141–150 of 219 posts
Earlier quoted context omitted.
This. So much this. It’s 2022. We still have meat flavoured train drivers and they have a massively constrained set of complexity to navigate.
It helps when those meat bags aren't on their phones just like in cars.
Earlier quoted context omitted.
Which law of nature self driving AI contradicts?
We know enough about the physical laws of nature that we can write them down and figure out where the limits of tech/automation are. We know nothing about how the human brain works. True artificial intelligence/AGI (which is needed for level 5 self driving) is a pipe dream.
Sooner or later AGI will be available, it's just a matter of time. It does not have to work like a human brain
https://www.smbc-comics.com/comic/fsd is the current definitive take on the subject. Watching this video I really had no idea how bad it was. My opinion is that we will never have full self driving without infrastructure updates; basically putting the cars on rails and forbidding non-self-driving cars and pedestrians. Even then it's dicey. At least part of the reason is that a self-driving car can never have skin in…
Earlier quoted context omitted.
If it's safer on balance than average drivers (for some reasonable definition of "safer"), I think the answer is "yes". Do you want to expose your children to unnecessary risk? I don't posit that it is safer or that it should be legal, but we should frame the conversation in terms of actual safety metrics and not cherry-picked incidents. Of course, our media isn't equipped to facilitate these kinds of nuanced convers…
But HOW do we know is safer? We should not assume is safer by default and I see many problems with Tesla fanboys numbers 1 the count for the times drivers intervene to prevent a crash are secret, or are not counted at a crash. 2 I don't want on my village/city some american car that is safer then the american driver that included 16 year old drunk teens and people that don't even have to pass exams and medical tests.…
We measure. This requires transparency on the part of Tesla, but if we frame it as "if you want legal FSD, you have to comply with transparency". We regulate like this in all other industries so there's no reason to think we can't do it here.
> We should not assume is safer by default
No one is advocating "assume it is safer by default".
> I see many problems with Tesla fanboys numbers
Right, but by framing it with metrics rather than cherry-picked incidents, we can have a national conversation about which metrics are most appropriate. If Tesla fanboys bring numbers that you take issue with, you can debate them. As an aside, it's hilarious that my earlier comment was being downvoted for advocating a debate based on data rather than cherry picked anecdotes, but I should know my audience better :)
> the count for the times drivers intervene to prevent a crash are secret, or are not counted at a crash.
See my transparency remarks above.
> I don't want on my village/city some american car that is safer then the american driver that included 16 year old drunk teens and people that don't even have to pass exams and medical tests.
Ignoring the obvious barbs at Americans, if we quantify the safety of any given FSD system, we can compare that against local safety and prohibit FSD systems which score worse than the local average such that FSD will only ever be a boon.
> updating this cars should invalidate previous approvals, as a software developer I know well how updates include bugs not only features.
Seems reasonable to me.
> Anyway I am against companies deciding from themselves how is a capable alphya testers and how they should run alpha tests.
In this case, I agree.
> Fuck shit statistics and shit companies and fanboys using them.
I love these kinds of constructive comments. Keep 'em coming.
Earlier quoted context omitted.
Cold fusion is a hoax. Did you mean regular fusion? Both Cruise and Waymo are legally operating cars without drivers that are available to the public today.
Whoosh (first part). I think that was their point. In one sense the cars are not available to the public. They are available as services of course, so you’re right in that sense. But we do not know what is going on in the background (such as a team of “safety drivers” monitoring every inch of the drive with the ability to do who knows what at any time). As far as I can tell, Cruise has an appearance of autonomy to re…
Cruise launched publicly in San Francisco yesterday.
I believe both they and Waymo are using remote safety drivers but at a ratio of one for every 20 cars on the road and they expect that ratio to continually decrease over time.
Abridged history of Elon's self-driving predictions: 2015: "We’re going to end up with complete autonomy, and I think we will have complete autonomy in approximately two years." 2016: "I really consider autonomous driving a solved problem, I think we are less than two years away from complete autonomy, safer than humans, but regulations should take at least another year." 2018: "I think probably by end of next year (…
FSD will never work for ordinary cars in cities. But it may be good enough for self driving tanks fighting in WW4 - AD2050 That is really the only scenario where the high collateral damage won't matter. "We didn't know we where training kill-bots"
Earlier quoted context omitted.
Which law of nature self driving AI contradicts?
Probably the one that attempting to process an unconstrained system in a constrained model, regardless of the method, always results in unhandled edge cases. Apply tons of metal and fragile entities made of meat to the problem and someone’s going to get hurt.
In finance they say past performance does not indicate future success, same should apply to AI in both directions. Just because its applications were too narrow in the past predicts nothing about the future.
Earlier quoted context omitted.
We know enough about the physical laws of nature that we can write them down and figure out where the limits of tech/automation are. We know nothing about how the human brain works. True artificial intelligence/AGI (which is needed for level 5 self driving) is a pipe dream.
I'm sorry, but it's you who know nothing. Insulting entire fields is neither nice nor constructive. Sooner or later AGI will be available, it's just a matter of time. It does not have to work like a human brain