No Priors Ep. 87 | With Co-CEO of Waymo Dmitri Dolgov

[@stri8ted]

Incredible episode. You guys are really hitting it out of the park with the last few episodes. Domain expert guests with domain aware questions.

[@kellymoses8566]

No Priors is such a nerdy name. I love it.

[@NickEllis-nr6ot]

Hearing Dmitri Dolgov's thought and experience was outstanding!

[@Arcticwhir]

28:17 so glad you brought this up, autonomous vehicles are very cool and useful in many ways, but I dont believe it will decrease traffic, it might actually increase traffic definitely through this transition, in the future im not sure. I do think it will make the roads safer - only after a certain threshold of cars on the road. I think we're all quite deaf to the amount of deaths and crashes there are on the roads in america, i literally see 1 or 2 EVERY SINGLE DAY on my way to/from work. 43,000 deaths per year and we collectively lose our shit over 1 plane mishap. We have normalized and accepted this status quo. many are stuck in a mindset of individual responsibility, attributing these accidents to individual choices, rather than a systemic issue. Ofcourse its both, but most think its 80% individual. I don't think autonomous cars are completely bad, lets say a drunk person calling one after a long night, instead of calling an uber - many would feel more comfortable with that.

[@etbadaboum]

If you have autonomous vehicles then you can have autonomous buses! Even Tesla presented a Robovan. So the real future of autonomy without congestion, and hence efficiency and productivity, is public transport!

[@Hamzairshad5]

Please add subtitles without difficult for me and lot of other people to understand Please consider my request and keep it up

[@tubekpful]

CC available, Click CC located next to settings wheel.

[@davab]

scale is bottlenecked by trust and safety.

[@bcan9lick]

WAYMO F1 racing.

[@Notrocketscience101]

I thought i had the speed set on 2x😂 He's not only brilliant but his mouth can't even keep up with his mind.

[@Stock2003]

Love it but the amount of time the guys MHMM is now implemented in my head

[@r.m8146]

I love Waymo, but it's scaling too slowly.

[@user-kl8vr8io3b]

Наш человек ..

[@umuti5ik]

How to talk for 45 minutes without saying anything.

[@hugopennmir]

Turn off the mute button

[@BadWithNames123]

when you are in a vocal fry competition

[@jeffgorchynski]

Apparently It's a requirement for AI experts

It's a west coast thing

[@philipkopylov3058]

A lot of couch experts in the comments, I guess time will tell if Dmitry's bet will pay off or not. Granted that they are backed my a biggest money printing machine in the world, I'd say they might as well have enough time to get it thru the doors. Everyone talks about scale and driving anywhere, but reality is that people live in densely populated areas and cities. Yes waymo cannot drive in corn fields of Ohio, but who needs it there anyways? Regarding the Lidar and mapping, nothing stops waymo from training their own version of end to end model, because they do have the data (in fact even better data since it's been captures with LIDAR, so there is no need to infer "mapping").

[@gemini_537]

Tesla FSD hasn't reached even 1mile of self driving without a human in the driver seat.

[@davab]

sure

[@ARog34]

Do you own a Tesla? Or ridden in one recently?

[@Matt-yp7io]

lol wtf does this guy know about scaling , waymo is like the antithesis of scale haha

[@davab]

lol, nail on the coffin

[@dave_lister]

They have been scaling rapidly. 10x in 1.25 years.

[@Matt-yp7io]

@@dave_lister haha good one

[@overnightclassic2]

Waymo's problem is I feel like they have less than they sound like they do. Just like tesla exaggerates their capabilities, I bet Waymo has a bit of dirty things they are hiding from consumers. When that girl rode "24 hrs in a waymo" it exposed the fact that waymo can only drive for 83 miles because the sensors and compute in the car waste so much power. Not scalable. They claim they can remove everything except vision and still drive. To what extent? I estimate over 4000 watts of compute and sensors. That doesn't work at scale. Waymo doesn't talk about their car lifespan but I suspect they are trashing batteries a minimum of every 3 years based on their use case. Waymo is not scaling not because of safety but because they do not have the capital to scale. They are losing money on every ride, their system is not profitable and they can't just grow and continue to bleed money. Tesla is very far from achieving self driving, but they have one advantage that competitors don't. They can make cars for less money than anyone. BYD is the only real threat because of their low battery prices.

[@zooq-ai]

Dude, you sound like a pathetic Elon simp who simply didn't get what Dmitri was saying. Waymo has closed the loop of building and evaluating. The flywheel has started. They are doubling every 3 months. They have 10000x more data about consumer behavior, pickup and dropoff. Dmitri even specifically mentioned the pitfalls of relying on pure NN

[@charliedoyle7824]

