is actually loaded in your device >> and you lose your device and somebody
54:39 else puts it on, how much of their personal of your personality or how much
54:43 of your memories or how much of >> your abilities is it absorbing as you
54:46 sort of augment as a human and you externalize stuff. >> Um and so I always thought that was a
54:50 really interesting almost like conceptual book. Like again there's lots
54:53 of really interesting ideas in that book. Um and this was one of them and is
54:58 one of the earliest versions of you know um glasses and human augmentation and all that kind of
55:04 stuff. So it's this interesting old concept that seems to again people are
55:06 iterating against. >> William Gibson had a similar concept
55:13 with um in his um virtual light the bridge series of like San Francisco
55:18 dystopian future. Um and he had this thing called God's little machine or
55:22 something. It was like a little zeppelin that flew above you in like a
55:26 threequarter shot like the Sims recorded your entire life indexed it and you
55:30 could pull it up and say, "Oh, when did Elad and I first meet or what was it
55:34 like when I, you know, the the first date I went on with my wife and
55:37 >> you could just kind of replay everything and then you have, of course, um,
55:41 Strange Days, the Katherine Bigalow film where you can put on a neural headset
55:45 and >> record your experience, then somebody else could put it on and live stream
55:51 your experience andor relive it. So these ideas are there and it really goes
55:56 to show how much of our consciousness and processing in the world is done
55:59 visually >> and that whole system >> could be uh the unlock for intelligence
56:08 and consciousness in some way is what some of the you know biology uh yeah I'm
56:14 sorry strange days um and it was called the squid superconducting quantum
56:18 interface device. Thank you, Producer Claude, for [laughter] coming to the
56:23 rescue here. Um, I do have some concerns about these things or being always on
56:27 and recording everybody all the time. >> I don't know if I'm going to win that
56:32 battle, though. You have concerns about this? No, >> you know, I'm in general a very paranoid
56:39 person, so I always have concerns. Um, but I think that uh it just comes back
56:43 to utility, right? What I think one thing that's been proven to us is um
56:47 people fluctuate between willing to sacrifice privacy in order to get access
56:50 to more stuff. That was Google Gmail, right? There was all these concerns
56:54 around Google reading your email reading it not really but a machine reading it
56:58 to generate ads and that was an issue but then you know hundreds of millions
57:00 of people adopted Gmail and then Facebook similarly people had concerns
57:03 about their own personal data and privacy and I think for a while um
57:08 Facebook actually was doing fine on it and probably what happened is because I
57:11 you know I sold my first company to Twitter and that was around 2009 and so
57:16 this was right as Facebook started paying attention to Twitter and I always
57:20 wonder if there's some alternate timeline or history or something else
57:22 where instead of trying to go after Twitter and media and public content,
57:26 >> Facebook leaned into its strengths because with the time it was a very
57:29 trusted privacy brand, right? It was your identity and it was your friends
57:32 and your friend group >> and it was viewed very differently. And
57:36 what if they had gone into banking and payments? What if they gone into
57:39 healthcare? What if they gone into all the things that were privacy native
57:42 almost what Apple ended up doing? >> That's a very interesting alternative
57:45 timeline for Facebook. >> Then maybe they will be the trusted
57:48 identity. Yeah. I don't know. It's it's an interesting concept, right, of like
57:51 they decided to go after Twitter. What if they decided to go after banking
57:55 >> at that moment in time when they were very trusted as like a you know, they
57:58 had all your stuff, but that was fine, right? Because it was just your friends
58:00 seeing it. >> It was your it was your social graph. It
58:05 was closed. They weren't trying to serve you ads. And Zuckerberg, in fact, hated
58:09 advertising. >> That was like he thought it was terrible. He didn't want to have ads on
58:15 the service until at some point somebody told him, "Well, you can make it free
58:18 and get billions of people to use it." >> And he was like, "Okay, fine." And then
58:21 Charles Samberg comes and the rest is >> and and this could have been an
58:24 alternate timeline. Maybe you would monetize through payments. And this has
58:26 happened a few times, right? I've talked to some of the people who worked on the
58:29 early browsers >> and it was the same thing. They're like,
58:31 "Is it going to be microp payments or ads?" >> And they ended up going down the ads
58:35 route, but there was an alternative universe where it was like microp
58:38 payments, right? And you you'd pay a >> maybe not kind of. I mean the technology
58:42 wasn't there [laughter] as the founder of a shortlived microp payment startup
58:47 back in 200 whatever you know I mean credit cards just weren't weren't there
58:50 I think the idea was good but I think we needed essentially >> you had one you had a microp payments
58:54 startup >> oh you did not you startup >> yeah it's called it's called contenture
59:00 we founded it at techrunch wrote about it we launched got customers and then we
59:03 failed and I had to go back to college >> oh okay wow you you were just too early
59:09 I kind of feel like yeah they were doing the brave attention token for a while.
