Oct 24, 2024 · 44m · no-priors
No Priors Ep. 87 | With Co-CEO of Waymo Dmitri Dolgov
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Waymo Co-CEO Dmitri Dolgov joins the No Priors podcast to discuss the technological breakthroughs, multi-modal AI architecture, empirical safety frameworks, and operational strategies driving the mass commercial scaling of autonomous robotaxis.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 22.9% of the talking time here. How this is scored →
The hosts as informed peer 5.6 Guest teaching 5.9 Guest disagreement 1.1 The hosts pushing back 1.4
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 35:45 Dismantling the premise of simpler AV domains
Dmitri strongly rejects the idea that long-haul trucking or deliveries represent simpler autonomous problems, arguing that core physical complexity cannot be bypassed.
Hardest push from the hosts ▶ 15:31 Challenging Waymo's approach via Karpathy's critiqueElad directly confronts Dmitri with Andrej Karpathy's framing that Waymo is an overly hardware-centric approach compared to Tesla's software-driven methodology.
Biggest teaching moment ▶ 38:47 The long-tail 'nines' fallacy with off-the-shelf AIDmitri delivers an insightful breakdown of how off-the-shelf modern VLMs give a false sense of rapid progress while remaining inadequate for the vast long-tail distribution of real-world driving.
The host holds their own ▶ 18:39 Inquiring about sensor pruning analytical scaling lawsElad demonstrates sophisticated AI systems knowledge by asking whether quantitative scaling curves can predict the deprecation of physical sensor modalities relative to model capabilities.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Origins from the DARPA Grand Challenges to Google | 6 | 5 | 1 | 1 | Elad demonstrates solid domain knowledge by citing Sebastian Thrun's Stanford lab lineage and early pre-AlexNet neural net history. Dmitri provides deep historical context from the 2006 DARPA Grand Challenges through Google's early prototyping milestones. | |
| Generational Hardware Skipping and Fleet Evolution | 6 | 6 | 1 | 2 | Elad asks about the recent inflection point to 100k weekly rides and why Waymo chose Arizona over California. Dmitri explains the generational hardware jumps from Firefly to Gen 4 Pacificas and Gen 5 J-Paces, contrasting deployment testing with system stress-testing. | |
| Navigating Operational Complexity and Urban Environments | 6 | 6 | 1 | 1 | Sarah and Elad query the hardest driving environments and technical AI shifts. Dmitri breaks down the interaction of density, speed, and weather, emphasizing how transformers and holistic evaluation infrastructure enabled the Gen 5 breakthrough. | |
| Waymo's Safety Framework and Empirical Performance Metrics | 5 | 7 | 1 | 1 | Sarah asks how Waymo's internal evaluation contrasts with regulatory safety cases. Dmitri outlines Waymo's Readiness and Safety Framework, citing empirical stats from 22 million rider-only miles and an 85% airbag-level crash reduction compared to human benchmarks. | |
| Regulatory Dialogue and Responsible, Trust-Based Scaling | 7 | 6 | 2 | 4 | Elad pushes Dmitri on why rollout speed cannot accelerate faster given proven safety, and invokes Andrej Karpathy's critique contrasting Tesla's software-first approach with Waymo's hardware focus. Dmitri firmly reframes hardware as a sensory advantage while asserting Waymo is fundamentally AI-first. | |
| Multi-Modal Sensing Physics and Sensor Suite Optimization | 6 | 7 | 2 | 2 | Elad asks whether an analytical scaling law allows AV systems to eliminate sensor modalities like LiDAR using vision alone. Dmitri explains the physics of multi-modal sensing and asserts that taking away LiDAR degrades safety below acceptable autonomous licensing thresholds. | |
| Ride-Hailing Strategy, Ecosystem Modalities, and OEM Partnerships | 5 | 5 | 1 | 1 | Sarah probes commercial expansion beyond ride-hailing and CapEx implications. Dmitri clarifies that while ride-hailing is their near-term focus, Waymo builds a generalizable driver stack applicable to trucking and personal vehicles through OEM partnerships. | |
| Societal Transformation, Land Use, and Transit Integration | 6 | 5 | 1 | 1 | Elad and Sarah explore urban land use transformation, car ownership shifts, and OEM relevance. Dmitri details how scale reallocates parking infrastructure and mentions recent initiatives incentivizing AV trips to public transit hubs. | |
| Passenger-Centric Interior Design and Novel Cabin Use Cases | 5 | 4 | 0 | 1 | Elad asks about radical interior form factor changes like bidirectional seating or built-in exercise bikes. Dmitri explains how Gen 6 cabin designs prioritize flat floors, side-sliding doors, and user privacy for work calls without human drivers. | |
| Community Trust Bottlenecks and Robotics Generalization | 5 | 6 | 1 | 1 | Sarah asks whether capital or consumer trust is the primary bottleneck to scaling. Dmitri emphasizes trust asymmetry, and responds to Elad by discussing how Waymo's perception and simulation research translates into generalized robotics. | |
| The Core Complexity Fallacy of Alternative AV Domains | 6 | 7 | 2 | 1 | Sarah asks what lessons emerge from companies attempting supposedly simpler domains like long-haul trucking. Dmitri dismantles the idea of easy AV sub-domains, detailing high-speed freeway edge cases like debris and spinouts. | |
| The "Nines" Problem and the Limits of Off-the-Shelf AI Models | 6 | 8 | 2 | 1 | Dmitri explains the 'nines problem,' pointing out that fine-tuning off-the-shelf VLMs allows anyone to build an impressive prototype quickly, but fails completely on the safety-critical long-tail required for driverless scale. | |
| Simulation Architecture and Surpassing Imitation Learning | 6 | 7 | 1 | 2 | Sarah asks whether Waymo's iteration cycle is purely data flywheel curation or architectural. Dmitri explains why simple open-loop imitation learning inevitably plateaus, necessitating closed-loop simulation and intermediate synthetic representations. | |
| Transitioning from Validation to Scaling the Safety Mission | 4 | 4 | 0 | 0 | Elad asks where the industry goes next. Dmitri summarizes Waymo's transition from foundational technology validation to commercial optimization and geographic scaling. |
Statements from this episode (19)
Google's self-driving project pivoted from ADAS to full autonomy in 2013
“And actually our first product that we thought was going to be viable was, you know, what nowadays you would call kind of an advanced driver system, right? And we had some expectations of, you know, a small number of years that it would take for us to get ther…”
Dolgov: Waymo completes over 100,000 trips and one million miles weekly
“You know, a 100,000 trips per week, more than a million miles per week, and growing, you know, exponentially.”
