SPEAKER_00
Dmitry Dalgov is co-CEO of Waymo. He joined Google's self-driving car project in 2009 as one of its first engineers and was repeatedly promoted until he took it over in 2021. Waymo is Google's most successful moonshot and now provides over 500,000 fully autonomous rides each week. Cheers, by the way. [SPEAKER_01] Yeah, cheers.
SPEAKER_01
You grew up in Russia, right? [SPEAKER_00] Yes, I grew up in Russia.
SPEAKER_00
[SPEAKER_01] Then I was actually Soviet Union.
SPEAKER_01
[SPEAKER_00] Right, exactly. [SPEAKER_00] My dad is a physicist, so the Soviet Union started falling apart and then he had a position, a visiting position, in Kyoto University for a year. We moved there as a family and then he went to Berkeley and I tagged along. And then I graduated from high school, I was thinking about the next thing I wanted to do. And I really liked that technical school in Russia. The Russians are serious about the physics.
SPEAKER_00
They are, they are.
SPEAKER_01
So I went back to Russia and I got my bachelor's and master's there. What year was this that you went back to Russia?
SPEAKER_00
1994.
SPEAKER_01
Okay. So that was almost peak Russian optimism in a sense where it was opening up.
SPEAKER_00
It was. Yeah. No, I actually remember talking to my mom about it.
SPEAKER_01
[SPEAKER_00] And of course, my parents grew up in the Soviet Union. They've seen it.
SPEAKER_00
[SPEAKER_01] They were born right before the war and then they saw they lived through some really tough times.
SPEAKER_01
[SPEAKER_00] And I remember talking to my mom and saying, she, in fact, I got my green card here in the US before I went back and she insisted that I do it. And I was actually at the time, I wasn't thinking of coming back. [SPEAKER_00] But then I was pretty excited about where Russia is and trajectory it's on. And being young and naive, there's no turning back. And so why did you decide to come back? There's more of a play-by-play than. Yeah, no, school, it was pretty clear to me I wanted to continue studying math and computer science.
SPEAKER_01
And while the undergrad and master's that I got in physics and applied math, I think that was still an incredibly strong foundational school of Russian math and science. Graduate school, it was very clear to me that the best way to do it was in the US.
SPEAKER_00
[SPEAKER_01] So I came back.
SPEAKER_01
I'm struck by the founders of the two most valuable UK companies are Russian math nerds who both went to the same school. Nikolai at Revolution and Alex Gerko at XTX. [SPEAKER_00] But yeah, it's a strong diaspora.
SPEAKER_00
[SPEAKER_01] There is a company not far from here where one of the founders also has a similar pedigree. [SPEAKER_01] Mm-hmm. [SPEAKER_01] Exactly. The classic engineering interview question of what happens when I type google.com and hit enter, talk me through, whatever you like. HTTP and DNS and BGP, you can go down to whatever level of stack you want. Do you want to maybe just describe when I take a ride in a Waymo today, what's happening at a technical level?
SPEAKER_01
[SPEAKER_00] Like, what is the architecture?
SPEAKER_00
Let me answer your question.
SPEAKER_01
What's happening in real time?
SPEAKER_00
Yeah. [SPEAKER_01] But this is going to be only a part of the story because we're going to be talking about the inference, the real-time inference part of it. And if we want to have a deeper, richer technical conversation, I think it would be interesting also to zoom out and talk about the entire ecosystem of what goes into building, evaluating, and deploying the Waymo driver. But when you're driving around or being driven around, we think about what we're building as a driver. Obviously, it's not a car. So it has a number of sensors that are positioned around the vehicle.
SPEAKER_01
[SPEAKER_00] We use three different sensing modalities. There's cameras, there's LiDAR or lasers, and there are radars. Those are the primary ones. [SPEAKER_00] There are also microphones, directional microphone arrays, but those are the primary three for sensing the world. They all have very nicely complementary physical properties. They all have 360-degree coverage around the vehicle, so the Waymo driver sees 360 all the time. So all of the data goes into a computer, as you would expect. And there's the software that processes it. Now it's all AI, specialized AI in the physical world.
SPEAKER_01
So it processes the sensor data. Nowadays, we talk about it using AI terminology as encoders that take this data in. And then there's the decoder, the action, the generative part, if you will, in the car. And the generative task there is to figure out how to drive. Right? And that is, of course, connected through a specialized interface to the car where we can actuate the vehicle. And that's why you see the steering wheel turn and it drives you around. Okay. So I get into my car. There's three main families of sensors, LiDAR, radar, and cameras.
SPEAKER_01
And then it is using that to first build a model of what's going on in the world, where are all the other cars and things like that. And then you make decisions and then actuate that with the car. That is the system that you're living in. And is all that inference done locally or presumably yes, nothing's in the cloud? Nothing real time. Nothing real time in the cloud. And there are some things that can happen in the cloud, but they're not required. Got it. What's an example of a nice to have that happens in the cloud? You can imagine a situation where we do some of it is not directly related to the driving.
SPEAKER_01
But let's say after you leave the car, we want to check that the car is not dirty. You didn't leave anything there. If you did leave an item, well, if you left in a mess, then I want to send the car to one of our depots, get it cleaned up. If you left an item there, your phone, all right, we want to detect that and then send it to our lost and found and let you know. So that we do with a model that actually lives off board as opposed to having to put it on the car, right? Because it's not a real time task related to the driving. So that's one example of something that.
SPEAKER_01
There are all these debates that go on on Twitter around self-driving. So I can think of end-to-end versus the more modular approach. There is cameras only versus array of sensors. And I can't tell, are these debates actually interesting to an expert in the field? Or do you think these are just settled matters and they're just grist for the algorithm?
SPEAKER_00
[SPEAKER_01] So we do that by asking a model that actually lives off board as opposed to having to put it on the car, right? [SPEAKER_01] Because it's not a real time task related to the driving. So that's one example of something that. [SPEAKER_01] There are all these debates that go on on Twitter around self-driving. So I can think of end-to-end versus the more modular approach. [SPEAKER_01] There is cameras only versus array of sensors. And I can't tell, are these debates actually interesting to an expert in the field? [SPEAKER_01] Or do you think these are just settled matters and they're just grist for the algorithm?
SPEAKER_00
[SPEAKER_01] I understand where the questions are coming from. I do find that often the way they're posed and the way the debate happens is losing a lot of the nuance and a lot of detail that really matters. [SPEAKER_01] So I think the question that's most interesting to me is that the most interesting technical questions are at that level. Because the way we think about building the Waymo driver, it starts with a large off board foundation model.
SPEAKER_01
[SPEAKER_00] I can imagine building a big model that understands how the physical world works and understands the important properties of what it means to drive, the social aspects of driving, and what it means to be a good driver as opposed to a bad one. [SPEAKER_00] So that's the foundation. Then we specialize it into, let me call it, three main off board teachers. There are still large high capacity off board models. [SPEAKER_00] There's the Waymo driver, there is the simulator, and then there's the critic, right? And those then get distilled into smaller models that you can run inference on faster.
SPEAKER_01
[SPEAKER_00] So the Waymo driver becomes the backbone, the main backbone of what's in the car. The simulator, of course, is what powers our synthetic generative environment that can run on the cloud for training and for evaluation of the system.
SPEAKER_00
And the critic, does the simulator ever run locally? No.
SPEAKER_01
[SPEAKER_00] No, it doesn't. However, what I think is interesting, the way the decoder works, the way the model works, if you think about the generative task in the simulator, of creating those realistic worlds and how other people behave, how cars, pedestrians, cyclists move, and the task that you have to solve on the car in real time,
SPEAKER_01
there is this fundamental shared capability of understanding how these objects relate to each other and predicting what they might do in the future if you are running on the car, and then generating those, sampling those probabilistic behaviors in the simulator. So it's different models, but there is a reason the shared foundation model is able to power both. And similarly, if you think about the critic, the job of the critic is to find interesting events and then be opinionated about what's good behavior and what's bad behavior.
SPEAKER_01
Similar fundamental understanding, right? If you're running inference on the car, you still have to figure out which of the multiple hypotheses of these future worlds you want to take action to steer towards. [SPEAKER_00] Yes. Okay. And these are all downstream of the same foundation model? That's right. So start with the foundation model. Yep. Then you specialize and fine tune, still off-board model. Those are the teachers. And then you distill. Each one of the teachers distills and trains its own student.
SPEAKER_00
[SPEAKER_01] Yes. [SPEAKER_01] The driver, the simulator, the critic. [SPEAKER_01] Yes. [SPEAKER_01] You started working on self-driving 20 years ago. [SPEAKER_01] Yeah.