59:12 That didn't take long. >> Oh, that's right. Yeah. >> Uh BAT, there's been a couple shots at
59:18 this and you know, one of the things I've learned is when we just talked
59:21 about it with, >> you know, Sergey being so ahead of the
59:26 curve that he was too ahead of the curve and then probably some
59:30 >> accounting wonks were like, "We're seven years into this, Sergey. Like, let's
59:33 pull the plug." And he was probably like, "All right, fine."
59:36 >> Um when if he had just stayed till year 11, it would have been like, "Oh, okay.
59:40 Now we own low earth orbit satellites. We had Starlink or >> which was Whimo, right? Yeah, which was
59:44 Whimo. But yeah, they could have had a couple others. So again, very forward
59:48 thinking company. I think they don't get enough credit for that.
59:51 >> I think this is where like Sergey and Larry just need to say we will be
59:56 willing to lose money on projects for 20 years, not just seven. And we don't care
60:01 what you think. Remember they were doing Google Fiber >> and I think that one got deprecated as
60:06 well. That would have been fascinating if they kept going along and pulled
60:09 [clears throat] that string because then you'd have fiber plus
60:12 >> um >> satellites, you'd have fiber, we'd have
60:16 flying cars, basically everything would be done. >> Well, you'd be out of business. We
60:20 couldn't invest anymore. >> Well, exactly. Be like Google's world.
60:24 But also that was like Larry had to go off balance sheet and use his personal
60:28 balance sheet to do >> he did three of them, I think. >> Kittyhawk, right?
60:32 >> Kittyhawk. Was he also in Archer? I think he might have also invested in
60:35 Archer. Yeah, I'm not sure. >> He was in at least two of them. Um, just
60:41 fascinating how close being too early is. But I I'm always encouraging
60:44 founders to to look in that startup graveyard. I think actually micro
60:49 payments as silly as it sounds built into a social network, a browser for
60:54 content. You know, this might be like Substack or, you know, a YouTube
60:58 competitor might be able to do very well with this where you could set your
61:02 >> what if you could set your browser to just say if I watch more than one minute
61:06 of a YouTube video, >> send the person one penny >> and uh if I watch, you know, a penny for
61:15 every second I watch of anything. And if I, you know, if I stay on a new a
61:19 document and I scroll down to the end, yeah, give them a penny. There's a lot
61:22 of people who might buy into that if there was some reward for it. I don't
61:26 know what that reward is. >> Well, I mean that was kind of what we
61:30 were trying to build, Jason. So, it was back then you couldn't do one penny. So,
61:33 you would put in $5 to your content account and then as you went around the
61:36 internet, we would hand out the money to people that you visited that had our
61:40 integration and people kind of liked it, but it was hard to get any sort of
61:42 momentum. >> Think about stable coins. Like, I wonder if coins are the unlock here. What do
61:48 you think about stable coins? >> Yeah. >> Yeah. I've been uh long interested in
61:53 them. I mean uh it's interesting to see some of these second order effects of
61:56 those now, right? In terms of the biggest holders of treasuries in some
62:00 cases amongst the biggest holders or are stable coin companies.
62:02 >> Yeah. >> So it's it's been fascinating to see
62:05 both their growth. I remember when Tether came out initially everybody
62:08 thought it was super sketchy >> are very uncertain in terms what the
62:10 assets are actually holding against the money. Yeah. >> And now obviously they're it seems like
62:14 they're fully collateralized between a mix of like gold, Bitcoin, and um
62:17 treasuries. >> They've done a great job cleaning it up.
62:21 Yeah. Yeah. So I, you know, it seems like it's going to be a really central
62:25 thing in general. And I think one of the things that people have underappreciated
62:28 about crypto, particularly people in the US, is the degree to which globally, for
62:32 example, that people don't have access to dollars. >> Yeah.
