3 ms·

Nice to see Hotz own up to his own hubris here and share what seems like a pretty level-headed opinion on the space after trying to put his money where his mout

Nice to see Hotz own up to his own hubris here and share what seems like a pretty level-headed opinion on the space after trying to put his money where his mouth was. EDIT: Having watched further in the video I think the best point he makes here is that the cost part of it simply doesn't make sense - the cars are still super expensive and good level 4 autonomy relies on things like carefully generated, high-precision map data that can easily become inaccurate and also are very expensive to produce.

zouhair 7y ago
Can you imagine the guy that invented the locomotive that told investors that it is the future you only need to build specific rails for them to work and you have to build miles and miles of them and no one can use those rails as they are meant only for trains and you also need to maintain them.
- kevingadd 7y ago
  We already have trains and roads, if you're proposing another massive infrastructure investment where the consequences of underinvestment include 'people get hit by self-driving cars due to outdated map data', the bar you have to hit there is a lot higher than 'here's my idea for this locomotive, we build some rails in empty grassland and it'll take you somewhere 10 times faster than before'. It's not comparable. Self-driving is about convenience and PERHAPS scale, it's not presently enabling entirely new scenarios. And if the car is super expensive (as he points out, and as it currently is), it leads to a situation where the self-driving cars are owned by central powers instead of individual civilians, so now the cost equation for driving is different and you're giving up control over your own vehicle. They were talking about releasing waymo cars with no steering wheel!
  - zouhair 7y ago
    Not really a counter-argument. The real counter-argument would be trains are not for private use, but cars are.
- gmueckl 7y ago
  Rails and train cars were relatively old technology. By the time the steam locomotive was invented, there were already horse drawn trains [1]. So your example is not quite backed by history. [1] https://en.m.wikipedia.org/wiki/List_of_horse-drawn_railways https://en.m.wikipedia.org/wiki/List_of_horse-drawn_railways
  - zouhair 7y ago
    Even now we have a lot of maps, roads, electronics and much more already made.
darawk 7y ago
That's a terrible point, though. The way the costs come down is by slowly chipping away at them, and scaling up production of all the important components. This is how all new technologies arrive.
- kevingadd 7y ago
  That's true for things like LIDAR scanners, but if in the end you need a bunch of cars (human-operated for a while, maybe forever?) constantly mapping and re-mapping roads, you can only optimize the cost of that so much. If you need a bunch of humans checking and massaging data, human wages only go so low. It's important to note that in this video he's not calling it impossible, just calling it a scam. Whether you agree with that is up to you - I think he's overstating it - but it is definitely the case that we've been made promises over the last 5+ years by many people (including him) that didn't hold up and the self-driving cars we have now are less autonomous than promised and more expensive than we were told to expect.
  - darawk 7y ago
    Right, but there are potentially more efficient ways to go about that. Particularly if you have the buy-in of local governments to instrument their roads for you.
    - kevingadd 7y ago
      Their current method is pretty efficient - getting the general public to annotate photos to mark stoplights, etc. So they've solved the cost problem by making the entire planet pay the cost for them. I think a company like Google can do this without expecting us to directly pay the price for it. And a truly functional, high level automated driving system with high safety WOULD be highly valuable, so maybe the price is fine? But it's hard to deny that if you look at all the pieces, as a whole, it is a very, very expensive proposition.
      - Retric 7y ago
        Compare the costs of physically building and maintaining roads vs having a sensor equipped vechile drive past them. That’s orders of magnitude cheaper and thus no real impediment. Assuming it’s both needed and people are willing to pay for self driving cars. 4.1 million miles of road in the US vs 263.6 million registered vehicles. Assuming it costs 10$ / mile you could make every road in the US every day for about 60$ / year per car. And that’s with insane cost estimates anything reasonable is going to end up being cheap.

zouhair 7y ago

Can you imagine the guy that invented the locomotive that told investors that it is the future you only need to build specific rails for them to work and you have to build miles and miles of them and no one can use those rails as they are meant only for trains and you also need to maintain them.

kevingadd 7y ago
We already have trains and roads, if you're proposing another massive infrastructure investment where the consequences of underinvestment include 'people get hit by self-driving cars due to outdated map data', the bar you have to hit there is a lot higher than 'here's my idea for this locomotive, we build some rails in empty grassland and it'll take you somewhere 10 times faster than before'. It's not comparable. Self-driving is about convenience and PERHAPS scale, it's not presently enabling entirely new scenarios. And if the car is super expensive (as he points out, and as it currently is), it leads to a situation where the self-driving cars are owned by central powers instead of individual civilians, so now the cost equation for driving is different and you're giving up control over your own vehicle. They were talking about releasing waymo cars with no steering wheel!
- zouhair 7y ago
  Not really a counter-argument. The real counter-argument would be trains are not for private use, but cars are.
gmueckl 7y ago
Rails and train cars were relatively old technology. By the time the steam locomotive was invented, there were already horse drawn trains [1]. So your example is not quite backed by history. [1] https://en.m.wikipedia.org/wiki/List_of_horse-drawn_railways https://en.m.wikipedia.org/wiki/List_of_horse-drawn_railways
- zouhair 7y ago
  Even now we have a lot of maps, roads, electronics and much more already made.

darawk 7y ago

That's a terrible point, though. The way the costs come down is by slowly chipping away at them, and scaling up production of all the important components. This is how all new technologies arrive.

kevingadd 7y ago
That's true for things like LIDAR scanners, but if in the end you need a bunch of cars (human-operated for a while, maybe forever?) constantly mapping and re-mapping roads, you can only optimize the cost of that so much. If you need a bunch of humans checking and massaging data, human wages only go so low. It's important to note that in this video he's not calling it impossible, just calling it a scam. Whether you agree with that is up to you - I think he's overstating it - but it is definitely the case that we've been made promises over the last 5+ years by many people (including him) that didn't hold up and the self-driving cars we have now are less autonomous than promised and more expensive than we were told to expect.
- darawk 7y ago
  Right, but there are potentially more efficient ways to go about that. Particularly if you have the buy-in of local governments to instrument their roads for you.
  - kevingadd 7y ago
    Their current method is pretty efficient - getting the general public to annotate photos to mark stoplights, etc. So they've solved the cost problem by making the entire planet pay the cost for them. I think a company like Google can do this without expecting us to directly pay the price for it. And a truly functional, high level automated driving system with high safety WOULD be highly valuable, so maybe the price is fine? But it's hard to deny that if you look at all the pieces, as a whole, it is a very, very expensive proposition.
    - Retric 7y ago
      Compare the costs of physically building and maintaining roads vs having a sensor equipped vechile drive past them. That’s orders of magnitude cheaper and thus no real impediment. Assuming it’s both needed and people are willing to pay for self driving cars. 4.1 million miles of road in the US vs 263.6 million registered vehicles. Assuming it costs 10$ / mile you could make every road in the US every day for about 60$ / year per car. And that’s with insane cost estimates anything reasonable is going to end up being cheap.