4 ms·

Could he really say anything else given his current position? Car manufacturers and VCs have pumped enormous cash into this space for over a decade and have wha

Could he really say anything else given his current position? Car manufacturers and VCs have pumped enormous cash into this space for over a decade and have what amounts to vaporware to show for it. Assisted highway cruise and crash avoidance are very good features, but we are perpetually “a few years away” from the promise of significant autonomy.

sneak 7y ago
To be fair, that was the state of speech recognition, too, until it wasn’t. We’ll get full self-driving eventually, but it may take 5-15 years. It’s hard but not intractable, like supersonic passenger air service. I’m not patient either.
- seibelj 7y ago
  I think everyone here knows that speech recognition tech is pretty garbage, even today. I have trained myself to speak as clearly and plain as possible to (Siri / Alexa / random customer services) and they are still only so-so. Very annoying technology.
- rayiner 7y ago
  Speech recognition still sucks. I have a coworker whose name starts with “S.” Every time she calls on my desk phone and I pick up with “Hi S.” Siri on my iPhone kicks in. It can also never understand the simplest reminders I try to set.
  - JoeAltmaier 7y ago
    Nah that's not speech recognition. That's Siri's owners dedicating only milliseconds of their server time to each request. To save a few bucks. Because they're cheap.
    - sneak 7y ago
      The wakeword detection runs locally, first.
      - JoeAltmaier 7y ago
        Sure but that was just the first example. Then there's failure to get a reminders right etc.
- jacquesm 7y ago
  The state of speech recognition is still quite terrible. For limited domains and very clear diction it works most of the time. For random speech and the way people actually speak it still sucks.
  - sneak 7y ago
    I find it to be mindblowing how accurate youtube’s automatic captioning ends up being most of the time. I also remember all of the attempts in the 90s and early 2000s, and we are at a place now I never thought we would get to.
  - RaceWon 7y ago
    > The state of speech recognition is still quite terrible. I think we should just "move fast and break things"... make all cars self-drivable via speech recognition. I mean that would would serve to limit car ownership... to those with a death wish.
    - bluGill 7y ago
      I don't wish death, which is why I walked to the bus stop. Because of the busy roads I had to cross (without any traffic control) I'm the one who has a death wish. 3000lbs of metal would protect the driver very well when I'm hit, but just ensures I die.
      - RaceWon 7y ago
        > I don't wish death, which is why I walked to the bus stop I live in NY. About 40 miles as the crow flies from NYC. Public transportation is unusable in my town... buses run about once an hour, give or take an hour on weekdays--forget about Sundays. How would you like trying get a sick child to a doctors appointment then going to a pharmacy? Need milk before a snow storm.. good luck. My point is the Vast majority of the USA is not geographically located in metropolitan areas. As a side: Metro North which I used when I worked in queens... is fricken expensive as hell and slow as molasses.
        7y ago
- Jagat 7y ago
  "until it wasn’t" => After the AI winter, we had to wait for the phone revolution which enabled data collection at scale, and google et al's big ad dollars in order to fund speech recognition. Moreover, speech recognition isn't a product you can't launch in a faulty state. You can sell a bad product, make a profit, and iterate without any consequence. It's not the same with self driving cars, which won't be profitable till you get rid of the driver entirely. And you can't launch a ok product and iterate. It has to be almost perfect when it comes to safety. Barring geofencing or legislative changes, you have to get it in a perfect state before you roll it out. When the VC money dies down, and when such big companies are no longer optimistic at having a viable self driving car, the research is going to slow down drastically, till a new wave of funding comes along from some other boom to resuscitate the research.
  - I_like_2lurk 7y ago
    >And you can't launch a bad product and iterate. I think Tesla is doing just this. They are putting self driving features on their cars, putting the onus on the driver to supervise, and using the experiment to gather data. I don't expect their current hardware to ever be able to deliver full autonomous driving, but it will likely provide a useful stepping stone.
  - xoa 7y ago