62:34 >> Because you're not allowed to by your government, and this is a way to buy
62:37 dollars, and this is a way to do simple transactions. Like all the stuff that we
62:40 take for granted here. >> But it also shows the path forward with
62:45 tokenizing shares and token, you know, things we've been talking about for
62:49 years. it feels like are finally going to happen. So, it's a it's a very
62:52 exciting era, I feel. >> Here we go. Tether is according
62:56 [clears throat] to producer Claude. Tether 127 billion in treasuries as of
63:02 Q2 2025. >> Number 18 global ranking ahead of South Korea and Circle 18 billion is in the
63:09 top 30. >> Yeah. >> Yeah. It's very under discussed how many
63:13 treasuries are actually held by crypto companies now. So one of the biggest
1:03:17 The crypto industry is now one of the largest purchasers of US government debt… what does that mean? Who owns who?
63:17 buyers of US government debt is the crypto industry. Like think about that.
63:20 What are the implications of that in the long run? If crypto gets 10 times bigger
63:26 and that 120 billion goes to 1.2 trillion. >> Mhm. >> What does that mean in terms of US
63:32 government's relationship to crypto? >> Think about that.
63:34 >> Yeah. >> Super interesting. >> Who's your daddy? It's kind of like the
63:37 China relationship. >> It's Tether. It's Tether. Tether.
63:39 >> Tether's your daddy. >> Paul is your daddy, I guess, in that
63:43 case. Well, and the interesting thing about Tether is um yeah, they did have
63:49 that really um spicy I'll use the word spicy. I don't want to get sued here,
63:52 but they were banned in a lot of regions. A lot of like regional action
63:55 taken against them. A lot of questions about those reserves. And they have gone
64:01 >> like on a five-year process of cleaning all that up. And there's like I think
64:06 they have according to the bill, the stablecoin bill that Sax was involved in
64:10 or talked about on the pod a bunch of times. >> I think they have a three-year window to
64:14 clean everything up in order to participate in the United States.
64:19 >> Uh so circles got the clear lane along with I guess their partnership with
64:23 Coinbase, but they have to Tether >> is going to clean up and the the guy
64:26 Paulo is his name. >> Paulo. Yes. who was like living in some
64:31 non, you know, the speculation was like none of the executives were available
64:34 and they were living in jurisdictions where you couldn't be deported
64:37 >> or picked up by a red notice. I don't know how much is that true. There were a
64:41 lot of tether truthers out there, you know, speculating. Um, but them figuring
64:48 out how to go legit if these stories of illegitimacy were in any way true is
64:52 like a story for the ages. That's going to be the next Sto movie, I think.
64:57 >> Yeah, it's pretty good. Um, I think one of the most
65:00 underhyped players in crypto right now perhaps is Stripe, >> who I think have been doing amazing
65:06 things in terms of um, stable coins and bridges and all the rest of it. And, you
65:10 know, I first invested in Stripe >> in 2010, I think. So, it's been kind of
65:16 a long road, but um, what an amazing >> next to act. I mean, obviously the
65:19 payment stuff is going to keep going and the SAS stuff will keep going, but I
65:22 think what they're doing in crypto is super interesting right now. They're
65:25 never going to go public and everything else. >> No, they definitely have. They're never
65:28 going to go public. This is my new prediction. >> Poly market.
65:33 >> I just >> infinite duration uh >> perpetual perpetual options on
65:38 hyperlquid. >> I Well, here's the thing. If you have that fluid of a market, if you're an
65:47 expert in transactions and commerce, what's to say they couldn't build and
65:51 you know the infrastructure to let people who are qualified purchasers
65:56 trade in and out of their private company and then slowly buy back all
65:58 their shares >> based on massive profitability and then
66:03 have a blackbox >> company that nobody understands what's
66:06 going on there. It's a tremendous advantage. And the Coulson brothers are
66:12 uh unique in their perspective of the world. I don't think they're
66:18 particularly driven by some outrageous um you know personal net worth on a
66:23 public market and the traditional accuchant of being an entrepreneur. They
66:28 seem very missiondriven and have very specific >> principled things they want to work on,
66:33 you know, like their think tanks, their science think tank, etc. So I
66:37 >> that could be amazing. I mean, I'm trying to think of the equivalent of
66:41 what's the largest private entity. >> Yeah, there's a couple of them. There's
66:45 like Coke Industries, there's Cargill. There's a bunch of these. Yeah, it's
66:49 kind of interesting. Bloomberg, right? >> Yes. >> Tons of cash and it's private. So,
66:57 >> I think people under um appreciate espec you know, Europe like a lot of the
67:00 market cap is actually these uh middle shot private companies in Germany for
67:04 example, >> these family businesses. VTOL Swissbased Dutch multinational energy and commodity
67:11 trading company. >> Yeah. Revenue $400 billion dollar in
67:18 >> Coke. Isn't Coke like >> Coke is number five? 125 billion.
