Transcript
0:00 AGI's been here. The singularity has started and now we are at least four steps in, four human steps in. >> But if they're competing with Uber and they're able to charge even a $150 a mile, we think the Cyber Cap could cost around 25 cents a mile to operate. So many cool things happened this week. Cyber Cab AGI Millennium problems being solved and Twitter spats that follow. another researcher claiming that AI is going to kill everyone and quitting their job. And then of course the Apple debut. Are they raising the stakes in AI or are they folding? let's start with Cyber Cab. Brett, what do you want to what do you want to take away from this event?
0:46 >> Well, so Cyber Cap, we have other content. Tasha is talking about Cyber Cab here, but the the launch event is, hey, this is a two seat, zero steering wheel, zero pedals vehicle that they are, you know, putting out into the world and it's running around at on a per mile basis, half the cost of an Uber. which is, you know, we think it'll actually probably get even cheaper over time. and the cyber cab is clearly Tesla's push into the robo taxi future. This is a vehicle that was designed specifically for robo taxiing around. Meaning you can make all kinds of neat engineering optimizations if you're going to try to reduce cost per mile as aggressively as possible for Robocat taxi. Now cost per mile is different from price per mile. So there's this controversy that's out there of why on earth would Tesla sell these cyber caps? Why would they let other people have them if they're going to be so lucrative to the business? Why not own them themselves?
1:49 And I want to clear this up for people. So, the way this is going to work is Tesla's going to charge a platform fee just like Uber does. The platform fee will be some percentage cut of the gross revenue that people go. Tesla will want the people who buy Cyber Cabs to get a good return on capital and they'll own some themselves and they can control the return on capital that people people get by changing that platform fee. But if they're competing with Uber and they're able to charge even a$150 a mile, we think the Cyber Cab could cost around 25 cents a mile to operate. So there's a$1.25 per mile that Tesla can keep some for itself and give others to CyberC owners. and so if you are operating the vehicle at 25 cents a mile and you get 5 cents a mile on top of that, that's you know a 20% return on every mile that it's operated. A platform asset owner would be super happy with that kind of return.
2:50 Meanwhile, that still lives a leaves a$120 a mile for Tesla to embed into its own profitability. So I think they'll manipulate the mix over time. But that's the solution to the puzzle of like why is Tesla selling these cyber cabs? I think they want people to be able to operate these assets particularly holding them for their own use during non-arch periods but then getting induced in during surge periods to provide them to the system as a whole. And I think these things are going to be everywhere and they're going to make sure that they appear everywhere. Tesla is going to treat those asset owners as you know favored customers who can get an abnormally high return on their investment because there's so much lucrative economics in the business generally and Tesla can be so profitable on the back end of facilitating cyber cap miles. So that's kind of like my take on why they're selling them. Sam, what did you think about it and how stuff can happen? AI infrastructure as a service- like returns, but just other infrastructure, transportation infrastructure as a service returns. I thought it was extremely interesting to hear all about the unboxed process here. You know, over 6 miles per kilowatt hour of energy. So, a hyperefficient vehicle. You know, back in the day, we we showed the progress and where we thought it was going to get. And I remember us having a conversation saying, "Is it is it really going to get that high? That seems seems maybe it's too much." And here here we are. no brake fluid. So total electric actuation of the brakes.
4:32 Another design choice to reduce complexity in manufacturing. I think the cost is going to surprise people, right? Obviously, there's going to be a ramp up that's going to be more painful than even Elon expects as is customary. and then the end result is going to be more mind-blowing than people can imagine. I think there's interest. Yeah, go ahead. >> I think there's there's this, you know, just just imagine what you can do to cost optimize a vehicle per mile when you're designing to cyber cab and how when you're competing against a company like Whimo that is taking offtheshelf products and then super gluing sensors onto them kind of the difference in optimizations you can make. So the cyber cab doesn't have to have like a consumer acceptable range. It has to have a range that is just sufficient to meet the intended operating conditions over the course of a day before getting back to charging. you can, it doesn't have to be able to go, you know, 60 to 80 mph with any PEP at all. because that's not like the operating conditions it's going to be typically doing. So, you can change the power electronics to optimize for like how these things are actually going to operate. and kind of because they're designing from scratch, they can design and manufacture this two seat vehicle in a hyperefficient high throughput way. Net the if you look at the cost per mile economics of a of running a cyber a robo taxi network, the biggest biggest input to those economics is the actual cost of the vehicle.