SPEAKER_01
[SPEAKER_00] As you think about the tech evolution, is this just a scaling law story where we had to be able to throw enough compute at it? Were there architectural approaches we needed to wait to have invented? [SPEAKER_00] Was it just a story of we needed 20 years of going down the wrong cul-de-sacs before we eventually arrived at the right approach? You know, could you, knowing what you know now, could you have a successful Waymo in market in 2015? Or was there some enabling technology? [SPEAKER_00] No. Technology breakthroughs that happened over the years were critically important, primarily in AI, but also in other areas, such as compute. [SPEAKER_00] Yes.
SPEAKER_01
[SPEAKER_00] Have you considered the neural computer? Yeah. Now, I wouldn't characterize it as going a thousand different dead ends and then having to retract and then finding the one right path. I would characterize it as iterative learning and evolution. Yes. And then transformers came around, but transformers, for example, are a very general architecture, right? Yep. Powers of the lens, powers our models, but how you apply them to that space, I think this is where it matters. It doesn't just fall out of transformers. Exactly, right.
SPEAKER_01
And of course, people talk about architectures, but architecture is important, but really a lot of it comes down primarily to your metrics, to your evaluation mechanisms, to all of the training recipes, and of course, new data. Yes. LMs are good at text or tokens specifically, and obviously perform best at domains that have some single corpus of text they can work on, like coding, where it's very helpful that everything was already textual. And part of the success has been creating textual representations for domains, such that we can then put a lens against them. Can you describe how you encode the world that you're seeing?
SPEAKER_01
Yeah, are you just building a 3D map, a 3D bitmap essentially, or? So this is where I think we can get into the question of what is the interface between the encoder and the decoder parts. And I think that touches also on the thing you flagged earlier, where people debate end-to-end or not end-to-end. Yes, yes.
SPEAKER_00
[SPEAKER_01] So the way, let's talk a little bit about end-to-end, and then get back to what is the interface between those two, right? [SPEAKER_01] So when you say end-to-end, what do we mean? [SPEAKER_01] We mean that it is some large ML model. Typically, you don't build them monolithically.
SPEAKER_01
[SPEAKER_00] You have different parts and different subgraphs. [SPEAKER_00] But what's important is that you can propagate and backprop the gradient and the loss function all through the different layers. [SPEAKER_00] So every layer you can learn the weights and the representations that matter for the final task. You don't force it through some narrow funnel between, let's say, the encoder and the decoder. Yeah, I think of a simple view of end-to-end being pixels go in and car actions come out, which is maybe a bit of an oversimplification, but yeah. Yeah, that's exactly right. And if this is the basic vanilla version of it, right? Yeah.
SPEAKER_00
[SPEAKER_01] There, if you think about what will it take to build the driver that's capable of fully autonomous operations. [SPEAKER_01] Yeah. [SPEAKER_01] If you think about this entire ecosystem of the driver, the simulator, the critic. [SPEAKER_01] You don't force it through some narrow funnel between, let's say, the encoder and the decoder. [SPEAKER_01] Yeah, I think of a simple view of end-to-end being pixels go in and car actions come out, which is maybe a bit of an oversimplification, but yeah. [SPEAKER_01] Yeah, that's exactly right. [SPEAKER_01] And if this is the basic vanilla version of it, right? [SPEAKER_01] Yeah.
SPEAKER_01
There, if you think about what will it take to build the driver that's capable of fully autonomous operations. Yeah. If you think about this entire ecosystem of the driver, the simulator, the critic. [SPEAKER_00] Yep. [SPEAKER_00] If that's all you do, pixels in, trajectories out. [SPEAKER_00] It becomes very difficult to do all of those three and achieve the high level of safety and performance that we require. [SPEAKER_00] And it becomes very difficult to do it at scale. [SPEAKER_00] Yeah. [SPEAKER_00] And, however, that's a very easy way to get started, right? [SPEAKER_00] Mm. You collect some data, and I'll add you to the LLM world, right?
SPEAKER_01
The easiest thing you can do is have a model.
SPEAKER_00
[SPEAKER_01] Yeah.
SPEAKER_01
The easiest way to get started nowadays would be to take a VLM. It already has a language aligned camera encoder. [SPEAKER_00] Yep. And then it has a decoder that can predict and generate text.
SPEAKER_00
[SPEAKER_01] And you can fine tune it and say, hey, instead of text, generate trajectories. [SPEAKER_01] Very doable. [SPEAKER_01] In fact, a while ago we published a paper. [SPEAKER_01] Yes.
SPEAKER_01
[SPEAKER_00] Called AMMA that did exactly that. Yes. And it will actually, in the nominal case, drive pretty well, which is mind-blowingly impressive. [SPEAKER_00] That is very funny. Yeah. [SPEAKER_00] And I mean, there's some intuition. [SPEAKER_00] You're saying you can take an off-the-shelf model, which has nothing to do with driving to start with. [SPEAKER_00] That's right. [SPEAKER_00] And you'll get these good results. That's right. In the nominal case.
SPEAKER_00
[SPEAKER_01] I just want to be clear, it's orders of magnitude away from what you need.
SPEAKER_01
Yeah, you should not try it on the street, but it works. But for example, if you want. It's a talking horse. It's impressive that it's talking, right? Exactly. Exactly. And you can actually, if the product that you wanted to build was a driver-assist system, not a fully autonomous system, then maybe that's all you need to do. [SPEAKER_00] Yep. And then, for that, you don't need all this other machinery of the simulator and the critic, because the number of nines is drastically lower. But this is interesting, because there is some intuition behind why that works. If you think about the hard parts of driving, it's not unlike having a conversation. Mm-hm.
SPEAKER_01
Except in the LLM world, right? [SPEAKER_01] You're modeling language or maybe modeling a dialogue in the space of sentences and words. What makes driving hard is also this multi-agent social interactive part of it. And if I do something, that's going to affect you, it's going to affect somebody else. And the history matters. It's not local and just geometric. Context matters. Semantics matters. But it's in a different space. It's not in the language of words, it's in the language of body language, if you know what I mean. Right? And we see that empirically validated if you do this approach. Okay.
SPEAKER_00
[SPEAKER_01] So then, let's say we build this thing. [SPEAKER_01] Just cameras, camera encoder, pixels go in, trajectory go out.
SPEAKER_01
But the quality is sufficient to drive. In the normal case, it's not sufficient to deal with the long tail of the edge cases and hit the high bar of superhuman safety that we require. So then, you start asking the question, what else do you need?
SPEAKER_00
[SPEAKER_01] Yes. [SPEAKER_01] And if all you did was observing how other people drive when you trained the system, maybe observing passively how people drive and how they interact, maybe also driving the car yourself and then using imitative learning to train it. [SPEAKER_01] Mind that that's not enough. You have to do something in closed loop.
SPEAKER_01
[SPEAKER_00] You have to do things like RLFT. RLFT? RLFT.
SPEAKER_00
[SPEAKER_01] Reinforcement learning based fine tuning. Okay. So, similar to the reinforcement learning with human feedback in the LLM world, right? You want to do a closed loop driving where you explore all kinds of different situations and then you give it a reward signal to keep it in distribution. [SPEAKER_01] For that, then, you need a realistic simulator.
SPEAKER_01
Right? Yeah. [SPEAKER_00] Also, if you want to have a good RL system, you need to have an opinion for the reward function. [SPEAKER_00] This is where the critic comes in, right? [SPEAKER_00] If you have a purely end-to-end system, let's look at the simulator. [SPEAKER_00] Now, what do you do? [SPEAKER_00] You have to be constrained to go from pixels to trajectory, right? That's all you can run the system on, right?
SPEAKER_00
[SPEAKER_01] Yeah.
SPEAKER_01
It's a very high dimensional space. It's a hard problem to generate everything. But even if you solve that, it becomes incredibly inefficient to run it in the full way of pixels to trajectories and simulation for training or for evaluation. So, this is when intermediate representations come in. There are some intermediate representations in the world in this task, in the physical world we know are correct. Yes. [SPEAKER_00] They are not sufficient, but they're not generally limiting, right? Yes. In other words, there's an object here, there's a concept of a road, there's signs, there's speed limits.
SPEAKER_01
[SPEAKER_01] So, this is where augmenting that learned representation, those learned embeddings from the encoder decoder with structured representation is what we do. And we find that this gives us additional knobs to simulate in that space, pixels to trajectories. It allows us to have additional safety validation layers in real time. And it also gives us additional mechanisms to specify the reward function, for evaluation of the critic or for training. Yes. Yes. In other words, there's an object here, there's a concept of a road, there's signs, there's speed limits.
