-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Chuck Cook and FSD featured in New York Times
- Thread starter clydeiii
- Start date
I subscribe to the Times, I think this link should work. I am allowed to "Gift" so many articles.
What Riding in a Self-Driving Tesla Tells Us About the Future of Autonomy
We took a ride with Tesla’s experimental self-driving system. It was by turns intriguing, impressive and scary.
www.nytimes.com
Daniel in SD
(supervised)
Did they say when he’s going to test the unprotected left again? I’ve got a beer riding on 90% success…
Yeah they have an entire section dedicated to Subie's and Dan's beer betting. It's fascinating.
Daniel in SD
(supervised)
“It is very easy to solve the first 90 percent of the problem, very hard to solve the last 10 percent,” Mr. Clare said, referring to the decades-long effort to create self-driving cars. “You need to be able to handle those edge cases gracefully.”
I trust the experts. I can almost taste the beer.
I trust the experts. I can almost taste the beer.
Olle
Active Member
Thanks for sharing!
I met Chuck at Teslacon. He struck me as very positive and enthusiastic about FSD. Somehow NYT managed to twist the ride into something negative. out of 6 hours of driving posted the clips where it looked scetchy. And quoted a so called expert that put another wet blanket on top. Don’t think it happened by chance. Most people who’ve ridden FSD with me lately were overwhelmingly amazed/impressed.
Can definitely smell some beef…
Daniel in SD
(supervised)
Yeah, I don't remember Musk saying this: "After releasing the new beta, Mr. Musk softened his claims about the immediate future of the technology. He now says that the technology will not be widely available until next year — and that regulators are unlikely to approve it for use without hands on the wheel."
This is what he said just last month: "I think we'll be pretty close to having enough 9s that you're going to have no one in the car by the end of this year. And certainly, without a question, whatsoever in my mind next year. I think we'll also have an update next year to be able to show to regulators that the car is safer much so than the average human."
KArnold
Active Member
Daniel in SD
(supervised)
Agree, but if they’re going to report on what Musk says they should be accurate. I don’t think he’s ever said that regulatory approval is unlikely.
During the Oct 2022 earnings call discussion he reportedly said this:
"The car will be able to take you from your home to your work, your friend’s house, the grocery store without you touching the wheel,” he reportedly said.
“It’s a separate matter as to will it have regulatory approval. It won’t have regulatory approval at that time,” he added.
Daniel in SD
(supervised)
“We -- as I said earlier, we're expecting to release the full self-driving software to anyone who orders the package by the end of this year. So, a separate matter as to will it have regulatory approval. It won't have regulatory approval at that time.”
This year. The NY Times says he was talking about next year. Fake news.
AlanSubie4Life
Efficiency Obsessed Member
srs5694
Active Member
The fact that the car drives well most of the time is impressive from a software engineering point of view, but from a safety point of view, it's absolutely imperative that it avoid the sort of stupid mistakes that were shown in the NY Times piece. Highlighting those flaws is wholly appropriate, since those are the areas that need work and that are preventing a wider deployment of self-driving technology. The NY Times piece also detailed how Tesla has addressed some specific problem areas, like the left turn across the wide divided road. I did not take the tone of the piece as being overly negative; I took it as explaining where the technology is now and why it's going to be years before we have true Level 5 autonomy. Namely, there are too many edge cases to be fully tested and fixed for a quicker deployment.
I have the FSD beta, and I rarely use it because it does stupid and dangerous things far too often. It is impressive what it can do; but if it drives fine for 5:59 and then crashes into a parked car because it can't tell the difference between a road and a parking lot, that's not what I'd call a successful 6-hour drive. Yes, that's 5:59 of impressiveness; but that last minute of crashing counts for a lot more.
An observation regarding the article. The author’s cite an Artificial Intelligence expert who believes an AGI won’t be viable in our lifetime or our children’s lifetime. However, the article does not mention that FSD is not an AGI, but a specialized AI. Although there is much debate on TMC about how soon we will see autonomous vehicles everywhere, I don’t think it will take another 30-50 years.