6:22 because that also affects your maintenance. and operating costs. Then it's also the efficiency of the vehicle on an electric basis and your ability to charge it like those two crossed. And so Tesla has here the optimal vehicle. They already have much better charging infrastructure than any of the marginal competitors out there. I think that they will be able to should be able to to price at a level that is on the order of a dollar a mile or less.
6:54 and that might be a price where Whimo literally cannot get gross profitable at that price point given the cost of their sensors, their cost of manufacturer, everything else. So it could be that kind of Tesla, you know, prices enough to keep Whimo limping along in the business for a while and then just turns the screws on them as they accelerate and scale. And then I do want to just a a meta point here which is what was the event, right? It's like the cyber cab was already announced.
7:32 They didn't launch they I guess they launched some into the road. So there's that. That's exciting. This was not a widely attended event. It was a it was a hyper exclusive crowd that attended. I think you know maybe the you could say it was a unboxing manufacturer release which is arguably one of the most important elements here. I think it's super exciting. I think of the adage never sell a dull or boring market. And I think Tesla became boring and that was pretty clear, right? I think the last earnings call, the last two I would say, all of the Tesla influencers got bored and they were distracted by SpaceX AI and all of the exciting things happening. And just as Cyber Cab is getting ready to, you know, transform transportation entirely, people are saying, "Okay, you know, who knows when this is going to happen?
8:34 Another year, another quarter." Like, you know, I'm I'm looking elsewhere. And I think it is true that never ignore what seems boring. There's tons of stuff going on. I think it is going to be extremely exciting and happen faster. It's happened way slower than everyone expected and now it will happen way faster than people expect. >> Yeah. I mean and I do think it's they are capability gated on releasing the next software release which we think is going to happen end of year. if you look I part of the reason to do a more less flashy event is because they're not yet ready to accept all the masses and to drive demand into the app and and really provide this service at scale. If you look at kind of app usage trends just on the strength of the cyber cab launch, you know, they've gone from never having done above, you know, 25k active users on the robo taxi app and now they're sitting at 90,000. So they four and a half timed their active user base on the strength of a very low PR event. I I think partly it's because now you can select the cyber cab in the app and you get, you know, a price that I think it's 40% cheaper than riding in a Model Y and half as half the cost of driving riding in an Uber. So, you know, we could begin like we're going to begin testing demand elasticity at these lower price points. and and you if you're in Austin, you can flag a cyber cab and soon you'll be able to do it everywhere. I can't believe I can't wait for these things to be available in Los Angeles so that I can ride around in them rather than driving myself or having my Tesla drive myself but still drive having something else drive me you know that I'm not responsible for finding parking for at the very least.
10:36 >> Yeah. >> And it's like 500 of these on the road in a city will seem like they are everywhere just because of it's like this was true in San Francisco. There were not that many wayos and yet you could always look out and you'll be like, "Oh, there's a Whimo." One, they're distinct looking so you'll notice them. Two, the utilization is going to be, you know, five to 10x more than a car that gets parked and hidden for most of the day. So, I think it will strike people pretty rapidly once they start rolling out.
11:11 >> Yeah. >> All right. >> Yes. >> A AGI Astra >> AGI is here. >> AGI is here. AGI's been here. We the singularity has started and now we are at least four steps in four human steps in Brett Astra. >> There's this idea. Yeah. Yeah. Yeah. So, OpenAI launched Astra which is their newest version of or their newest build pre-trained build of their model set and it is wildly performant. It's so performant that they threw 88 hours worth of its work at solving one of the Millennium problems, which were a set of math problems set out with a million-doll bounty if you solved it.
11:55 and the the reason to do it, the explicit reason to set out these problems was because these were problems that, you know, computers wouldn't be able to solve. So, whoops, that was wrong. and so and to inspire mathematicians to pursue these big, thorny, complex proofs. and Astra has solved it's called Navier Stokes which is a problem in kind of modeling fluid flows. If you've ever done system dynamics I think is where I took had to deal with Navier Stokes. you actually use kind of like call it cludgy implementations of it. the the so these systems are wildly wildly wildly performant. There's this idea called recursive self-improvement that people talk about. Are are we going to enter a land of recursive self-improvement and OpenAI put out a blog on their own internal use of these models? And I think it's worth broadly considering that one recursive self-improvement is not somewhere we're at yet where you set the model to improve itself and it goes off and improves itself. instead and I think for a very long time we'll be in this domain where it's like humanass assisted like accelerated improvement where you you don't like there is there are always going to be like corners and byways that these things get stuck into and you will need human oversight to kind of like coordinate them to point them in the right direction to unstick them and that's where we're going to continue to live for some time it's just the frequency of human oversight is going to reduce over time but for open AI and anthrop IC and you know maybe SpaceX as it catches up. The fact that they internally have access to these tools well ahead of when anybody else has access to the tools because the tools are still in testing and they're still like making sure the tools are safe to release gives them this huge competitive advantage in capital efficiency relative to the entire software space and all of the other like non-frontier AI developing companies. and and so there's this, you know, idea in the industry that you have to eat your own dog food, meaning you have to accept your cludgy, terrible products because that's how you'll be forced to improve upon them. Here you have to eat your own. You have to eat your own.