SPEAKER_01
So, this is where augmenting that learned representation, those learned embeddings from the encoder decoder with that more structured representation is what we do. And we find that this gives us additional knobs to simulate in that space, just pixels to trajectories. It allows us to have additional safety validation layers in real time. And it also allows us to give us additional mechanisms to specify the reward function for evaluation of the critic or for training. Yes. So, this is again, we've gone full circle of it. Is it end-to-end? Yes, it is.
SPEAKER_01
Yes. But if you want to do it at scale, for full autonomy, it's augmented with all of this other stuff. That's very interesting on the simulating point. It's just very hard to simulate for an end-to-end model because it's easier to deal in intermediate representations, rather than coming up with the pixel perfect view of the world.
SPEAKER_00
[SPEAKER_01] You need both. [SPEAKER_01] Yeah. [SPEAKER_01] So, having end-to-end architecture that's augmented with that structure allows you to play in both of those worlds.
SPEAKER_01
Yeah, yeah, yeah. Hmm. So, what are you looking to do as a self-driving car? I mean, it sounds funny, but I think people maybe don't realize that there are many different things that you're looking to solve for, where you're looking to get the person to their destination, you're looking to get them there reasonably promptly, but also drive quite smoothly, and also have many lines of safety, not annoy other drivers and get honked at. So, what are some of the reward functions or things you're optimizing for that maybe are not obvious to people? So, safety is the primary focus, right?
SPEAKER_01
Safety is the fact that people can't have nice things or not everyone is nice to the robots. And so whether you're driving through a dodgy area or getting blocked, or maybe I'm not going to drop you off here, maybe I'm going to go around the block and drop you somewhere better. But all of these, as you say, other human issues, how do you go about solving this? A lot of the ones that you mentioned are just things that we need to work on and understanding. Honestly, if we're not dropping you off exactly where you want it to be dropped off, or we don't give you a good interface to tell us, that's on us. You just need to make it better.
SPEAKER_01
It feels like the drop-off is actually a pretty nuanced part of the self-driving journey, the highway stuff and the 35-mile-an-hour roads, that is all nailed, but there's a lot of nuance in the drop-off experience. I'd say they're all hard. You picked freeways and you picked drop-offs. For different reasons. For drop-offs, you're absolutely right, there are a few things that are not obvious when you think about this problem. But it's understanding where you want to go, and making it as convenient as possible for you. Pick-ups from drop-offs, it's not exactly symmetrical. But then it's also understanding the context of the situation, where do you stop?
SPEAKER_01
You don't want to block a driveway, you don't want to double park, although in some cases where if it's a quick one, maybe it's okay. So there's a lot of nuance that goes into doing that well, so that it's a smooth, frictionless experience for the rider, as well as other folks. [SPEAKER_00] Yeah. [SPEAKER_00] Freeways, for most of the time, not much happens. [SPEAKER_00] They're very well structured, because we designed them that way. But there is still that long tail of really complicated stuff that happens. Yes. Where the consequences of a bad event are much more severe. [SPEAKER_00] Right? [SPEAKER_00] The speed is much higher, everything is quadratic in speed.
SPEAKER_01
[SPEAKER_00] But we see a lot of stuff. [SPEAKER_00] Imagine grills falling off of freeways, imagine people getting into accidents and spinning out of control. [SPEAKER_00] You see one of those flatbed trucks with a bunch of stuff piled in it, and you're driving behind it. I always find it very nerve-wracking.
SPEAKER_00
[SPEAKER_01] It looks a bit… I know. Yeah.
SPEAKER_01
[SPEAKER_00] Yeah. [SPEAKER_00] And we've seen them leave a trail. [SPEAKER_00] Yes.
SPEAKER_00
Yeah. Okay. So it's a different set of problems. [SPEAKER_01] But I feel like the general sentiment with Waymo is that the driving has mostly now been solved by you guys, and it's a question of scaling up, and maybe some super long tail stuff, really snowy conditions. [SPEAKER_01] Is that your sense internally, or is there actually much more nuance to it than that? [SPEAKER_01] I would say the… [SPEAKER_01] Yeah, it's not like we're done with engineering.
SPEAKER_01
Yeah. I would say that we've clearly moved past the stage of scientific research and deep core technology development to this new phase of accelerated global scaling and deployment.
SPEAKER_00
Yes. So, we still have work to do, right? Yeah. [SPEAKER_01] But I don't see today any limitations or any gaps in the core technology. [SPEAKER_01] The driving is good enough now. [SPEAKER_01] The core technology I think is good enough that I can't think of any aspect of driving that is not supported by the fundamental technology. [SPEAKER_01] Now, that said, there is a lot of work to do in specialization and in validation before we can deploy responsibly, right? [SPEAKER_01] technology development to this new phase of accelerated global scaling and deployment.
SPEAKER_01
[SPEAKER_00] Yes. [SPEAKER_00] So we still have work to do, right? [SPEAKER_00] Yeah. But I don't see today any limitations or any gaps in the core technology. The driving is good enough now. The core technology I think is good enough that I can't think of any aspect of driving that is not supported by the fundamental technology. Now, that said, there is a lot of work to do in specialization and in validation before we can deploy responsibly, right? We're not driving everywhere in the world.
SPEAKER_00
[SPEAKER_01] We are planning to start operating in London and in Tokyo this year. [SPEAKER_01] Do we have a driver that you're using today in San Francisco that we can just plop down in London and go? [SPEAKER_01] No, right? [SPEAKER_01] But what we're seeing is incredibly encouraging from the perspective of whether the core technology is there.
SPEAKER_01
Yes, right? So now it's a matter of collecting the data, doing some specialization and validation. The signs are different. In both of those places, people drive on the other side of the road. But that's actually not that hard for computers, right? Core technology generalizes really well, but you still have work that you have to do. [SPEAKER_00] What generalizes least well? Increasingly, we're finding, especially now that we're able to hook the Waymo AI to the digital world and the VLMs and inherit the general world knowledge from VLMs, we're seeing really strong results from zero shot or few shot learning because of that general knowledge that we bring in.
SPEAKER_01
But there are a few things, like cold weather, cold winter weather, where it affects the entire stack. Right? It's not just the AI. We actually have to. [SPEAKER_00] Hardware, yeah. [SPEAKER_00] You need the hardware, you need to have the proper cleaning solution, heating elements in it. [SPEAKER_00] And then you think about things that are completely solvable by computers, like motion control and slippery surfaces, right? So that takes a bunch of work. [SPEAKER_00] You don't get that for free from just pulling in some VLM decoder. Yes.
SPEAKER_01
[SPEAKER_00] Was it the case that in the early days, there was a lot of San Francisco specific work or Phoenix specific work in the early markets, whether it be mapping or something else, and that you guys seem to have either solved that in generalizing it, or just scaled up your ability to do the city specific work? [SPEAKER_00] How much enabled the rapid city expansion? We usually think about the capability of the Waymo driver, as well as deployment, not primarily and directly in that space of cities or zip codes. I think about the operating domain. [SPEAKER_00] Right? And that's freeways, cold weather, snow, rain, fog, density, et cetera.
SPEAKER_01
And that's what we are building, that's where we're evaluating, and that maps to a city, a particular city, whether it be within the operating domain or outside of it. [SPEAKER_00] Yeah, right? So if we rewind history a little bit, our initial deployment where we started offering a fully autonomous commercial service for the first time was in 2020.
SPEAKER_00
In Chandler, Arizona. [SPEAKER_01] And that was on what we called the fourth generation of the Waymo driver. [SPEAKER_01] This was the Pacifica minivans with different hardware, different software. [SPEAKER_01] We were super focused on doing the whole thing end to end.
SPEAKER_01
[SPEAKER_00] Right.
SPEAKER_00
[SPEAKER_01] Learn how to build the driver, evaluate it, deploy regularly, operate it end to end, 24-7 with customers, learn from the customers. [SPEAKER_01] And we were very focused on that operating domain of mostly Chandler. Yeah. [SPEAKER_01] Which is a medium, low complexity one. Yeah. [SPEAKER_01] Then when we made the jump to the fifth generation of our system, which is what's on the Hyundai today, we really wanted to take a huge bite out of that operating domain. Yeah.
SPEAKER_00
[SPEAKER_01] We collected data all over the United States, all different states, different cities. [SPEAKER_01] And we chose to deploy in the hardest parts of San Francisco, hardest parts of Phoenix. [SPEAKER_01] We made a big jump on the hardware side and most importantly on the software, the AI side.