67:23 >> Cargill should be on there. Like there's a bunch of these.
67:26 >> Cargill number four 160. >> Yeah. >> I'm trying to find a tech one. Huawei
67:31 number 10. I didn't know that they were private. >> Yeah. I mean I guess bite dance is a
67:35 private. Vance 14. >> That company. >> Yeah. >> Hey, a lot on the uh on the stripe
67:42 stablecoin point cuz I agree very much that they're they're leading. They
67:46 bought uh was it bridge for 1.1 billion and they just announced tempo their
67:50 their own kind of blockchain or L1 for for stable coin payments? I'm not asking
67:54 for internal information here, but like what do you think they're trying to do
67:57 with that? Is that just to make a better foundation, better rails for stables to
68:01 fit into the kind of existing e-commerce world or are they trying to solve a more
68:04 >> I I don't know fundamental issue with stable coins today that is not well
68:08 served by existing L1's and L2s. >> Yeah, I you know I think it's best to
68:12 just look at what the company says about all this stuff. Um >> because I I don't want to over interpret
68:17 or misinterpret what they're trying to do. Um I think fundamentally if I look
68:21 at what they've been building as a company over the years it's really this
68:26 focus on providing uh a piece of global infrastructure like Patrick used to talk
68:28 about it almost like containers ships and that enabled global commerce
68:32 >> for the world and so I think they're very missiondriven around you know this
68:36 interconnected web of commerce isn't that good for the world and how do we
68:39 help make it really easy for people and in particular developers to build things
68:44 against that so >> absolutely extraordinary what else do we
68:47 have on the docket Uh, we got maybe one more story left in us. What do we got?
68:51 >> Well, because we have a lot here, why don't we talk a little bit about the uh
1:08:53 Anthropic responded to JCal’s Bestie David Sacks… Is Dario Amodei a doomer? Fearmongering?
68:55 the the fracus, Jason, with Anthropic and uh your bud David Sax.
68:59 >> Oo, that's spicy. >> Yeah, I know. >> We also have the OpenAI browser. Less
69:04 spicy. Let's go for the spice. >> Okay, let me just pull that up real
69:08 quick. I got to find >> I have some thoughts on it. >> I'm shocked. I'm shocked that you do.
69:14 >> You might not believe this, but I have opinions. I know for better or worse.
69:18 >> It's good. It's good. >> All right. So, uh, Anthropics Dario
69:24 Emmed wrote a public post this week responding to his critics. But to
69:27 understand what's going on, we have to go back in time a little bit to the one
69:31 big beautiful bill. Jason, we discussed here on the show that there was a
69:35 provision that eventually didn't make it in, but that would have banned states
69:38 from having state level AI regulations for a period of 10 years. Ended up
69:42 getting voted down 99 to1 in the Senate. Fast forward to recently, Jack Clark,
69:46 another co-founder of Open, sorry, of Enthropic, wrote a post, a summary of a
69:50 talk that he gave in which he said that he thinks other stakeholders in the
69:53 world should have a bit of a say in how we build AI as we build it. He's not a
69:57 doomer. He's definitely an optimist, but he thinks that we should listen to other
70:01 people. Uh, and then your friend David Sax comes in and he says that quote,
70:04 "Anthropic is running a sophisticated regulatory capture strategy based on
70:07 fear-mongering. [snorts] It is principally responsible for the state
70:10 regulatory frenzy that is damaging startups." Fast forward through some
70:14 back and forth and other people weighing in and we got Daario's um open letter
70:18 essentially saying, "Hey, what the hell's wrong with you guys? We love
70:21 startups. We work with them. We're in favor of a federal standard." And this
70:25 has become such a big deal, such a back and forth that CNBC is covering it. And
70:30 so Jason, I'm I'm incredibly curious why we have left the era of founder
70:33 friendliness and we're now seeing venture capitalists attack some of the
70:36 most successful companies in the history of startups. you know um this industry
70:43 left its boutique phase when Facebook hit scale. Uh Facebook just had far far
70:50 too much impact on society, elections, our children, Instagram. I think that
70:56 was the end of the innocence. Um, just if I were to pick, you know, a
71:01 demarcation point a lot of when >> we became so big that we couldn't be
71:06 looked at as the weirdos in California building cool stuff and then it became,
71:10 oh, how are these things damaging society? We should be questioning