I saw no lies or anything false in the article. The video clips of its shortcomings weren’t faked.
The article does well in contrasting reality from the claims Elon has repeatedly made for years to the public, including “robotaxi service by December 2019”.
We all know…there will be no robotaxi service in 2022, or 2023, or…
For one example..
The article does well in contrasting reality from the claims Elon has repeatedly made for years to the public, including “robotaxi service by December 2019”.
We all know…there will be no robotaxi service in 2022, or 2023, or…
For one example..
Olle
Active Member
I totally agree with you, it does stupid stuff. NYT gives a snapshot in time of the stupid stuff, which I mean gives a negative twist.
Instead, the salient point is the rate of the improvement over the two years that FSD beta has been out. Based on that rate, extrapolate to where one believes it will become safer than human drivers.
Do you feel the same way? how long have you had the FSD beta?
srs5694
Active Member
My point is that even a small percentage (say, 1%) of stupid stuff far outweighs 99% good driving, because it's that 1% that will cause accidents. That's not a negative twist; it's the reality of successes vs. failures in the problem space of driving a car.
I've had the FSD beta for just over a year. It has improved in that time, but given the rate of improvement over that year, I expect it will be several more years before it has any real utility. Both a year ago and today, it's an impressive technology, judged by the standards of a computer program that operates in the real world; but it does stupid things far too often and is not a useful technology. I don't trust it at all on busy city streets; I use it only in low-traffic conditions on roads that aren't even remotely tricky; and even then, it's rare for it to go more than about five miles without my having to disengage it. Claiming that FSD is both impressive and useless may seem contradictory, but the evaluations are coming from different directions -- it's impressive in an academic/theoretical sense, which is very different from the perspective of a user who just wants something that works.
"Safer than human drivers" is an obvious but peculiar metric. Tesla's FSD might meet that metric for most drives in (at a wild guess) 3-5 years, given the improvements I've seen; but there will be corner cases where a human will do better than FSD for far longer than that. If you unexpectedly hit one of these corner cases once every 100 miles, then how useful is FSD, really? You certainly can't turn your Tesla into a robotaxi, which might be expected to drive well over 100 miles a day. Even if the car shakily makes it through such corner cases 90% of the time without getting into an accident, those statistics would mean that individual robotaxis would be crashing once every few days. As a driver assistance feature (say, level 3 automation), FSD might be useful in 3-5 years; but I'd be concerned that human drivers would over-estimate the car's capabilities, be inattentive as a result, and therefore end up in accidents that would not otherwise occur.
My expectation is that Tesla will have to upgrade cameras, upgrade the FSD computer, and/or add more sensors (lidar, say) before they can get FSD to work at level 4 or above. I can't say I'm certain of this, but given the slow progress on the problem (at least, compared to Tesla's promises), it's looking something like Zeno's Paradox, and hardware improvements may be necessary to get past that.
I'll add that my view of Tesla's FSD has gone down in the 3.5 years since I bought my Tesla, and in the year since I gained access to the FSD beta. I bought the feature for $5,000 in early 2019 with the knowledge that Musk and Tesla had been making claims and predictions that were unlikely to be met, certainly on the timelines being mentioned at the time. Nonetheless, I thought that I'd get something useful out of my $5,000, and maybe I'd get something worth more than that amount. Today, although the FSD feature of my car does more than it did in 2019, I'm more skeptical that I'll get my money's worth out of the FSD purchase. As it is right now, FSD requires more effort to use than normal driving; it's the Flaky Student Driver feature, as it requires the sort of attention that's required to monitor a 16-year-old who's never before driven. (Or so I imagine; I admit I've never taught anybody to drive.) The timeline for improvements has not only been slower than what Musk and Tesla were promising in 2019, it's been worse than I imagined it would likely be at that time. If I were to buy a new Tesla tomorrow, I would not throw away $15,000 on FSD, and I can't recommend that anybody else do so, either.
The only question is, do you think you'll ever stop bringing this fact up? I mean, we get it.
Similar threads