SPEAKER_01
I would say that was the big discontinuous jump. And that's what you're seeing now after we've scaled up and iterated on all aspects of building and deploying the driver. This is now why you're seeing us go in parallel and scaling in the US. [SPEAKER_00] So driver version five was just a much more generalizable stack than version four. [SPEAKER_00] And what was it about it that made it just have been trained on a much wider dataset? It was when we made this big bet on AI. [SPEAKER_00] Yeah. There was a lot more of small AI models and ML models in the fourth generation. [SPEAKER_00] Got it. We made a much bigger bet and jump to AI as the backbone for the fifth generation.
SPEAKER_00
[SPEAKER_01] AI is the backbone as the core engine, meaning you're saying that Gen 4 had lots of small subsystems.
SPEAKER_01
[SPEAKER_00] Okay. [SPEAKER_00] Yeah. And that's been, so we made that jump and we've been iterating and improving the model since then. [SPEAKER_00] As Waymo rolls out widespread autonomy, it has second order changes on the entire system. [SPEAKER_00] In this case, traffic patterns or other drivers' behavior, or eventually how cities are laid out. [SPEAKER_00] And autonomous systems are coming in many domains. In commerce, soon agents are going to be transacting without human intervention. We're basically getting driverless commerce. And Stripe is building the economic infrastructure for AI. And as part of that, we're letting payments be initiated by humans or by agents.
SPEAKER_01
So if you want to sell to agents, or if you want to let your agents spend money all around the web, check out Stripe's agent and commerce suite. [SPEAKER_00] Can we talk about hardware a second? system. In this case, traffic patterns or other drivers' behavior, or eventually how cities are laid out. [SPEAKER_00] And autonomous systems are coming in many domains. In commerce, soon agents are going to be transacting without human intervention. We're getting driverless commerce. And Stripe is building the economic infrastructure for AI. And as part of that, we're letting payments be initiated by humans or by agents.
SPEAKER_01
So if you want to sell to agents, or if you want to let your agents spend money all around the web, check out Stripe's agent and commerce suite. Can we talk about hardware a second? So lots of hardware questions, but one is maybe everyone in this space has a very charismatic demo of a vehicle that is custom made for self-driving. [SPEAKER_00] And so it's often the van with no steering wheel, seats facing in both directions. You guys have one. Tesla has the steering wheel-less cyber cab. Cruise have the cruise origin. And yet we're still driving in Jaguars that have a steering wheel in the front and are pretty similar to consumer cars.
SPEAKER_01
[SPEAKER_01] And it's interesting to me because if we were talking about this 10 years ago, we might say, well, yeah, developing a custom car, that's relatively straightforward. We know how to put a bunch of sensors on a new car, but the software will take a long time. And what's interesting is we've made huge progress in the software, but interestingly, the cars are still derivatives of cars that people are driving. And so I'm curious why you just think the custom hardware has not happened as of 2026.
SPEAKER_00
[SPEAKER_01] It's obviously a small improvement compared to Waymo, which is the big improvement, but it's just interesting that it still hasn't happened. [SPEAKER_01] Well, let's say our sixth generation of the vehicle and the driver is our version of that. [SPEAKER_01] Oh, no, I know it is. [SPEAKER_01] It is the Ojai platform, right? [SPEAKER_01] So that is, you know, it still has the, we can talk about whether you want to have the seats pointed backwards or not. [SPEAKER_01] I actually think it looks nice in a demo, but practically speaking, it's maybe not the way to go.
SPEAKER_01
But that is a custom designed vehicle. [SPEAKER_00] And it is, we put a lot of thought into moving away from a car that's designed around the driver, to a car that's designed around the passenger. And it's much more spacious, but it's happening.
SPEAKER_00
[SPEAKER_01] It's not open to the public yet.
SPEAKER_01
But I took a ride in it the other day, fully autonomously, and that's coming this year. Yes. How much better is it as a passenger experience? You'll tell me once you give it a try. I love it. [SPEAKER_00] Okay. [SPEAKER_00] It's all about the space.
SPEAKER_01
[SPEAKER_00] And the convenience of ingress and egress and the screens and the interface of the passenger. So we put a lot of thought into every aspect of it.
SPEAKER_00
[SPEAKER_01] Yes.
SPEAKER_01
So it has sliding doors. It's very easy to get in. [SPEAKER_00] It has a flat floor. [SPEAKER_00] If you sit in the back, you can fully stretch out. [SPEAKER_00] And there's so much space there. [SPEAKER_00] And it looks from the outside fairly big. [SPEAKER_00] Yes. [SPEAKER_00] Right? [SPEAKER_00] But the actual footprint of that is barely, barely, barely larger than the I-Pace. [SPEAKER_00] So it's amazing that you walk in, it feels like you're in the living room. [SPEAKER_00] Yes. [SPEAKER_00] I guess my question is, Waymo does 25 million rides a year, right, with the Jaguar I-Pace.
SPEAKER_01
[SPEAKER_00] And it's interesting that so much scaling has happened with self-driving so far on the old retrofit. [SPEAKER_00] Yeah. [SPEAKER_00] Maybe that's to be expected. [SPEAKER_00] I think, well, it matches the high, I don't think it's a given. [SPEAKER_00] You're right. [SPEAKER_00] I think, but if you think about the value proposition, right, of course, there is the safety of it. [SPEAKER_00] Yep. [SPEAKER_00] You don't have to worry about it. [SPEAKER_00] Yep. [SPEAKER_00] There's also the privacy.
SPEAKER_00
Yep. Being in the car by yourself, with other folks, but not having to share the space with another human, right?
SPEAKER_01
[SPEAKER_00] Maybe you want— [SPEAKER_00] No, Waymo is a great product. Yeah. [SPEAKER_00] But I guess this is why we're seeing such consistency in the car, you know, it drives well, very predictable. And you can go beyond that, right? And you specialize even more to make the experience even more magical around the rider.
SPEAKER_00
[SPEAKER_01] But I guess it's, it would have been disappointing if, without the specialized car, and I think I would have been surprised if we leveled off at some other much lower level of customer adoption. [SPEAKER_01] Because a car seems like more of an optimization improvement, but the core of the value proposition comes from those other factors. [SPEAKER_01] Yes, yes. [SPEAKER_01] I guess it's just take risk on one thing at a time.
SPEAKER_01
We'll start by doing the software layer, and then we'll build a specialized car or something like that. [SPEAKER_00] That's right. That's right. Yeah.
SPEAKER_00
Yeah. [SPEAKER_01] It's also— [SPEAKER_01] Yes. [SPEAKER_01] I mean, as you said, it's a big investment. [SPEAKER_01] Yes. [SPEAKER_01] So you have to de-risk the fundamentals. [SPEAKER_01] Yes. [SPEAKER_01] And throughout our history, we were very focused on setting the most, the biggest goal for the company to de-risk the most important questions, right? [SPEAKER_01] We talked about the third generation, where we wanted to deploy something and go end to end. [SPEAKER_01] We talked about what was the goal with the fourth generation, and then, sorry, the fifth generation. And then there's the sixth generation, right? [SPEAKER_01] Yes.
SPEAKER_00
[SPEAKER_01] As you said, it's a big investment. [SPEAKER_01] Yes. [SPEAKER_01] So you have to de-risk the fundamentals. [SPEAKER_01] Yes. [SPEAKER_01] And throughout our history, we were very focused on setting the biggest goal for the company to de-risk the most important questions, right? [SPEAKER_01] We talked about the third generation, where we wanted to deploy something and go end to end. [SPEAKER_01] We talked about what was the goal with the fourth generation, and then, sorry, the fifth generation. And then there's the sixth generation, right? So it was the sixth generation where it made sense to go and spend all this effort into the custom—
SPEAKER_00
And the sixth generation is both the custom vehicle. Is it also a new generation of the driving stack? Yeah. It is the new hardware. [SPEAKER_01] Yep. [SPEAKER_01] The sensors, the hardware, the software and hardware they're putting on the Ojai vehicle is the sixth generation. [SPEAKER_01] Yep.
SPEAKER_01
It is very different from the fifth generation.
SPEAKER_00
[SPEAKER_01] It is simpler.
SPEAKER_01
It is more capable. It is much lower cost. It's a fraction of the cost. It's comparable to what you would get with a fancy ADAS system nowadays, the driver assist system. [SPEAKER_00] Yeah. The software is pretty much the same. So when we talk about generalizability of the Waymo driver, we talk about weather conditions. Yes. We talk about cities, but it also generalizes well to different vehicle platforms and different sensor configurations. Okay. So Gen 6 is a new vehicle and a new sensor stack, but similar software. It's almost a tick-tock cycle happening here.