71:14 everything about AI. We just did today talking about Amazon and job
71:17 displacement. We should be thinking about kids using these products. We
71:21 should be thinking about people getting oneshotted and getting, you know, and we
71:25 should be thinking about the truth and the search for truth. I think all these
71:28 things are valid discussions. I think what's at stake here is there is covert
71:34 um I don't know lobbyists uh and covert political actions to try
71:41 to regulate things. And though that does make me nervous. I just want to see all
71:46 the evidence of it um if that's actually happening or not. And I I I told Sax
71:51 this. He wasn't too pleased with me, Elad. But if this is the most important
71:57 technology of our lifetime, why wouldn't states have a right in the format of the
72:04 United States of America to, you know, chime in on it? You know, if they kids
72:09 access to it or how it's used in healthcare, how it's used in terms of
72:13 job displacement. What's your take on all this? Yeah, I think um at at this point in
72:19 time and I think industries evolve over time and the nature of how you regulate
72:22 those industries should evolve with it. But early on in industries, it's a bit
72:27 of a disaster to let each incremental state chime in >> on how to regulate it because
72:31 fundamentally there's a handful of states that effectively by fiat govern
72:35 what the rest of the country does, whether those the rest of the country
72:38 adopts things or not. It's it's almost the flip. Why should California be able
72:42 to govern AI for everyone in the US which effectively almost happened
72:46 through the Scott Weiner bill. So I actually view it on the other side of
72:49 this one where I think um the states should not intervene for now. It should
72:53 be federally mandated. There are both national security reasons to do that.
72:57 There are national competitive dynamics that drive that. So, uh, you know, my
73:03 view would be, um, and and this literally almost happened, right?
73:06 California was going to >> try and regulate how the rest of the
73:09 country deals with AI, which is a really bad idea. So, >> and it was based on the number the
73:15 amount of compute you have as if that is some sort of >> Well, the question is why should
73:21 California regulate it for at all a and then b for everyone else, which is fact.
73:24 I mean, it would have been national, which means it would have been global.
73:26 So, should California have the right to set national or global AI policy
73:30 effectively? >> Yeah. And the car example is a good one.
73:34 And so, that's one where I happen to be in favor of it. So, when you look at
73:38 these, it's always um >> uh it may have been the right thing to
73:41 do. The question is, should California have been able to do it? In other words,
73:45 >> correct. Yes. I was in favor of it. Yeah. >> Yeah. Yeah. It's like, let's separate
73:49 the do we agree with the thing? >> Yes. >> From how should that thing get done,
73:52 right? And I think those are two separate things. So, >> yeah. Yeah, and that's what I've been
73:56 trying to parse in my head because I, you know, yeah, we should have higher
74:01 gas standards and who cares if cars cost a little more. We we, you know, having
74:05 smog over cities and people losing, you know, months to years off their life
74:09 and, you know, asthma and all this kind of stuff. Yeah, it's obviously a tragedy
74:12 of the commons. It's a good thing that California did that in my mind, but
74:17 should Texas be forced to not buy the trucks they want that have the hauling
74:21 power and the price performance ratio that they want? I I picked Texas as a
74:26 place to live because I like less regulation. Um, and so it really does
74:31 come down to the operating system of America where the states are supposed to
74:35 have some sovereignty but then you have to super oppose it against well if we
74:41 are in fact in a global race with China, can we can we afford to slow ourselves
74:47 down? And that's like a very interesting this is >> probably there's a global race with
74:50 China and then there's the um there's the political side there's the the
74:54 military side of that there's the economic side of that there's the
74:58 political and cultural side of that >> and so I I disagree as well with hey we
75:01 should have every country on the planet weigh in on how AI models should get
75:05 built in the US. I think each country should have its own sovereign AI model
75:09 >> and there's lots of ways to accomplish that right like if I was running a