14:12 >> Yes. You have to you have to eat the most expensive food. You have to eat like at a Michelin threear restaurant and every other schmo is out there trying to like eat Burger King and survive, you know. And so I do think that there's people talk about what are the moes in AI and how is it that these companies, you know, could possibly carve out economic advantage over time. Well, if the primary moat right now is your ability to accelerate, it's basically like actions per minute cross with your ELO score. Well, here they have they can parallelize kind of their existing internal tooling stack which is much better than you know aside from maybe Anthropic and maybe SpaceX anybody in the world and continue to accelerate. So, one of the things I take away not just from Astra but from their own internal blogs on on use of their tooling is that I think the closed model ecosystems are actually going to blow away the openw weight models very contrary to people's assertions and expectations as in the time gap between them is going to elongate. maybe now judged by how they're packaged performant in the hands of users. And so kind of like Aster itself is really impressive, but it's also impressive because it's good at computer use. You need to have computer use like natively embedded in the app in which you're using it. and so it becomes a more general purpose tool.
15:43 their ability to develop that app quickly and and make it performant is predicated on their ability to point these tools at that app and say make it better, make it better, make it better. and so Astra, you know, along with the other model releases, you know, the the the rate of performance improvement per dollar spent has inflected here, call it a a 99.99% year-over-year cost decline in so far as we can measure it against, you know, Agentic type benchmarks. And Astra is a demonstration of that. and they solve Navier Stokes not with Astra but with another model they have in the bag that they're able to use internally that nobody else has access I feel like that's the biggest that's the biggest takeaway is it's like they're not even working we don't even have access to Astra and they're already working with the next generation one and using it for internal research and they're 10xing to 100xing their capability and Brett what do you make maybe we'll I want to throw we don't want to ignore muse meta's muse I think we'll throw that to next week but AI doomerism anthropic researcher quitting. I feel like this is the consistent quarterly quit and warn the world that you know death is coming from AI. What what do you make of it?
17:06 >> I don't understand that one. I think the idea of alignment so I'll just summarize it or synthesize it as I understand it. The idea is these are super powerful tools and AI could become quote unquote misaligned and if it gets misaligned then we mere humans would be an inconvenience to it and so it would somehow systematically seek to eliminate humans or accidentally eliminate humans on the way to whatever its set of goals is out there in the universe. and it is a mixture of like fantasy that we are ultra powerful and kind of like overinfected by too many sci-fi stories.
17:50 I I think dumerism is an appropriate term because the it's kind of like saying the rise of corporations would cause the extinction of humans because corporations are kind of in some ways they are super intelligence. It's basically a coordination function for a bunch of brains to work together on a single pro problem. And yes, corporations seek to profit maximize or you could interpret it as power maximize. In doing so along the way, you know, they run up against kind of like the gray areas of the law and and then some corporations operate outside the bounds of the law and sometimes by doing so they injure people. And I do think that kind of like as with any powerful tool, some people will be injured by AI systems directly or indirectly. But the incentive structure of accessing resources within the planet Earth and beyond is it's set up such that if you operate within the legal frameworks, it makes it much easier to access and accumulate those resources. And so just like a corporation will tend to particularly as it gets bigger and has more at risk operate within the bounds of the economic rules that have been set out. I think AI systems will be exactly the same. And and if you look at the more powerful AI systems, they're ones that are kind of already being pre-guided to operate within those boundaries and frameworks. And in some ways I think that people have failed to ex update kind of their set of expectations because they they previously thought that these AI systems would look like formal logic systems where you give it a an input you say mac you know make me some paper clips and it's a formal logic system where it's like I have to make paper clips and it just like expands expands expands until it's turning the entire universe into paper clips. And the reality of these AI systems is actually they are deeply heavily encoded with the kind of ethical, legal, and moral frameworks of us humans because they're built on the language of humans.