SPEAKER_00
It's similar software. That's right.
SPEAKER_01
[SPEAKER_00] That's right. [SPEAKER_00] And then we're going to put the sixth generation Waymo driver on other vehicle platforms, like the Hyundai Ioniq that's coming later in the year. [SPEAKER_00] What is different about the sixth generation hardware stack and how did you make it cheaper?
SPEAKER_00
[SPEAKER_01] So it still has the same three sensing modalities, but we've made significant optimizations in all three.
SPEAKER_01
Yeah. Unification, simplification, and riding the commodity curve. [SPEAKER_00] Yeah. Is it a classic case of manufacturing scale where we're not using more than the other cars? [SPEAKER_00] Well, scale hasn't fully come into place, but if you think about the supply chains and industries, cameras are pretty mature. Yeah. So now you can get a decent automotive radar for tons of dollars. Mm-hmm. There is a variant of the automotive radar called the imaging radar and it gives you richer data. So that has also come down in cost drastically, but it's a little bit behind your standard automotive radars. LIDARs are following the same very predictable, well-known trend.
SPEAKER_01
So we're riding that and we're also learning from the previous generation to make improvements, simplifications, and optimizations. [SPEAKER_00] So I have a very silly question. [SPEAKER_00] What are LIDARs versus radars better at in a self-driving context?
SPEAKER_00
[SPEAKER_01] Are they complementary? [SPEAKER_01] They're very complementary. [SPEAKER_01] Yeah. [SPEAKER_01] They both involve blasting photons out there and then they bounce off something and come back. You measure what comes back. [SPEAKER_01] The frequencies are very different. [SPEAKER_01] Yes. [SPEAKER_01] So laser gives you very high resolution.
SPEAKER_01
You can think of it as a laser beam that goes out and spins around. Yeah. It shoots out millions of these laser pulses per second. And then each one comes back and you're sampling the 3D structure of the world with very high resolution. So LIDAR for very fine-grained mapping. That's right. Radar has much lower resolution, but because of the physics of it, it degrades much better in adverse weather conditions. Yeah. So if you're driving on a freeway, radar will give you really good returns for cars that are essentially invisible in the camera space. [SPEAKER_00] That's interesting.
SPEAKER_01
[SPEAKER_00] So does that mean there are some environments where you'll be relying significantly more on radar?
SPEAKER_00
It's a combination of the sensors, right? So we rely on each one, and each one is noisy, right?
SPEAKER_01
Yeah. [SPEAKER_00] How the noise characteristics show up in different environments is different, but it's not like we switch from one to another. It's not like we estimate what's happening with the world through cameras and through radars and through LIDARs and then compare. No, there's an encoder for camera, there's an encoder for LIDARs, there's an encoder for radar, and they all go into the system that gives you jointly the best view of what's happening in the world. [SPEAKER_00] So if it's a nice, bright, sunny day, cameras are very valuable. If it's pitch dark or you have sun in your face or you're blinded by headlights from an oncoming car, then camera will degrade.
SPEAKER_01
There's still some noisy signal, but it will degrade. Yes. [SPEAKER_00] And radar and LIDAR are completely unaffected. Right.
SPEAKER_00
[SPEAKER_01] Are there technical problems that are your white whale or you're still chasing, or are you particularly interested in solving, even if they're niche? [SPEAKER_01] We really want to have driving when it's actually snowing nailed. [SPEAKER_01] If it's pitch dark or you have sun in your face or you're blinded by the headlights from oncoming car, then camera will degrade. There's still some noisy signal, but it will degrade. [SPEAKER_01] Yes. And radar, late, LiDAR is completely unaffected. [SPEAKER_01] Right.
Are there technical problems that are your white whale or you're still chasing or you are particularly interested in solving, even if they're niche for the, we just, we really want to have driving when it's actually snowing nailed or steep hills in San Francisco, or are there problems you've been very interested in historically or still are? [SPEAKER_01] I'm super excited right now about the accelerating global expansion. [SPEAKER_00] Mm-hmm. [SPEAKER_00] More cities in the United States and going internationally. So being, I don't know, I understand I'm not answering your question about the knowledge, I'll come back to that. [SPEAKER_01] [SPEAKER_01]
SPEAKER_01
But really that's the thing that I'm, today most excited about. Getting to a place where any major metropolitan area, you can fly into the airport and then take away more and go anywhere you want to go—that is insanely exciting to me right now.
SPEAKER_01
So technically, what I'm most excited about is all of the rapid progress in AI and the world models, the foundational model work. It is just such a massive boost to how much we can simplify the system, how much we can bring down the cost and how we can scale globally. And there's some magic that happens that I don't think I would have anticipated a few years ago. Yeah. So that I find from the technical perspective, just insanely thrilling. Yes. When you talk about the progress in AI, what are the most fun parts of it for you these days?
SPEAKER_01
I think it's seeing the capability and the scaling laws from this approach of starting with that cornerstone of the foundational model and then specializing to teachers and then distilling. You get such big wins in performance across the board. You invest something into the architecture, you get better data or training recipe. [SPEAKER_00] And then you invested that early stage and then it just has massive amplification and ripple effects. And so that in some ways is magical. And then you see it on the car and I've had some moments where the car does something and I look at a log and I've been surprised. It does things that I didn't think it was capable of doing. Mm-hmm.
SPEAKER_01
All right.
SPEAKER_00
[SPEAKER_01] So it's that—when you see emergent behavior, that's a proud moment. [SPEAKER_01]
SPEAKER_01
One example. Yeah. When you build a system and then you think you understand how it works and you understand fully the limits of its capability and performance, and then it does something almost magical. Yes. It's exhilarating.
SPEAKER_00
Yes.
SPEAKER_01
[SPEAKER_00] So one example I can give you, I think I've shared some videos of that publicly in some talks—this example where the situation that happened in San Francisco, a fairly benign situation where at an intersection, our light is red, there's new cross traffic, a bus goes by and it stops partially blocking our light. Our light turns green. So we start to go, we're nudging around the bus and then you see a pedestrian being detected on the other side of the bus. Mm-hmm. All right. And then your car responds appropriately. It slows down, goes a little bit wider. Yep. And then a pedestrian actually emerges from the bus and we go on our own way.
SPEAKER_01
So the first time I looked at that log, I thought what's going on here? I know we have pretty good sensors and the software is very capable, but we don't see through stuff. [SPEAKER_01] Yeah, yeah, yeah. Right? That's not how cameras or LiDARs and radars work, right? I saw the pedestrian through the bus. You saw the pedestrian on the other side of the bus. Yeah, yeah. And it's not, you look at the windows and, you know, radars shouldn't—it's a massive metal box.
SPEAKER_01
Yeah. [SPEAKER_00] Yeah.
SPEAKER_00
Yeah. [SPEAKER_01] Look at the sensor data. [SPEAKER_01] Yes. [SPEAKER_01] It just shouldn't—radar shouldn't be able to go through it, right? Camera, you can't see in the camera because there's reflections and there's people on the bus. So it's not like you can see through the windows. [SPEAKER_01] Right.
SPEAKER_01
So what is going on? Maybe it's noise or some coincidence. And the first time I saw it, I couldn't actually believe it. It's like, no, there's something. Yeah. It doesn't smell right.
SPEAKER_00
So what actually turned out to be happening is that our peripheral LiDARs bounce under the bus. And there was just a little bit of very noisy reflection of the movement of the person's feet. That was enough for the AI models. They detect it—there's likely a pedestrian there and I'm going to detect it as such.
SPEAKER_01
Yeah. And moreover, there's enough data there to predict what they're going to do. Yes.
SPEAKER_00
[SPEAKER_01] It just blew my mind. [SPEAKER_01] Is this the perfect example to explain what we were talking about earlier? The value of sensor fusion across a sensor suite, but then secondly, building, I mean, relatedly, [SPEAKER_01] And there was just a little bit of very, very noisy reflection of the movement of the person's feet. [SPEAKER_01] That was enough for the AI models.
SPEAKER_01
[SPEAKER_00] They likely detect a pedestrian there and I'm going to detect it as such. Yeah. And moreover, there's enough data there to predict what they're going to do. Yes.
SPEAKER_00
[SPEAKER_01] It blew my mind.