    >It's not the same with self driving cars, which won't be profitable till you get rid of the driver entirely. And you can't launch a ok product and iterate. It has to be almost perfect when it comes to safety. No, it merely has to be as good as humans, though it should be mildly better. The massive economic and feature advantages it will bring to the table mean that regular people will paper over problems real quick, just as with, well, current cars. In fact if anything I wonder if a higher rate of fatalities and accidents would actually be psychologically better than "almost perfect", because in the modern environment things that are super rare and low risk tend to get inordinate media attention and political response vs things that happen with a certain frequency and thus people become habituated to. I mean, again cars themselves are the big example of that. In America alone, estimates are around 40,000 died last year, which itself was actually around a 1% drop from the previous few years. Which means hundreds of fewer deaths. But that didn't register remotely like a handful of high profile super unlikely dramatic deaths would. But the bonuses are the main things vs any negatives, because humans are good at motivated reasoning to deal with risk when it's something they want. And a reliable included always available personal chauffeur that means things like car access for kids without disruption of parents' schedules, continued car access for the elderly (note, one of the most powerful and reliable voting segments of the electorate), elimination of issues due to drinking/tiredness/phone usage, just plain letting people do other productive things or flat out sleep on trips, etc etc etc., is a powerful set of incentives to quickly move past someone else's "bad luck".
- OnlineGladiator 7y ago
  > To be fair, that was the state of speech recognition, too, until it wasn’t. I regularly yell at Google when I ask it to play me a song, oftentimes even when I spell out words letter by letter, painstakingly slowly, and it still fails me. Speech recognition doesn't work all that well for me. Anyway, I think nuclear fusion is a much better comparison. There is an inflection point where it just works, and until you reach that it's not really useful to anybody except as a research problem. And based on what I hear from my peers who work on self-driving cars (including at Cruise), it's still decades away. Most of them already quit because they just stopped believing it was ever going to happen.
  - jstanley 7y ago
    > I regularly yell at Google when I ask it to play me a song Well, you catch more flies with honey than with vinegar.
    - OnlineGladiator 7y ago
      https://xkcd.com/357/ https://xkcd.com/357/
  - rapnie 7y ago
    A hydrogen car economy is another example of something that has been promised for ages, and is always just around the corner.
noobermin 7y ago
He might be trying to thread a needle of sorts, ride the wave of recent anti-car articles in the press in recent weeks while still managing to get some good press for his start-up.
zaroth 7y ago
There may be a chasm which needs to be crossed before the steering wheel can fully be removed from a vehicle, but... I have been very impressed with the steady incremental real-world improvements to AutoPilot over the last 13 months since I got my Model 3. They have added up to a dramatically more adept system which has become an extremely competent highway driver handling lane changes, merges, being cut off, handling poor visibility, lane closures / traffic cones, even things like scooting slightly over to one side of the lane to give a bit more room when passing a large truck. One example of behavior which has gotten better over the last couple months is during merges where the right lane would appear to widen, prior versions would steer right to try to re-center and then have to steer left to get back in lane, where now the car understands how to keep its path in these situations, even if there isn’t a square-dashed-line drawn to indicate the right side of the right lane continuing. I think the majority of the miles driven on my Tesla have been driven with AutoPilot. (Actually that would be a really cool feature if the car could show that number.) It’s become indispensable to me, and it keeps getting marginally better every month. I’m not saying that it is full autonomy, or even that we will get there on the next couple years. But the cumulative progress over the last 12 months has been incredible, and if it continues anywhere near this rate of advancement over the next 24-36 months, I think AutoPilot will be a truly remarkable achievement — as in, approaching a watershed human achievement akin to spaceflight.
- ezzzzz 7y ago
  Yeah but, if I'm not mistaken, autopilot only works on highways. Self-driving in such an environment may be 'close' to a solved problem, but if we are talking about a service capable of picking people up at their door... yeah, I'm not convinced. >I think AutoPilot will be a truly remarkable achievement — as in, approaching a watershed human achievement akin to spaceflight. Just want to point out the irony in this statement. Ok sure, fully autonomous self driving may be a great technological achievement, but it will be an achievement that benefits rent-seeking corporations rather than humanity at large. Sure, the benefits of such a service could allow car-free living at a price cheaper than leasing/owning a car, but this is a benefit that will be reaped by those with means (middle class) while the poor will still be riding the bus or walking. It would also make investment in public transportation even less likely, which would actually result in a net benefit for all.