19:55 And so they they have much more context than people thought they would have of what kind of it means to be in the world. so I think it's you know a combination of people having gotten a lot of resources for being in the system wanting to get off the ferris wheel maybe some pressure from friends and family who feel like corporations are eagle evil and generally people just being kind of like narrow-minded and inane just like so Dario CEO of Anthropic I think we already are due to have 40% job loss due to AI on his forecast like there is 0% chance that's going to happen. And in fact, it's totally backwards. It's the opposite of what's going >> I I want I don't want to go down this rabbit hole, but I think AI dumerism is an elitist take, right? It's it's like a classic case of being in an ivory tower privileged lazy take. And it's like the reality is AI is improving the lives of so many people and will continue to do so in an incredibly meaningful way.
21:01 And you have AI researchers who are getting paid tens of millions of dollars a year saying, "I think it's too dangerous." Right? It's like, "All right, buddy. Like, why don't you go why don't you go work?" I think I think like yeah the deep seed there and this is true with a lot of disruptive technology is people see the extrapolated event where they're going to become less meaningful in this the story of humanity or their own story of meaning and meaning making and and so it feels very uncomfortable if you if you have a lot of power and ability to guide things today and the world is accelerating in a direction that you don't know where it's going to go. You're intrinsically worried that it will mean that your power or maybe the power of your kids, your ability to kind of like be this feudal lord over everybody else will be diminished. And so then you begin to throw sand in the gears of the technology. It's like I' I'm comfortable being at the top of the pecking order where I'm at. So I would prefer that this would not accelerate quite so quickly so I can kind of prolong and and build my dynasty effectively. And again, it's like, yeah, there is a huge amount of human welfare to deliver with AI. I like an unimaginable amount. and it does not necessarily mean the person that's on top is going to stay on top, particularly if they take their foot off the gas.
22:30 >> All right, moving on to Apple's most recent announcement. speaking of context, a Apple Watch that is always listening and can transcribe the last 15 seconds of this conversation. and of course, I think the biggest announcement, the one more thing, the folding iPhone. that is crazy crazy expensive, right? >> Are you going to get are you going to get are you going to buy an iPhone Duo?
23:03 >> No. I people should know this. I'm a late adopter of Apple products. I I I am pro Apple. >> I'm an early adopter of Apple products despite the fact that I trash them pretty consistently, but I'm not going to get phone duo. >> I'm not going to get a duo. I'm not going to get a new phone, which will be the first cycle I've skipped in a while. >> but I do think I want to focus on actually on the watch side of things because I think increasingly >> the watch slash health side of things is going to be incredibly incredibly useful. And you know, we've already spoken previously about how, you know, AI is arguably going to be a better health provider than a existing doctor.
23:50 I think if you're wearing a Apple Watch or other device and you're collecting data 247, it's like right now maybe not so needlemoving. I think over time it could become incredibly useful and just having the longevity of context for yourself, I think there is more upside there than people appreciate. >> Sure. But I I think this is where the Apple has succeeded over time because they took this approach where they would internalize most of everything. you know they're vertically integrated but also from a even with what they're going to do with the watch they're going to restrict kind of the ability for third party developers to get access to the underlying data in any meaningful sense. and so if you think about how Tesla Tesla's approach to FSD over time was we buy this you know for a while they were using radar but then the compressed reads coming off the radar weren't as useful as getting the raw so that they could you know train ML and AI against that raw data or the reason their cameras can see into the glaring sun is because rather than accepting the compressed reads coming off the chip they're going straight to the optics or I think that's roughly right? Or they're they're they're not they're basically the pre-processing that Apple is insisting that it needs to do in part due to privacy, but also to basically keep control, I think, is going to increasingly restrict kind of like the ability of these systems to be universally useful. and kind of the consumer experience with technology is going to be increasingly driven by how easily plug inable is that technology into the AI agents that are running their lives. And I already say like so what do I use Grockbot for? Basically to extract myself out of iCloud into Google Drive because that has a better set of connectors with both OpenAI and Grock.
25:52 and so you know the fact that the only way to get to that iCloud information is whenever Apple releases it, which will be who knows how many years from now, some connector that's their sanitized approved version of getting access to that information. That's going to be too late in AI space. and so I do think like the watch is a compelling product. I'm not going to buy the AirPods with their cameras on them, but having an always on listening device, I think, is going to be something that people will both accept and want over time. and maybe it's their only hope that people will accept the Apple Watch version. My fear would be they'll so heavily constrain and be underpowered in their ability to extract meaningful information out of it that then it can't it won't be a first option or first look for people. So then it becomes a novelty. So then developers don't develop to it because people are never using the feature and then it just kind of like withers and that's that's the Siri story effectively. they tried to launch like Siri skills and you know all this stuff where you could make Siri more powerful but they're they they restrict your ability to open into the ecosystem so much that now when the number of ways you can plug things in together is expanding massively it really makes their platform less valuable over the medium and long term.