SPEAKER_01
Is this the perfect example to explain what we were talking about earlier? The value of sensor fusion across a sensor suite, but then secondly, building an intermediate representation of what's going on, where if you're just dealing with pixels, the person behind the bus does not exist in pixel space. And so you need to have some representation of the world that exists to be able to reason about the person behind the bus. I think it's an example where giving it an intermediate representation to boost the level of performance of all parts of the model is what's happening here. Right. Just imagine solving this problem with a black box, purely open loop, imitative system.
SPEAKER_01
[SPEAKER_00] Yeah.
SPEAKER_00
[SPEAKER_01] Be hard to impossible. [SPEAKER_01] Is it impossible? No.
SPEAKER_01
Yeah.
SPEAKER_00
[SPEAKER_01] But in practice, what would it take to achieve that level of performance?
SPEAKER_01
[SPEAKER_00] Yes. Very, very difficult. [SPEAKER_00] What metrics can you share on just where the business is at today in terms of rides, revenues, cars on the roads? [SPEAKER_00] We have about 3,000 cars on the roads. [SPEAKER_00] We're doing about half a million rides per week. That translates to about over 4 million fully autonomous miles per week. We are operating in a fully autonomous mode in 11 cities in the U.S. And 10 of those we have riders, public riders. [SPEAKER_00] What's the ghost city? The ghost city is Nashville. Okay. We just started there. Okay. So we just opened it up to riders in four new cities in one day.
SPEAKER_01
That was one of those super exciting moments where I thought back to the history. How long did it take us from the first time we started fully autonomous rider-only operation to the first time we had external riders in four cities? That's about eight years. And then just the other week, we just launched four in one day. Yes. Yes. It seems now clear that in 15 years, most miles that are driven will be autonomous. There'll be some burning period and there's lots of old cars on the road. I think it'll actually take a little while and some of that will be by level four, level five systems expanding in new cities and that expansion continuing.
SPEAKER_01
Some of it will be existing driver assist systems and getting up to level two and level three and existing systems across current car brands getting more and more capable.
SPEAKER_00
Mm-hmm.
SPEAKER_01
[SPEAKER_00] What do you think that working your way up from the lower levels versus working your way expanding from existing products like Waymo? [SPEAKER_00] What will that convergence look like? [SPEAKER_00] Yeah. [SPEAKER_00] Because we're going to eat it from both sides. [SPEAKER_01] I don't believe we will. Hmm. And I actually think this- That's a great answer. Yeah, cars will get smarter. [SPEAKER_00] There's going to be advances in driver assist systems. Yep. And if there is at the same time from level four autonomy, there is simplification and the sensors of today are not going to be the sensors of tomorrow. [SPEAKER_01] They'll be much more integrated.
SPEAKER_01
[SPEAKER_00] They'll be simpler. [SPEAKER_00] There'll be much lower cost. So from that perspective, there is a path of convergence. [SPEAKER_00] Yep. And there's also a path of convergence from the product lines. Mm-hm.
SPEAKER_00
[SPEAKER_01] There's right hailing and you can take a ride through the Waymo app today. Eventually, they'll be on your personal car. So that I see. [SPEAKER_01] And talk about the technology and the- [SPEAKER_01] I see it just as fundamentally two different problems. [SPEAKER_01] There's driver assist systems and then there is full autonomy.
SPEAKER_01
And I think it's deceptive to think of them as incremental on one spectrum of complexity. Okay. But you think one cannot work one's way up from driver assist systems to full self-driving? [SPEAKER_00] You think you have to start building a full self-driving system? [SPEAKER_00] I think you have to tackle. If I think about the hardest parts of building a fully autonomous rider-only system, they are very different from what you do for a driver assist system. Yep. [SPEAKER_01] Right. [SPEAKER_01] And of course, some work in this space helps you, right? So I'm not saying you can't make the jump, but it is a qualitative jump. [SPEAKER_00] Yes.
SPEAKER_01
[SPEAKER_00] When can I buy a Waymo so that I don't need to wait for it when I want to go? I can just, when I'm ready, I can walk out the door and it's there. [SPEAKER_00] I'm not going to give you a date today, but you're not the first person to bring this up as a- [SPEAKER_00] That's my product request. [SPEAKER_00] As a product request. [SPEAKER_00] Yeah. [SPEAKER_00] Do we note it? [SPEAKER_00] Okay. [SPEAKER_00] I'll add it to the list. [SPEAKER_00] Yes.
SPEAKER_00
That waiting for the car, it should be nice just in the garage there and keep your stuff in it and everything. It's not the first time you've heard that request.
SPEAKER_01
I can just when I'm ready, I can walk out the door and it's there. [SPEAKER_00] I'm not going to give you a date today, but you're not the first person to bring this up as a product request. [SPEAKER_00] I'll add it to the list. [SPEAKER_00] You know, that waiting for the car, it should be nice just in the garage there and keep your stuff in it and everything. It's not the first time you've heard that request. So how, it seems to me operationally very intensive and very hard. A self-driving car is actually not self-driving. It takes a village.
SPEAKER_01
And then there's just keeping the cars clean and keeping everything running in that regard. Can you describe what the operational infrastructure that sits behind Waymo looks like? [SPEAKER_01] Sure. And I will say that we are overall in all of those areas on a path of increasing efficiency and automation. So the number of manual steps that one had to do five years ago to launch Waymo versus where we are today is drastically different. But nowadays, if you look at one of our depots as a fully automatically orchestrated dance of autonomous vehicles.
SPEAKER_01
[SPEAKER_00] So the way it looks today is cars will automatically go on there to pick up their riders and serve their trips. If for some reason they need to come back, maybe they're low on energy, maybe somebody left a mess on the car, they will automatically come to the depot.
SPEAKER_01
If it is, so cleaning today is a manual process. So it'll get flagged in the car. We have fleet management systems, say, hey, car number 378 needs cleaning. And we'll actually on the sensor dome, we're able to display on console, show you like a little emoji. And there are people whose job it is to clean the car who still come and clean up. If cleaning is not required and it's just charging, we'll say, pull automatically into a charging stall and we'll say, hey, I need charging. We don't yet have automated charging. In the future, you can imagine that being fully automated. But a person will come in and plug in a cable and the car will charge and say, hey, now I'm ready to go.
SPEAKER_01
[SPEAKER_00] And it will get unplugged and the car will pull out of its parking stall and then go on its merry way. One of the new Porsches, I think it is, has inductive charging, like your iPhone, where you just drive over the charging mat. I was amazed that that works at car scale, but yeah, presumably in the future, they'll just be able to drive on the charging mat. Or do you think robotic plug-in will be easier? We'll see. I don't know. I think there are some questions about efficiency and how that plays into the overall cost and which one will be most cost beneficial remains to be seen.
SPEAKER_01
[SPEAKER_00] How well behaved are the Waymo riding population in terms of not leaving a mess in the car? [SPEAKER_00] We have wonderful riders. We have the most amazing customers in the world. So generally, I would say they are very good. I think there is something about not having a person in the car. It's not somebody else's car, but in some ways, you want to preserve the nice aspects of it. Generally, people want to preserve the nice aspects of it.
SPEAKER_00
It's so clean to begin with. Yeah. I think that's the general trend that we see.
SPEAKER_01
And because there's not somebody else's space, you're in it, it feels like it's your own, so you don't want to mess up your own space. I don't want to speculate too much on the psychology. However, I will say that it varies, and you can imagine a college town on a Saturday night and that's a different distribution. Will I be able to get Waymo at any address that has USPS service in the US, or will there be some head-tail dynamic where Ketchikan, Alaska is just never worth it?
SPEAKER_00
[SPEAKER_01] Eventually it will absolutely. There's no doubt in my mind. I think it's just a matter of when and what modality would make the most commercial sense. It's not a technical problem. Your technology is solved. But if you're in the middle of nowhere and there's just not enough density of trips, does it make sense for Waymo as a ride-hailing service to have cars on standby? Probably not. They can be deployed somewhere else and you probably don't want a horribly bad ETA. This is where a personally owned vehicle that is equipped with the Waymo driver is maybe how you will see it materialize. Relatedly, what will the second-order effects of majority autonomous traffic be like? It feels like a lot of things will work better. You know, when someone merges into a lane very poorly and everyone all the way back has to slam on the brakes, that's anti-social behavior. So it feels like higher quality and more pro-social driving will reduce traffic.