Dolgov: Waymo evaluates autonomy across operating domains rather than zip codes
“The way we think about it in terms of the development and evaluation of the driver is kind of in the space of the operating domain, not necessarily areas or zip codes. And then you kind of take cities and you map that to the operating domain and you deploy it,…”
AI model architectures are only enablers without the surrounding data flywheel
“The architecture is an enabler, but really to make it work at the level that we care about, you need like everything around it. The data engine, the whole, you know, flywheel of, you know, training the system, evaluating it. And you kind of have to think about…”
Dolgov: Waymo experiences 6x fewer airbag-deployment collisions than humans
“So for airbag, you know, deployment type collisions, we're about a factor of six better than human drivers.”
Dolgov: Swiss Re study showed 100% bodily injury claim reduction for Waymo
“We have done a study with Swiss RE, the global kind of thing, the biggest global ratio in the world. And we partnered with them, we shared the data, they've done the analysis, and they found that, you know, for damage, We had about a four X reduction versus th…”
Waymo doubled weekly autonomous miles from 50,000 to 100,000 in three months
“We are scaling exponentially. It took us about three months. To get from 50,000 miles to a 100,000 miles, right?”
Waymo's autonomous driving technology is fundamentally centered on AI, not hardware
“It's all about AI. Full stop. Like we talked about, you know, a few big breakthroughs that allowed us to get to where we are today. You know, content, transformers, big Ness now, you know, most recently kind of combining the Waymo AI with the congenital knowle…”
Dolgov: Waymo has achieved autonomous driving quality and shifted to optimization
“So far we have, you know, achieved Good quality of the driver. We have built all the machinery to evaluate it and know what it takes. So now for us, it's an optimization, right? It's a, it's an optimization, simplification.”
Dolgov: Waymo's sixth-gen hardware focuses on simplification and cost reduction
“Every generation of the hardware, you know, would increase capability, but also simplify drastically. And then the cost comes down, you know, every generation. So that was true, you know, for the previous generations we made a big jump from the you know, the p…”
Dolgov: Camera-only systems are not safe enough for full autonomy
“Same, same for us. We can, you know, take something away and we can answer that question. Like, is the performance, can you still drive? Of course. Is the performance good enough? For full autonomy, and is it good enough for our bar of readiness and safety? An…”
Waymo focuses on robotaxis but plans future trucking and delivery deployments
“Right hailing is the main one that we're focusing on. So we're, you know, focused on technology. We're focused on the product. We are learning from our users. We're every day, we're earning trust. And we're setting up the ecosystem of, you know, partnerships i…”
Dolgov: Waymo is incentivizing rides to public transit hubs
“In fact just in the last couple of days, we announced something that we're doing where we are incentivizing people to take Waymos in the cities where we operate to public transit hubs, and then, you know, everybody benefits.”
Dolgov: Consumer and community trust is the main bottleneck to scaling Waymo
“Primarily it's the latter. So we've always, you know, our playbook has been to you know, go about it responsibly and gradually and earn trust every step of the way and have this transparent dialogue. Again, this is a very new thing, new technology, new product…”
Dolgov: Waymo remains laser-focused on autonomous driving over general robotics
“We are very focused on the, you know, trillions of miles where we can have the positive benefits. So for us like I focus, I think focus is very, very important. So we are being very, you know, laser focused on driving.”
Dolgov: Trucking and delivery AVs cannot bypass full autonomy's core complexity
“And then they're a little bit different, but if we're talking about full autonomy, maybe there's second order differences, But the first order of complexity is still there. You can, ah, like the, if you think about, you know, the core, the heart of the problem…”
Dolgov: Autonomous vehicle complexity lies entirely in long-tail reliability 'nines'
“The complexity is in the long tail of the many, many, many nines. And you don't see that if you go, you know for a prototype, if you go for, you know, a driver assist system and this is where, you know, we've been spending all of our, that's the only hard part…”
Dolgov: Off-the-shelf AI models cannot achieve human-beating driverless safety records
“The power of transformers, the power of realism is mind-blowing, right? So with just a little bit of effort, you get something on the road, and it works. You can, you know, drive, I don't know, 1000 of miles, and we just, it will blow your mind. But then is th…”