75:12 country I'd be very interested in that. I want my cultural values represented in
75:15 local models and things like that. So I >> sure >> agree with that. But the flip side of it
75:19 is um >> if anything perhaps the last few decades have shown us that perhaps we don't need
75:25 to solicit everybody's opinion on everything >> both in the context of a company and the
75:29 context of a country and the you know and so this is almost a throwback to
75:33 things that uh haven't really been working. It feels >> Yeah. If you were to put this against
75:39 high-speed trains, Alex, like there's a reason why we don't have them in America
75:43 because everybody gets to chime in on the highspeed train and then the
75:48 high-speed train, you know, is like, you know, it's [laughter]
75:50 in other countries, they're like, high-speed train has to be a straight
75:54 line, >> or else it can't go fast. And you know, >> but in China, there's been a lot of
76:01 studies discussing how both province and central planning has led to a dramatic
76:05 overbuild of economically inefficient highspeed rail. And so, we can all give
76:09 a thumbs up to the Shanghai Beijing line. Why not? But when you see the what
76:14 is it eight pillars and eight cross beams, there's some model that I'm
76:17 translating poorly here for how to build highspeed rail in China. And it's led to
76:23 a lot of misplaced spend. And so I agree that we should >> of that sp like that's often also
76:27 corruption. That's a bunch of stuff. It's not just >> but if you want centralized power all if
76:33 you don't want people to have say you're going to end up with a central power.
76:37 >> Well really what the question is is what are the things that should be regulated
76:40 by what level of government? M [clears throat] >> and that's that is something that has
76:44 been established and adjudicated in the US for >> centuries now right and that this falls
76:49 into line of those things which is >> how much should states be able to govern
76:56 >> on their own overall technology progress for the US and for the globe and that's
77:00 effectively what this issue really is right >> interstate commerce
77:04 >> not just interstate commerce it's effective de facto regulation of the
77:08 entire country >> right no I'm saying that because it's
77:11 that it should fall prey to that and therefore it could be limited in that
77:13 way. >> Exactly. Exactly. >> But what's what's interesting though is
77:17 we're talking almost about the wrong the wrong foe here because I think this is
77:21 why anthropic is, you know, being dragged into a bit of the mud here. But
77:25 the the the provision to bl to block state level AI rex was voted down nearly
77:29 unanimously in the Senate and partially because Republicans from the south
77:34 wanted to protect local industries. So, how do we get Congress to listen to this
77:38 perspective and do something intelligent and pass something that will preclude or
77:42 maybe just do better than state level regulations? Because Daario said in his
77:47 letter that they disagreed with the idea because um a blocked action without
77:51 offering a federal alternative. So, how do we get to one? How do we fix Congress?
77:58 >> I mean, there is there is this >> um I I I'd say three things on it. One
78:02 is I tend to stay away from politics. I'm the least uh cogent person to ask
78:05 about >> how to fix Congress or how to change, you know, like I'm that's not the world
78:11 that I tend to spend time in. Um, you know, I do think there's a number of
78:14 different approaches that we've taken in the past or historically
78:20 to federally mandate things. And one route is executive orders where they can
78:24 be applied. The other route has been acts of congress. You know, there's
78:26 three or four routes that are just standard. And the question is what best
78:31 applies here? Then I think it's back there's also the meta issue of where do
78:35 states cross the boundaries in terms of the areas that probably shouldn't apply
78:39 to them and is that a legal framework is that something else that already exists
78:45 and how do we interpret um you know the existing legal structures of the country
78:49 relative to some of these topics >> I think that's the granular piece that
78:53 you're talking about a lot I think is the key piece of this which uh downside
79:00 what concern do we have over AI that is not addressed by an existing law. If you
79:04 were to go in an AI and you ask it right now, how do I build a chemical weapon?
79:08 How do I build a biological weapon? It says, hey, I'm not going to do that.
79:11 It's already been solved by the teams. They've >> self-regulated.