[SPEAKER_01] have cars on standby yes, yes, probably not right? They can be deployed somewhere else, and you probably don't want a horribly bad ETA. This is where a personally owned vehicle that is equipped with the Waymo driver is how you will see it materialized. Relatedly, what will the second order effects of majority autonomous traffic be? It feels like a lot of things will work better. When someone merges into a lane very poorly and everyone all the way back has to slam on the brakes, that's anti-social behavior. Higher quality and more pro-social driving will reduce traffic a little bit, even for the same number of cars on the road, but presumably there'll be other second order effects. We'll want higher throughput traffic lights. How else will things change?
SPEAKER_01
The first thing I think you mentioned is a huge deal. I just need to think about traffic jams. What was that saying that the Navy SEALs and slow is smooth and smooth is fast? That's what traffic jams are like. You accelerate abruptly, then you come to a stop. Sometimes you have traffic, like what happened.
SPEAKER_01
Yes. An old lady crossed the road three hours ago and we still have the standing wave there, right? So if everybody was a smooth, predictable driver and a consistent driver, you would still have those traffic jams and at the time off. But the time constant to clean it out would be very different. But longer term, things like parking lots. Right now, if you look at what is our most interesting. Yes. Pieces of land allocated to parking lots and garages. Why is that? Because your car is just sitting there 90 percent of the time. If more cars become fully autonomous, then there's no that right. Imagine what you can do with your favorite city in the world if you don't have to spend that money, that huge fraction of it, on just keeping these chunks of metal sitting around.
SPEAKER_01
[SPEAKER_00] I don't think people often realize how big a deal parking minimums are for the layout of the urban landscape. The coffee shop near where I am would like to have outdoor seating but can't because it would reclaim parking spots.
SPEAKER_01
Yes, it would be wonderful. I only have a few more questions, but I'm curious to talk about Google's relationship with self-driving. Right now, Waymo is the most exciting thing happening at Google aside from everything else AI related, but it was a very long journey to get here. I feel like Google almost started working on it too early because you were saying there's been a bunch of recent enabling technologies. So did it require Google starting when it did, so early? Or could one have spun up this project in 2015, 2020? And how did Google keep the faith when it always felt like it was perennially two years away?
SPEAKER_01
No, I just have to give credit, huge kudos and gratitude to Sergey and Alphabet leadership. It is part of the culture and DNA of the company to have that vision and have the stamina and conviction to go the distance. To the other part of the question, was it too early? I don't know. What we've been seeing, clearly all of the breakthroughs that we've seen over the years have changed how we're building the system. But the complexity of the problem is such that you need to go through these iterative cycles. We've seen many waves of technology. There were breakthroughs in 2013 when ImageNet came around, and that is the right time to start a self-driving company. And transformers came around and VLMs, and all of those are super powerful and have applications in other spaces. In the digital world, they certainly have an impact on our AI in the physical world. But there are no silver bullets. They drastically reshape that early part of the curve. It's always been that the nature of this problem is very easy to get started. It's deceptively easy to get started, but it is super hard to go the full distance. There's the standard engineering rule of thumb that every next nine takes 10x more. Maybe there is a more optimal path, but I don't see that there's some magical moment where the true complexity of the problem goes away and then you can just take some off-the-shelf components. If that were the case, then I think the industry would look very different today.
SPEAKER_01
Last question I have: you've been promoted a lot at Google. It feels like Google really recognized your talents. What do you think Google does? Google is famously one of the very best in the world at technical talent. The current AI wave more broadly happening is either stuff happening at Google or generally Google alumni. What have you observed firsthand from how Google does this so well?
SPEAKER_01
[SPEAKER_00] I would say Google's culture of not accepting the status quo, having a big vision, and investing in technical talent, the people who can go the distance and realize the vision, that is part of the culture. I think this is what you're seeing with the breakthroughs in AI in the digital world and all of the early investments in transformers and other fundamental technologies, quantum computing, and other efforts as well. Thank you. [SPEAKER_00] Thank you. I appreciate you sharing this, but the transcript appears to contain only speaker labels without any actual spoken content to clean. There are no words, filler words, grammar errors, or punctuation to process.
SPEAKER_01
If you have a transcript with actual dialogue, please paste it and I'll clean it according to the rules you specified. Uh, I think it's an example where giving it kind of, uh, uh, an, using that intermediate representation to boost the level of performance of all parts of the, kind of the model. Yes. Is what's happening here. Mm-hmm. Right. Just imagine, you know, solving this problem with a black box, you know, purely open loop,
SPEAKER_00
Yeah.
SPEAKER_01
imitative system. Be, yeah. Hard to impossible. Is it, you know, impossible? No. Yeah. But in practice, what would it take to achieve that level of performance?
SPEAKER_00
Yes. Very, very difficult. What metrics can you share on just where the business is at today in terms of
SPEAKER_01
rides, revenues, cars on the roads?
SPEAKER_00
Um, we have about 3,000 cars, um, on the roads. We're doing about half a million, uh, rides, uh, per week.
SPEAKER_01
Uh, it, that translates to about, you know, over 4 million fully autonomous, uh, miles per week. Uh, we are operating in a fully autonomous mode in 11 cities, uh, in the U.S. Uh, and 10 of those, uh, we have, uh, riders, public, you know, riders, uh, and-
SPEAKER_00
What's the ghost city?
SPEAKER_01
The ghost city is Nashville. Okay. We just started there. Okay. So we just, uh, uh, opened it up to riders in four new cities in one day. So like it, that was one of those, you know, little, but super exciting moments where I, you know, I thought back to the history. Like how long did it take us from the first time we started fully autonomous rider-only operation to the first time we had external riders in four cities? That's about eight years. Mm-hmm. And then just, you know, like the other week, we just launched four in one day. Yes. Yes. It seems now clear that in 15 years, most miles that are driven, uh, will be autonomous.
SPEAKER_01
Uh, like there'll be some burning period and, you know, there's lots of old cars in the road. I think it'll actually take a little while and some of that will be by level four, level five systems expanding in new cities and, uh, that expansion continuing. Some of it will be, you know, you referenced existing driver assist systems and kind of getting up to, uh, you know, level two and level three and existing systems in across current car brands
SPEAKER_00
getting more and more capable. Mm-hmm. What do you think that working your way up from the lower levels versus working your way expanding from existing products like Waymo? What will that convergence look like? Yeah. Because we're going to eat it from both sides.
SPEAKER_01
I don't believe we will. Hmm. And I actually think this- That's a great answer. Uh, uh, yeah, cars will get smarter.
There's going to be, you know, advances in driver assist systems. Yep. Uh, and if there is, you know, at the same time from level four autonomy, you know, there is simplification and, you know, uh, the sensors of today are not going to be the sensors of tomorrow. So they'll be much more integrated.
SPEAKER_00
They'll be simpler. There'll be much lower cost.
SPEAKER_01
So from that perspective, they're going to, you know, there is a path of convergence.
SPEAKER_00
Yep.
SPEAKER_01
And there's also, you know, a path of convergence from, you know, the product lines. Mm-hm. There's, you know, right hailing and what, you know, you can take, you know, a ride through
SPEAKER_00
the Waymo app today. You know, eventually, they'll be on your personal car. So that I see.
SPEAKER_01
And talk about the technology and the- I see it just as fundamentally two different problems. There's driver assist systems and then there is full autonomy. And I think it's deceptive to think of them as kind of incremental, you know, on one spectrum of complexity. Okay. But you think one cannot work one's way up from driver assist systems to full self-driving?
SPEAKER_00
You think you have to start building a full self-driving system? I think you have to tackle, if I think about the hardest parts of building a fully autonomous, you know, rider-only system, they are very different from, you know,
SPEAKER_01
what you do for a driver assist system. Yep. Right. And I, of course, you know, some work in this space helps you, right? So, you know, I'm not, I don't want to say you can't make the jump, but it is a qualitative jump.
SPEAKER_00
Yes. When can I buy a Waymo so that I don't need to wait for it when I want to go?
SPEAKER_01
I can just like, when I'm ready, I can walk out the door and it's there.
SPEAKER_00
I'm not going to give you a date today, but you're not the first person to bring this up as a- That's my product request. As a product request. Yeah. I'll, I, uh, do we note it? Okay. I'll add it to the list. Yes. You know, that waiting for the car, it should be nice just in the garage there and keep your stuff in it and everything. It's not the first time you've heard that request. Um, so how, it seems to me operationally very intensive and very hard. Like a self-driving car is actually not self-driving. It takes a village.
