Hacker News Reader

Comment by wg0 | Hacker News Reader

At the moment - self driving technology is as good as LLMs are. When they work, they work wonderfully and when they don't, they don't but the problem is that there's no sure shot way of knowing when they'll work and when they won't.

Isn't it or something has changed lately?

View on HN

The problem is that, unlike with LLMs where a human always can and should fact check/reword before using, there is no such option with self driving. People's lives are literally directly at stake if things go wrong.

For limited scenarios such as geofenced to a city with good weather, it seems that self driving cars (at least Waymo's tech) is good enough, baring some small lapses which are more funny than dangerous. As a general thing everywhere? To me it doesn't look like we're even close (like 10 years) to being there. Also, the problems actually instead of supposedly being solved by self-driving are... very minimal. There will be no noticeable improvement in things that matter the most on the global scale (pollution, congestion, time spent), only marginal practical and maybe financial improvements (there might be more and cheaper taxis).

Self-driving electric trains are already a solved problem that could solve probably the majority of transportation needs, congestion and pollution all in one go.

Cheap taxis could lead to people abandoning car ownership and this a huge reconfiguration of American cities. (if they end up cheaper than self owned cars)

If everyone still takes the taxi everywhere, what would be the reconfiguration about and where would it come from? Everyone will be still stuck in traffic due to the physical impossibility of fitting everyone in an individual vehicle on the roads at the same time.

This is false.

It's been shown through computer modeling that road congestion emerges as a function of both vehicles/road segment/time unit and human driving behavior.

Remove the humans from the driver's seat, and your effective road capacity goes up significantly without doing anything else.

Do you have links to those computer models? Do they take into account induced demand?

And still, even if autonomous cars are drastically more efficient, with a passenger per car the space use is still drastically less efficient (even the biggest highways will have a fraction of the capacity of a proper metro).

> Do you have links to those computer models?

You're more than capable of doing this research yourself.

> with a passenger per car the space use is still drastically less efficient

Sorry, this simply isn't the metric for success. It might be your metric, but it certainly isn't the population's metric. A lot of people want to travel comfortably in cars, "mindless pursuit of space efficiency" be damned.

Ubiquitous self-driving vehicles will allow us to eventually eliminate most congestion, and you're just going to have to accept that cars aren't going anywhere.

How would they eliminate congestion when they're so space inefficient? Even if they're more efficient at traffic flow, there's only so much physical capacity on the roads that people want to travel on.

> you're just going to have to accept that cars aren't going anywhere.

Cars aren't going anywhere, but they're the wrong solution for a variety of reasons (congestion leading to time loss, pollution (including noise pollution), land use efficiency or lack thereof, etc.) for a very big part of transportation needs.

> A lot of people want to travel comfortably in car

And a lot don't and aren't given that option.

> "mindless pursuit of space efficiency" be damned.

Mindless? There's only so much physical space available in the places where people live and work. American cities are often comprised of massive parking lots taking up valuable space that can be put to much better use, simply because most people are forced to commute by car. Give them another option and many (of course not all) would prefer it, which would also make driving much more pleasant for the remaining drivers.

> How would they eliminate congestion when they're so space inefficient? Even if they're more efficient at traffic flow, there's only so much physical capacity on the roads that people want to travel on.

Sure, but that's not the limiting factor at all, which you'd understand if you bothered to do the slightest bit of research outside of your comfort zone.

> And a lot don't and aren't given that option.

That's very true, but it's not one-or-the-other. I am a strong supporter for intensive funding of mass transit systems....for those who want to use them. But that doesn't preclude continued usage of cars.

> massive parking lots taking up valuable space that can be put to much better use

"better" is in the eye of beholder. Car users are generally okay with how space is currently allocated.

If you want change towards increased mass transit, I highly recommend not being hostile about it. There is no need for mass transit to be positioned as a replacement for car use. When you threaten peoples' way of life, that's how you get people actively working against you.

It should come as no surprise that I'm a huge fan of getting around by car, but I would still happily have my tax money go towards bolstering mass transit systems and walkable/cyclable infrastructure for the portion of the population that wants to commute that way.

Would this not lead to induced demand where more people decide to start traveling more frequently once they perceive the roads as having cleared up? An additional consideration would be people incapable of driving would now have greater access to freely take rides.

The notion of shutting down all improvements due to fear of induced demand is ridiculous. Let's also stop building sewage systems because of induced demand and all defecate on the ground. It will reduce water use and be good for the environment by enriching the soil.

That’s not what I meant to imply. I was simply asking if removing humans from the driver’s seat and replacing them with autonomous vehicles that are hypothetically more convenient to take would actually increase the effective road capacity.

I’m not so sure it would actually solve the problem. I think a more effective and easier solution would be to design cities to be more accessible without cars. That way, traffic would likely be reduced due to fewer humans in the driver’s seats even if they continued commuting. To summarize, in order to achieve a greater road capacity, I think we should focus on replacing the method of commuting that requires taking the road altogether. Subbing it for another method that still uses the road will invoke induced demand.

People in America mostly view cars as a status symbol and a part of their identity, which is why everyone buys giant trucks and SUVs. Why would a better taxi change that? Only "the poors" use taxis

This is not as true in cities, where almost no one has a giant truck and there are not so many SUVs.

I mean, Waymo is doing hundreds of regular trips throughout all of San Francisco. I think the technology is getting pretty good.

That's not actually a blocking problem, as you have the same issues with human drivers today. At the end of the day, just like literally everything else in life, self-driving tech is a statistical process with corresponding costs.

The bar for success here isn't perfection. Rather, it's whether or not self-driving vehicles are cheaper than human-driven vehicles. Part of the cost of a self-driving vehicle hinges on how often the vehicles end up in failure states necessitating remote human intervention and/or insurance payouts. It's fine if they sometimes fail, they just have to do so infrequently enough to be cheaper. And if Waymo is to be trusted, they already appear to be safer.

> the problem is that there's no sure shot way of knowing when they'll work and when they won't

Partnering with Uber let's them offer self driving only within pretty well known constraints. Weather, locality, traffic, routes, time, etc.

This is actually not true at all. At Waymo we have a very good understanding of our behavior.

Reply on news.ycombinator.com