You're likely very wrong. To start, "Waymo is not scaling" is kind of true, but mostly wrong. In short, Waymo isn't yet scaling enough for you to see it because you're not paying attention to how fast they are adding cars and customers lately, and because the Waymo Driver is still a few years away from being ready for full-scale primetime robotaxi. They still are very careful and slow at verifying their Driver in new ODDs like freeways, and they don't yet have their next large-scale robotaxi ready. That will change soon when all necessary kinds of driving is fully verified, and the gen-6 Ioniq 5s start full production and service. Waymo in 2024 is still a pre-business, at the tail end of the development/testing phase, using expensive Jaguars with the prototype gen-5 Driver that are put together at a very small scale. They lose money per ride because the whole support operation is still inefficient; it's small scale with prototype hardware and lots of staff to make sure everything is safe. The cars are expensive, the maintenance is high, remote operations are expensive, they don't have a ton of service area yet, the fleets are limited in those service areas, and they are charging market prices where they lose money per ride to gain customers and prove the business will work when they reach scale. This early phase will likely end in 2026 or 2027, after they fully serve freeways in their early markets and fully serve the Phoenix metro area, over 1000 square miles, along with most or all of the Bay Area metro, and a big percentage of the L.A. metro, plus some amount of Austin and Atlanta. The Jaguar test-fleet with gen-5 Driver will be likely around 2000-3000 cars max. They'll be giving probably hundreds of thousands of rides per week, maybe close to one million miles per day, when this test/dev phase ends. At that point they will start the full-scale business phase, where they switch to Hyundai Ioniq 5 robotaxis with the cheaper and better gen-6 Driver, and buy them by the tens of thousands. These robotaxis are likely to be close to $50,000 total at first. Tesla won't be able to manufacture SAFE robotaxis for less than Hyundai. I think Hyundai/Waymo will have $30k robotaxis at some point in the 2030s. If they want an ultra-cheap little robotaxi like in China, Hyundai or another OEM will make them one. With the Gen-6 Ioniq 5, they will be able to fully serve all of their cities with many thousands of cars in each market, so they'll be giving millions of rides per day. The Driver will be getting so good that the remote interventions per mile will go way down, the driving-range of the robotaxis will be high per charge; and with partners like Uber and Hyundai, and potentially an IPO coming, they'll be able to add new metros quickly. I expect them to be able to add a new metro, like Dallas or Miami, from no operation to full service, in about one to two years by 2030. At some point adding a new metro in one year will be rather easy. Waymo will likely be making money by the early 2030s. Tesla's only serious problem is they won't have a safe ADS over large scale any time soon. They won't solve drivng with a few cheap cameras and just simple end-to-end AI making driving decisions directly off camera images. It won't be robust enough over the massively-long tail of difficult driving that comes with millions of robotaxi miles per day. And even if Tesla develops FSD into a real safe robotaxi driver, which won't happen until at least 2030, they still will need remote operations, customer service, a ride-hailing app that customers want to use, and maintenance/charging/cleaning hubs. There is no serious advantage by Tesla. Nobody can cut corners in robotaxi. It is way too dangerous, and all robotaxis will need similar physical work done at regular intervals. It would be like an airline company cutting corners. That obvously doesn't fly. Robotaxi services will be a commodity business, just like Uber/Lyft, car rentals, or the airlines. They will all be the same more or less.

[@Music_vibes-kw7xr]

If they're not making money, they're not scaling period. Too bad

[@deepblender]

Understanding what it takes to get full self driving to work properly and safely is the basis that is required. As far as I know, they are the only company having achieved that point. They understand what exactly is required and as such they can improve with every generation to optimize for energy consumption or cost or whatever else they deem necessary. If you don't understand how to get something to work, you can't understand how to scale it.

[@overnightclassic2]

@@charliedoyle7824 Things are getting more expensive, not less. Only one reducing cost on their cars is Tesla. Everyone else they are going up because they do not have good enough profit margins with EVs. Nobody except byd can compete with tesla's margins on their vehicles. BYD only because they have good battery prices.

[@martindbp]

The question they don't dare to ask directly: "now that Tesla is showing you can do self driving with cheap sensors, beautiful cars and lots of data, how do you compete?". And the answer is they can't. They have to really hope that Tesla never reaches a high enough reliability, because if they do Waymo is toast. They can't compete with cars that cost $150-200k that looks like a dorky, embarrasing nerdmobile compared to the slick and futuristic Tesla robotaxi for $30k, as well as the costly HD LiDAR mapping that's required for the Waymo.

[@ademolaobaude9651]

So where is Tesla's unsupervised self driving? Please don't mention FSD. Cos that's just another name for ADAS. If Tesla already solved self driving with cheap sensors, they'd be on the streets already right? Waymo on the other hand is actually offering a robotaxi service.

[@Nunya-lz9ey]

@@ademolaobaude9651if Waymo had it solved they would scale.

[@ademolaobaude9651]

@@Nunya-lz9ey they literally are already doing that. Offering rides in 4 cities and testing in many other cities. A simple Google search would have helped you.

[@jeffgorchynski]

@@ademolaobaude9651 Imagine celebrating a victory in the first period.

[@overnightclassic2]

@@ademolaobaude9651 that's not scaling. 4 cities is nothing.

[@Koyaanis]

After watching this interview, I feel even more confident that Waymo’s approach can’t scale. Expensive hardware, predefined maps and remote human supervisors is not practical or economical. They’re shut down within 5 yrs.

[@martindbp]

Yep, as soon as Tesla shows the scaling laws continue to hold, Waymo has nothing.

[@ademolaobaude9651]

Why are you Tesla fanboys obsessed with Waymo scaling. Unless you're invested in Waymo, I don't see why you should care about that. Why not worry about the fact that Tesla is still in level 2 and unsupervised fsd will continue to be a pipe dream

[@Koyaanis]

@@ademolaobaude9651 I didn’t even mention Tesla 😂😂

[@martindbp]

​@@ademolaobaude9651why do Waymo fanboys not care about scale and economics? That's what will win in the end.

[@ademolaobaude9651]

@@martindbp I'm not a Waymo fanboy. I don't think I qualify as one. I use Waymo every other day. I love the service and that's all I care about. While I want them to scale, it's not because I care about their balance sheet, it's because I want other people to enjoy the same experience. We're talking about a trillion dollar company. How much they are burning is the least of my concern

[@Finlaymacnab]

This guy is exuding a nervous vibe. The hosts seemed to be kinda disappointed with his answers which seemd surface deep to me. The Karpathy interview was 10x better.