SPEAKER_00
It takes a village to get a village to get a village to get a village to get a village to
SPEAKER_01
get a village to get a village to get a village to get a village to get a village to get a village. And then there's just like keeping the cars clean and, you know, keeping everything running in that regard. And so, can you describe just what the operational infrastructure that sits behind Waymo looks like? Sure. And I will say that we are overall, you know, in all of those areas on a path of, uh, increasing efficiency and automation. Yep. Right. So, you know, the number of manual steps that, you know, one had to do, you know, five years ago to, you know, uh, you know, launch Waymo, um, uh, versus where we are today is drastically different, right? So, um, yeah, yeah.
SPEAKER_01
But nowadays, if you look at one of our depots as like a fully automatically orchestrated, you know, dance of autonomous vehicles.
SPEAKER_00
So the way it looks, uh, uh, you know, what it looks like today, uh, is, um, cars will automatically,
SPEAKER_01
you know, go on there, you know, to pick up their riders, you know, serve their trips. Uh, if for some reason, you know, they need to come back, um, you know, maybe they're low on energy, uh, maybe somebody, you know, left a mess on the car, they will, you know, automatically come to the depot, right? Yes. If it is, so cleaning today is a manual process, right? So it'll get flagged in the car, you know, we have fleet management systems, say, hey, you know, car, you know, number, you know, 378 needs cleaning. And we'll actually, uh, on the sensor dome, we're able to, you know, display on console, you know,
SPEAKER_00
show you like a little, you know, emoji. Put the hand up, yeah. Yeah.
SPEAKER_01
And, you know, there's, you know, people whose job it is to clean the car still come, you know, clean up. If that's, you know, cleaning is not required and it's just charging, you know, we'll say, you know, pull automatically pull into a charging sole, uh, and we'll say, hey, you know, I need charging. We don't yet have automated charging. In the future, you can imagine that being fully automated, right? But, you know, a person will come in and, you know, plug in a cable and the car will charge and say, hey, you know, now I'm ready to go.
SPEAKER_00
And, you know, it will get unplugged and the car will, you know, pull out of the, you know, its parking stall and then go, you know, on its merry way. One of the new Porsches, I think it is, has inductive charging, like, uh, just like your iPhone,
SPEAKER_01
where you just drive over the charging mat. I was amazed that that works at car scale, but yeah, presumably in the future, they'll just be able to drive on the charging mat. Or do you think just robotic plug-in will be easier? We'll see. We'll see. I don't know.
SPEAKER_00
I think there's, you know, some questions about, you know, efficiency and, you know, how that plays into the overall cost and which one will be, you know, most cost beneficial, you know, remains to be seen, I think. How well behaved are the Waymo riding population in terms of not leaving a mess in the car?
SPEAKER_00
We have wonderful riders. We have the most amazing customers in the world. So, uh, generally, I would say, they, uh, are very good. I think, you know, there is something about, you know, I talked about not having, you know, a person in the car, it's not somebody else's car, but in some ways, you kind of, like, want to preserve the, the, I think, generally, people want to kind of, preserve the nice aspects of it, right, and, kind of, think of, as, uh, It's so clean to begin with, this, yeah, yeah, it's kind of, like, you know, uh, I think that,
SPEAKER_01
that's the general trend that we see, right, and, like, because there's not somebody else's space, you know you're in it it feels like it's your own yeah so you don't like want to mess up you know your own space i think i i mean i i don't want to yeah uh speculate too much on the psychologist thing however i will say that it varies and you can imagine you know a college town on a you know saturday night and yeah that's a different distribution yes yes will i be able to get waymo at any address that has usps service in the us or will there be some head tail dynamic where ketchkin alaska is just never worth it uh eventually it will absolutely right there's no doubt in my
SPEAKER_01
mind i think it's just a matter of uh when and you know what modality would make the most commercial sense right this is right you know it's it's not a technical problem i mean your technology is solved yes but then you know if you're in the middle of nowhere and there's just not enough density of the trips does it make sense for the right healing service that you know waymo is running to you know have cars on standby yes yes probably not right they can be deployed you know somewhere else and you probably don't want you know a horribly bad eta and this is where you know personally owned vehicle
SPEAKER_01
that is equipped with the waymo driver is maybe you know how you will see it materialized relatedly what will the second order effects of say majority autonomous traffic be like it feels like a lot of things will work better where as you say you know when someone merges into a lane very poorly and everyone all the way back yeah has to you know slam on the brakes that's kind of anti-social behavior and so it feels like higher quality and more pro-social driving will just i mean basically reduce traffic a little bit even for the same number of cars on the road but presumably there'll be other second order
SPEAKER_01
effects like we'll want higher throughput traffic lights and yeah how else will things change change so the first thing i think you know we that you mentioned is uh i think that's a that's a huge deal i just need to think about uh traffic jams yeah okay what was what's that saying that the the navy
SPEAKER_00
seals and uh slow is smooth and smooth is fast right that's what like you know traffic jams are like you
SPEAKER_01
you accelerate abruptly then you come to a stop and you sometimes you have a traffic like what happened yes well like you know an old lady crossed the road three hours ago and we still have the standing wave there right so if everybody you know was a kind of a smooth predictable driver and a consistent driver and you would still have those you know uh traffic jams and at the time off yeah but then the time constant to clean it out i think would be very different but uh longer term and you know things like uh parking lots right right now if you look at you know what is our most interesting yes yeah uh
SPEAKER_01
pieces of land allocated to you know it's parking lots it's garages and why is that well because again you know your car is just sitting there 90 of the time right uh if you know more cars become fully autonomous then there's no you know that right and like then imagine just imagine what you can do yes with you know your favorite city yeah in the world if you don't have to spend that money
SPEAKER_00
that huge fraction of it on you know just just keeping your these chunks of metal sitting around yeah i don't think people often realize how big a deal parking minimums are for the layout of the urban landscape the coffee shop near where i am would like to have outdoor seating but can't because it would reclaim parking spots yeah wouldn't it would be wonderful yeah i only have a few more questions but i'm curious to talk about google's relationship with um uh self-driving where uh again it feels like um right now waymo is aside from everything else a.i.i. related kind of the most
SPEAKER_00
exciting thing happening at uh google but it was a very long journey to get here i mean i feel like you
SPEAKER_01
could say uh that google almost started working on it too early because you were saying there's been a bunch of recent enabling technologies and so did it require google starting when it did so early or could one have spun up this project in 2015 2020 and then how did google keep the faith when it always felt like it was perennially two years away yeah no i i know on the latter part i just have to give credit
SPEAKER_00
huge kudos and gratitude to you know larian sergey uh and you know alphabet leadership center company uh
SPEAKER_01
it it it is part of the culture and the dna of the company is to have that vision and have the stamina uh and conviction to go the distance uh so um to the other part of the question uh yeah was it too early i don't know i think what we've been seeing you know clearly all of the breakthroughs that we've seen over the years have changed you know how we're building the system uh but it the complexity of the problem is such that like you need to go through these eater of cycles right it's not you know still and we've seen many waves of technology right there's you know breakthroughs in you know 2013
SPEAKER_01
image net came around and there's never okay like that is the right time to start yeah bsl driving company and you know transformers came around and you know vlms and that and it all of those are uh super powerful and you have applications and other um spaces like in the yeah in the digital world they certainly have an impact uh on you know our ai and the physical world uh but there are no silver bullets right they kind of they drastically reshape that early part of the curve yes like then it's always been that the nature of this problem is it's very easy to get started it's deceptively
SPEAKER_01
easy to get started but it is super hard to go you know the full distance and get it's you know
SPEAKER_00
the number of knives right that you have to get like there's the standard you know engineering rule of thumb that you know every next nine takes you know 10x more so i um yeah maybe there is a more
SPEAKER_01
optimal path but i don't see there's you know that there's some magical moment where the true complexity of the problem goes away and then you can just take some off-the-shelf components and your business if that were the case then i think the industry would look you know very different today yeah yeah last question i have you've been promoted a lot at google it feels like google really recognized your talents just what do you think google does like google is famously one of the very best in the world at technical talent and say you know um the the current ai wave more broadly happening you know is
SPEAKER_01
either stuff happening at google or generally google alumni um but just what have you observed uh
SPEAKER_00
uh first hand from how google does this so well yeah i would say uh google you know that that culture of google of not accepting the status quo having you know a big vision and uh investing in technical talent the people who can you know go the distance and realize the vision that is part of the the culture i think this is what you're seeing and uh with the you know the breakthroughs in ai uh in the digital world and like all of the early investments in you know transformers and under you know other fundamental technologies uh you know quantum computing yeah and you know i guess we are not unlike those efforts
SPEAKER_01
as well to measure thank you yeah thank you so
SPEAKER_00
so so
SPEAKER_01
so so Thank you.