sneak 7y ago

To be fair, that was the state of speech recognition, too, until it wasn’t. We’ll get full self-driving eventually, but it may take 5-15 years. It’s hard but not intractable, like supersonic passenger air service. I’m not patient either.

seibelj 7y ago
I think everyone here knows that speech recognition tech is pretty garbage, even today. I have trained myself to speak as clearly and plain as possible to (Siri / Alexa / random customer services) and they are still only so-so. Very annoying technology.
rayiner 7y ago
Speech recognition still sucks. I have a coworker whose name starts with “S.” Every time she calls on my desk phone and I pick up with “Hi S.” Siri on my iPhone kicks in. It can also never understand the simplest reminders I try to set.
- JoeAltmaier 7y ago
  Nah that's not speech recognition. That's Siri's owners dedicating only milliseconds of their server time to each request. To save a few bucks. Because they're cheap.
  - sneak 7y ago
    The wakeword detection runs locally, first.
    - JoeAltmaier 7y ago
      Sure but that was just the first example. Then there's failure to get a reminders right etc.
jacquesm 7y ago
The state of speech recognition is still quite terrible. For limited domains and very clear diction it works most of the time. For random speech and the way people actually speak it still sucks.
- sneak 7y ago
  I find it to be mindblowing how accurate youtube’s automatic captioning ends up being most of the time. I also remember all of the attempts in the 90s and early 2000s, and we are at a place now I never thought we would get to.
- RaceWon 7y ago
  > The state of speech recognition is still quite terrible. I think we should just "move fast and break things"... make all cars self-drivable via speech recognition. I mean that would would serve to limit car ownership... to those with a death wish.
  - bluGill 7y ago
    I don't wish death, which is why I walked to the bus stop. Because of the busy roads I had to cross (without any traffic control) I'm the one who has a death wish. 3000lbs of metal would protect the driver very well when I'm hit, but just ensures I die.
    - RaceWon 7y ago
      > I don't wish death, which is why I walked to the bus stop I live in NY. About 40 miles as the crow flies from NYC. Public transportation is unusable in my town... buses run about once an hour, give or take an hour on weekdays--forget about Sundays. How would you like trying get a sick child to a doctors appointment then going to a pharmacy? Need milk before a snow storm.. good luck. My point is the Vast majority of the USA is not geographically located in metropolitan areas. As a side: Metro North which I used when I worked in queens... is fricken expensive as hell and slow as molasses.
      - 7y ago
Jagat 7y ago
"until it wasn’t" => After the AI winter, we had to wait for the phone revolution which enabled data collection at scale, and google et al's big ad dollars in order to fund speech recognition. Moreover, speech recognition isn't a product you can't launch in a faulty state. You can sell a bad product, make a profit, and iterate without any consequence. It's not the same with self driving cars, which won't be profitable till you get rid of the driver entirely. And you can't launch a ok product and iterate. It has to be almost perfect when it comes to safety. Barring geofencing or legislative changes, you have to get it in a perfect state before you roll it out. When the VC money dies down, and when such big companies are no longer optimistic at having a viable self driving car, the research is going to slow down drastically, till a new wave of funding comes along from some other boom to resuscitate the research.
- I_like_2lurk 7y ago
  >And you can't launch a bad product and iterate. I think Tesla is doing just this. They are putting self driving features on their cars, putting the onus on the driver to supervise, and using the experiment to gather data. I don't expect their current hardware to ever be able to deliver full autonomous driving, but it will likely provide a useful stepping stone.
- xoa 7y ago
  >It's not the same with self driving cars, which won't be profitable till you get rid of the driver entirely. And you can't launch a ok product and iterate. It has to be almost perfect when it comes to safety. No, it merely has to be as good as humans, though it should be mildly better. The massive economic and feature advantages it will bring to the table mean that regular people will paper over problems real quick, just as with, well, current cars. In fact if anything I wonder if a higher rate of fatalities and accidents would actually be psychologically better than "almost perfect", because in the modern environment things that are super rare and low risk tend to get inordinate media attention and political response vs things that happen with a certain frequency and thus people become habituated to. I mean, again cars themselves are the big example of that. In America alone, estimates are around 40,000 died last year, which itself was actually around a 1% drop from the previous few years. Which means hundreds of fewer deaths. But that didn't register remotely like a handful of high profile super unlikely dramatic deaths would. But the bonuses are the main things vs any negatives, because humans are good at motivated reasoning to deal with risk when it's something they want. And a reliable included always available personal chauffeur that means things like car access for kids without disruption of parents' schedules, continued car access for the elderly (note, one of the most powerful and reliable voting segments of the electorate), elimination of issues due to drinking/tiredness/phone usage, just plain letting people do other productive things or flat out sleep on trips, etc etc etc., is a powerful set of incentives to quickly move past someone else's "bad luck".
OnlineGladiator 7y ago
> To be fair, that was the state of speech recognition, too, until it wasn’t. I regularly yell at Google when I ask it to play me a song, oftentimes even when I spell out words letter by letter, painstakingly slowly, and it still fails me. Speech recognition doesn't work all that well for me. Anyway, I think nuclear fusion is a much better comparison. There is an inflection point where it just works, and until you reach that it's not really useful to anybody except as a research problem. And based on what I hear from my peers who work on self-driving cars (including at Cruise), it's still decades away. Most of them already quit because they just stopped believing it was ever going to happen.
- jstanley 7y ago
  > I regularly yell at Google when I ask it to play me a song Well, you catch more flies with honey than with vinegar.
  - OnlineGladiator 7y ago
    https://xkcd.com/357/ https://xkcd.com/357/
- rapnie 7y ago
  A hydrogen car economy is another example of something that has been promised for ages, and is always just around the corner.

