12 ms·
Tesla pulled its latest ‘Full Self Driving’ beta after testers complained
- verelo 5y agoWhen do they start refunding people for this “feature” so many paid for but are not getting?
- yellow_lead 5y agoMaybe once there's a class action for it
- verelo 5y agoYeah, in all honesty i hope they figure it out. We have fsd on our model 3, and i am really looking forward to it, but I’d be lying if i didn’t feel a bit duped…especially since i found out it could be bought on subscription after we already paid for it.
- yellow_lead 5y agoSorry to hear that. I do think they screwed a lot of people over with the marketing - I'm sure they will get their punishment at some point...
- throwawayboise 5y agoYou bought a car from PT Barnum.
- verelo 5y agoHa, yeah i guess that might be.
- warning26 5y agoIt does seem ridiculous that they're still selling "full-self driving" capability. How have they avoided legal trouble for false advertising?
- justapassenger 5y agoTesla and Elon are tech darlings, and nothing sticks to them. For now.
- FutureZeitgeist 5y agoI’ve never understood this either. I guess it’s technically under a “beta” label so it gets a pass for now. But I can’t imagine this continuing without consequence forever, unless Tesla actually manages to solve self-driving with a vision-only approach.
- akkawwakka 5y agoProsecuting FSD claims, historical or present, seems like a slam-dunk case for the FTC…
- jason2323 5y agoIts a difficult problem! And i hope that they are able to figure out their issues and push forward the era of FSD
- iab 5y ago> It is impossible to test all hardware configs in all conditions with internal QA, hence public beta. 10k to be a beta tester
- shmoe 5y agoDepends when you bought.. 6k here :P
- Guest42 5y agoHopefully it's not 6k + your personal safety.
- MarcoZavala 5y ago
- maxharris 5y agoCorrection: $10k to be a beta tester on a technology that's worth $100k or more per car when perfected. I'd also like to point out that you can get it through a subscription for $99/month if you already have Enhanced Autopilot ($199/month if not).
- akerl_ 5y agoI bought FSD, but I’m really confused by the idea that it would be worth $100k when perfected. What’s the math look like on that?
- maxharris 5y agoThis is from 2019, and they say a robotaxi could make an owner around $30k in a year. https://medium.com/swlh/30-000-from-a-tesla-robotaxi-not-as-crazy-as-it-sounds-dafba4c6923c https://medium.com/swlh/30-000-from-a-tesla-robotaxi-not-as-...
- 5y ago
- clouddrover 5y agoWhat will be the last straw for people to demand a refund for the "full self driving" that is still not delivered? Will it be when Tesla says they now need lidar after all?
- maxharris 5y agoJames Douma explains why lidar is a waste of time for FSD: https://www.youtube.com/watch?v=urKMwhzivs8 https://www.youtube.com/watch?v=urKMwhzivs8
- lancepioch 5y agoHis argument makes sense and I think he's completely right. However, his argument also includes, "If the depth prediction is arbitrarily close to lidar". So we have to ask ourselves a bunch of questions related to it. Can we get it close enough? How feasible, reasonable, and cheap is that? How can we tell if it's good enough? And I'm sure there's more.
- kjksf 5y agoThere's a big disconnect between people who do not own FSD and those who do. Those that complain are 100% in the "don't own FSD" bucket. Prove me wrong. Find me one owner of FSD that is so dis-satisfied that he demanded a refund. Those who do pay for FSD use terms like "my mind is blown", "it's amazing". They do videos, evangelize the product and wait until midnight on Friday (that's when Tesla releases updates) so that they can test it at 2 am in the morning. You can make what you will out of this. To me it shows that product is great already and will be very successful when finished. The complainers that never used the product are noise.
- 01100011 5y ago> Those who do pay for FSD use terms like "my mind is blown", "it's amazing" Absolutely. Then you tell them you heard the car does odd things like randomly coming to a full stop and they say "well, yeah it does and it sucks, but it will totally be perfect one day." It's funny how people can change their attitudes towards something when they've sunk $10,000 into it.