1:19:12 Why Jason thinks AI companies need to self-regulate
79:17 >> So, okay, what's the next piece? Oh, um, how do we deal with the, you know, I
79:20 don't know, the termination >> the bioweapons stuff, the bioweapon
79:24 stuff is so overstated and I feel like that's the >> way that some of these companies have
79:31 been trying to uh cause distress and fear and regulators and others because
79:36 the reality is all the protocols for um building a a virus you can release
79:41 etc are on the web. >> Yeah. >> I mean I worked as a biologist for
79:44 almost a decade, right? and all these love protocols are out there. Um the
79:49 real issue is buying the equipment, using it, actually doing things in an
79:52 intelligent manner, ordering all the logos you'd need, etc. And you know, so
79:58 the extra information content in my opinion isn't high >> and the hard part is the physical world
80:03 stuff. And so whenever people raise that as an example, >> I view it as kind of a red herring where
80:09 where it's purposefully raised in attempt to cause fear in people so
80:12 they'll react and regulate something. So I do think those are bad bad examples
80:17 that people tend to use in general and it's purposeful. I think
80:19 >> it's per I agree and I always ask people tell me the next one tell me the next
80:24 one. Let's just get granular here for a minute. I have one like IP protection. I
80:29 think that that should be done. Uh that should be self-regulated by the industry
80:33 which they're starting to do. They're starting to say, "Hey,
80:37 >> even if you don't agree with how Sam's doing it at OpenAI," he's saying with
80:39 Sora, >> "Hey, if you're like, you know, the estate of Elvis or Martin Luther King,
80:47 we'll be gracious and turn off the ability to make Martin Luther King
80:50 saying inappropriate things." And they did. >> They didn't have to go the legal route.
80:55 Sam said, "We're good people. We're going to be aggressive in letting people
80:58 do things, but we're also going to be reasonable in turning them off."
81:01 >> I disagree with the approach. I think they should always get permission. But
81:04 okay, fine. He took a different route. They'll get the legal exposure. But
81:09 >> yeah, this is um a situation where we have to get into each one. I was just
81:14 thinking when you were talking about um you know, bombs and I was like, remember
81:17 fertilizer bombs? Everybody was like, "Oh my god, you know, the internet's
81:21 causing this, the tour network, uh you know, dark web, everybody knows." And
81:24 it's like, you still got to get the fertilizer. Every piece of every bag of
81:31 fertilizer has IDs in it. Like in the chips from what I understand, there's
81:35 like some secret way that they watermark every bag of fertilizer. So if you blow
81:39 something up and then if you buy them, you need to register and they've got
81:42 some, >> you know, I mean, who knows how the government is implementing this after
81:50 the 911 because the first before 911, they tried to blow up the World Trade
81:52 Center with a fertilizer bomb. They succeeded in blowing up Oklahoma City.
81:56 Two different groups of terrorists, domestic and international.
81:59 >> They they solved this problem. I bet you they have like ways of tracking
82:03 fertilizer where like the second the person leaves the fertilizer store,
82:09 there's like a camera on their license plate. They're being tracked by a drone.
82:13 Like the government took this seriously. >> There's so many like simple solutions to
82:16 this. So >> yeah, that's why I stopped fertilizing my my garden. I was worried that uh I
82:21 get tracked down. >> Yeah. So you you you laugh, but the only
82:25 time I've actually gotten pulled out at the airport and put into a small room
82:28 was after going home to Oregon to my parents house and I was walking around
82:33 the backyard a lot and I popped a uh some sort of test at TSA for possible
82:36 fertilizer. >> That happened too >> and they like
82:39 >> I've never been treated as a criminal quite like that. They're like, "Is there
82:42 anything in here that's explosive?" I'm like, "I don't know. weed, but
82:46 please don't find that." And it was so Oh, sorry. We're live. Uh edit that out,
82:51 I guess. Um, but it was just it was so surreal, but it it can catch you. And
82:55 >> they uh they literally asked me like, "Have you been doing gardening?" I was
82:59 like, "No, how's that relevant?" I didn't [laughter] know what they were
83:00 talking about. >> Like, I've been helping startups grow.
83:05 >> I don't get it. >> by spreading manure. >> Yeah, exactly.
83:10 >> Uh, all right. Listen, this has been another amazing episode of This Week in
83:16 Startup Salon. Thank you for joining us today. What a great guest. Uh, where can
83:19 people find you? I mean, you got a Twitter account. You're pretty active.
83:21 Yeah. >> Yeah, Twitter. They can find me uh at a log. And pleasure to be on. Thanks for
83:25 including me. >> All right. Uh, and we'll see you all next time on this week in Startups.