27:16 >> Interesting. So who do you think I mean does Garmin scoop this up? Is it a Whoop a Whoop aura world out there? >> I think so. Back to Astra for a second. One of the skills this thing is very good at that it previously wasn't as good at is kind of like CAD and 3D design. You can read kind of the way in which they're training and making these models more expert as also a read into the capabilities that OpenAI needs internally. So if you they're designing hardware devices like that's the kind of skill you would need your model to acrewue. I think that I think that and it could be you know a third party a a an entity that is more open to being developed to I think is the winning play if you're not one of the frontier labs.
28:07 so Slack for instance, Ghat I think meta switch from Ghat to flat Slack because Slack can operate more headlessly for agents to basically plug into the data streams. that's a a sign for how businesses are going to think about the tools they take on here. you know, is if Apple like some of the software companies is sitting in a spot where it's like no, we don't want people to plug into us. we'll provide the agent implementation on part on on top of our constellation and that's it you know and some AI software companies are like or sorry software as a service companies are like no you're not allowed to like plug into like the API layer and we're going to restrict your bot from doing computer use against us if we can cuz we want you to use our bot that like interprets our kind of surface for us. I think that's a losing strategy over any meaningful strategic time horizon given the the both the pace at which things are accelerating and then the core capability of the closed frontier labs. I just think you're in you're in a position of design irrelevance if you adopt that strategy.
29:17 >> And that's not to say the Apple Duo isn't very cool. I will certainly admire from the sideline someone who pulls it out next to me and say >> it's too it's why why can't we build like the whole direction here the reason the watch is more interesting than the Apple Duo is like I think we need to go back smaller I might sound like a crotchety old man now but like the the the phone's getting bigger and bigger and now it's like well it's smaller but it unfolds but the smaller is still wider than the iPhone Pro and so it doesn't really solve the canvas fit in your pocket problem in I think a satisfying way. plus it's like 2x the price. Not quite 2x the price, but is with a worse camera setup. It's definitely a no for me. And I think just make a smaller flip bony version of this thing, you fools.
30:10 All right, next week we've already got exciting topics to cover, including Muse. And I'm sure other wild things will happen between now and then. Thanks to everyone. >> And the flip phony version of it could be the listening device. Like this is what you know I think it's very it's highly likely that SpaceX is going to launch a product here as they go direct to consumer on the cellular side. I I think people are hungry for new form factors on this stuff. You don't need a phone to watch Tik Tok. You can watch Tik Tok in the cyber cab while you're rolling around in the back or in the front.
30:44 There you go. All right. See everyone. >> Arc Investment Management LLC is an SEC registered investment advisor. ARK, Wolf Financial, and Public are separate unaffiliated entities and do not have a relationship with respect to any company marketing or selling the products or services of the others. Ark, Wolf Financial, and public make no representation or warranties regarding and expressly disclaim any responsibility or liability for any acts, omissions, losses, or damages of the other entities or their affiliates, officers, employees, or agents. No company shall be liable for any indirect, incidental, consequential, or special damages arising out of or in connection with this podcast or its contents. The information provided in this show is forformational purposes only and should not be used as the basis for any investment decision and is subject to change without notice. It does not constitute either explicitly or implicitly any provision of services or products by ARC and investors should determine for themselves whether a particular investment management service is suitable for their investment needs. All statements made regarding companies or securities are strictly beliefs and points of view held by ARK and/or show guests and are not endorsements by ARK of any company or security or recommendations by ARC to buy, sell or hold any security.
32:01 Historical results are not indications of future results. Certain of the statements contained in this show may be statements of future expectations and other forward-looking statements that are based on ARK's current views and assumptions and involve known and unknown risks and uncertainties that could cause actual results, performance, or events to differ materially from those expressed or implied in such statement. ARK assumes no obligation to update any forward-looking information. ARK and its clients as well as its related persons may but do not necessarily have financial interests in securities or issuers that are discussed. Certain information was obtained from sources that ARC believes to be reliable. However, ARC does not guarantee the accuracy or completeness of any information obtained from any third party.