- Syonyk 5y agoOf course they did. There's only so much you can do with "feed more data in the machine learning pot, stir it, and see what comes out." We're doing a good job of proving what the limits of ML based self driving are... Computers aren't human, and as it turns out, human brains and visual processing systems are more than the "couple crappy cameras and a bit of a neural network" thing that a lot of SV types seem to believe it is. We're firmly into the "fusion" realm of self driving cars. A lot of money and a lot of time, effort, developers, etc, have been fed into the maw, and we've tried all the easy options - which it's clear don't work. I'll take the pessimal position of saying that the current approach Tesla is using (vision only) can't work, though I'm sure many will disagree with me. It comes down to a difference of opinion about the complexity and capabilities of the human brain. A sensor-heavy setup like Waymo is using has the greatest chance of working, though probably only in heavily-mapped areas where it can use the sort of peripheral data humans use to identify the lanes by building locations and such when the lanes are unclear. As long as not much changes, which probably requires very regular remapping of regions (or an awful lot of data upload from cars), that's the most feasible approach, though it's not a general purpose solution given the amount of data required to do high precision mapping of the entire country, much less the world. We'll see, but "a couple cameras and a car-scale computer" just doesn't seem a solution that's going to work.
- tablespoon 5y ago> Computers aren't human, and as it turns out, human brains and visual processing systems are more than the "couple crappy cameras and a bit of a neural network" thing that a lot of SV types seem to believe it is. Honestly, I think it's something approaching a general rule that most things are far harder than "SV types" believe they are. Maybe it has something to do with all the startups.
- coz131 5y agoI always find it odd that hacker news has such pessimism on technology advancements. It might not work now but give it another 5 to 10 years this may very well become reality. The pace of advancement in ML/AI is incredible.
- 5y ago
- sidibe 5y agoHaving such bad regression that it seems like all users noticed with new update shouldn't be possible for this kind of technology. Whatever your level of faith in Tesla, and some of you have a lot, you have to admit that it is reckless to have a release process that fails to test this internally enough that they would widely release dangerous behaviour that tons of users noticed immediately
- MuffinFlavored 5y ago> bad regression I wonder what their testing suite looks like...
- other_herbert 5y agoWe just found out
- m0zg 5y agoAlso it's interesting that they now canary it on drivers with "high" safety rating. I wonder if that introduces bias into the data they collect and "bad" drivers and driving environments become "out of domain", in which case the system will suck with those drivers and in those environments. Glad I didn't pay $10K for this though. Super satisfied with the car otherwise.
- brokenmachine 5y agoInteresting. How do they judge a drivers safety rating? Can a driver view their own safety rating?
- m0zg 5y agoI'm not sure how they do it exactly, but I imagine you could, as a first approximation, infer something like that from how hard the driver brakes and how reliably they keep distance from other cars, both of which Tesla's telemetry can and does collect. Tesla will beep at you if it thinks you're forgetting to brake. It also knows if you ran a red light or a stop sign, since it recognizes traffic lights and signs. It knows if you're speeding, too. And it knows if you're speeding near a school or in a construction area, as well. AI is a bit of a double edged sword, since if Tesla knows this, it's only a matter of time before this data makes its way into some sort of a "social score" metric, that will affect things like car and health insurance.
- brokenmachine 5y agoInternet of Snitches https://news.ycombinator.com/item?id=28158939 https://news.ycombinator.com/item?id=28158939
- sjg007 5y agoSo how do these systems rank? 1. Tesla 2. Super cruise 3. Ford? BMW? 4. Anyone else?