noobermin 7y ago

He might be trying to thread a needle of sorts, ride the wave of recent anti-car articles in the press in recent weeks while still managing to get some good press for his start-up.

zaroth 7y ago

There may be a chasm which needs to be crossed before the steering wheel can fully be removed from a vehicle, but... I have been very impressed with the steady incremental real-world improvements to AutoPilot over the last 13 months since I got my Model 3. They have added up to a dramatically more adept system which has become an extremely competent highway driver handling lane changes, merges, being cut off, handling poor visibility, lane closures / traffic cones, even things like scooting slightly over to one side of the lane to give a bit more room when passing a large truck. One example of behavior which has gotten better over the last couple months is during merges where the right lane would appear to widen, prior versions would steer right to try to re-center and then have to steer left to get back in lane, where now the car understands how to keep its path in these situations, even if there isn’t a square-dashed-line drawn to indicate the right side of the right lane continuing. I think the majority of the miles driven on my Tesla have been driven with AutoPilot. (Actually that would be a really cool feature if the car could show that number.) It’s become indispensable to me, and it keeps getting marginally better every month. I’m not saying that it is full autonomy, or even that we will get there on the next couple years. But the cumulative progress over the last 12 months has been incredible, and if it continues anywhere near this rate of advancement over the next 24-36 months, I think AutoPilot will be a truly remarkable achievement — as in, approaching a watershed human achievement akin to spaceflight.

ezzzzz 7y ago
Yeah but, if I'm not mistaken, autopilot only works on highways. Self-driving in such an environment may be 'close' to a solved problem, but if we are talking about a service capable of picking people up at their door... yeah, I'm not convinced. >I think AutoPilot will be a truly remarkable achievement — as in, approaching a watershed human achievement akin to spaceflight. Just want to point out the irony in this statement. Ok sure, fully autonomous self driving may be a great technological achievement, but it will be an achievement that benefits rent-seeking corporations rather than humanity at large. Sure, the benefits of such a service could allow car-free living at a price cheaper than leasing/owning a car, but this is a benefit that will be reaped by those with means (middle class) while the poor will still be riding the bus or walking. It would also make investment in public transportation even less likely, which would actually result in a net benefit for all.