- pkulak 5y agoCar journalists seem enamored with Super Cruise, though I've never tried it. I compared a few, including Autopilot, before I bought my last car and actually liked Volvo's the best. It doesn't try to be more than it is. It's just a very strong bias of the wheel for the middle of the lane. You can change your lane positioning by turning the wheel, and you'll get pulled back to the middle if you let off. But it won't disengage unless you really torque it. I wish is was less strong, actually, but then you wouldn't be able to rest you hand on the wheel for the presence detection.
- powerbroker 5y agoI'm not buying FSD. The non-FSD features in the car are way too unpolished. If Tesla can't get the easy stuff right, there is no hope for the more complex. Some issues I've noted: 1. Voice recognition fails to match the stated operation described in the manuals since 2019; 2. Cameras are unable to reliably distinguish traffic signals, namely red; yellow; and green, using 2018 camera hardware; and 3. It makes absolutely no sense for the battery to 'brick' or declare 'out-of-range' while regeneration is adding charge back to the battery -- which is how the Model S behaved in 2020. In other words, as the car nears '0 range' or low/no charge, it absolutely should not shut down while kinetic energy is being converted to charge as the car slows down. Based on may 5 years of experience driving two Teslas, I can't imagine this 'FSD problem' being solved prior to 2026.
- perl4ever 5y agoI don't think it's a Tesla problem, ultimately. I look at it this way. Every day I plug in my Android phone to my car and use the navigation. I think it's the best option out there. However, it's far from perfect. For instance, it frequently wants me to turn left on to a busy street from a side street with no traffic light. There are many other issues which aren't that big a deal because I'm in control. Sometimes it gets quite weird. One time my location was displaced like a block and a half, scrambling the directions. Even while acknowledging I'm ignorant of 99.9% of the research and what's being done, there is a simple equation that I believe in - a nigh-on perfect navigation system must precede self-driving. If nobody knows how to walk, running isn't going to happen. I don't have to know what people are doing in private to know that what they are doing in public isn't consistent with the aspirations.
- penjelly 5y agoyikes after watching the one of the videos in the thread this actually looks dangerous to be using on roads. Im not against public testing but if someone gets killed wont that be a huge setback? The video in this thread shows like 4 failures in 2 minutes, almost driving directly into a pole, almost removing the driver door of a parked vehicle, turning onto the wrong road, and missing an intersection turn. A month ago i saw similar videos of FSD turning towards pedestrians... im actually surprised things havent been worse
- penjelly 5y agocorrection: they do speed up some portions of the video. so "2 minutes" is inaccurate.
- yellow_lead 5y agoSo Tesla HQ issues a "revert" and everyone's car rolls back? Is it not optional?
- brokenmachine 5y agoLol, as if anything's optional.
- _009 5y agoAI doesn't scale well. Problems get worst as you make your model bigger and more generalized. To make things worst, data, model architecture, precision, hardware, affect your model performance in ways that are hard or impossible to anticipate. If you watch Tesla's AI presentation, https://www.youtube.com/watch?v=HUP6Z5voiS8 https://www.youtube.com/watch?v=HUP6Z5voiS8, you will notice that they have multiple AI's stacked on each other, which IMO is a step back from truly e2e multimodal AI system. So even with their custom fancy hardware, multimodal is too hard. I wonder, wouldn't it be better to use geo fencing (using H3), and have the car download the model depending on the zone where it is driving? And optimize multiple models based on "driver engagements"? This could fix the problem of zones where there are particularities in the driving, road, or human activities, and allow for model optimization to happen on a smaller vector space than the whole world. For example, why not have a model for US highways, LA, New Deli, UK, so on. Tesla also knows where the cars are, and control their expansion plans worldwide, which could inform model prioritization roadmap. In my mind, it will be easier to test, debug, label, optimize, and guarantee quality to users, that at the end of the day, without knowing exact statistics, I am dare to say spend more than 70% of the time driving around the same county/city/area/town?
- intended 5y agoIs it just me or do people tend to ignore that traffic data is cultural data, not some observation of the periodicity of a pendulum. Self driving level 5 will not work and cannot work.