“Tesla Is Reportedly Revoking Internship Offers to College Students Weeks Before Their Start Dates: ‘I Spent Thousands On Housing’”
“Tesla Is Reportedly Revoking Internship Offers to College Students Weeks Before Their Start Dates: ‘I Spent Thousands On Housing’”
I think you’re confusing what some Tesla fans breaking the law will do and post online and what an automotive product is officially certified as being able to do: some people showing that it can do it in certain conditions is not the same as it being demonstrated by experts (using a stricter standard than the “I’ve been able to make it works” of Youtube videos) as being able to reliably and safelly do it in all required situations, and hence being certified.
If all it took to formally define it as capable of doing it is “somebody posted a video of it doing it”, then a car falling down from a cliff would be formally the same as an airplane. Certainly if throwing clapped up Ford Fiestas of cliffs and filming it for Youtube became a fad, by that logic of yours they would be as capable as Cessnas since you would’ve seen lots of videos in Youtube of Ford Fiestas flying.
Lot’s of videos in Youtube is not significantly Statistically correlated with capability, it’s correlated with how long those cars have been out (10 years), number of them out there and the kind of user it is aimed at (tech early adopters who tend towards fanboyism and posting “look at what my cool gadget can do” videos, unlike the high-income established professionals targetted by Mercedes). Further, in Social Media people are far more likelly to show the selected “good” videos of things they feel reflect on their personal image rather than the “bad” ones - it’s the same as with vacation photos in Facebook: they’re almost never those photos that make a person look like an idiot that blew a ton of money of shit vacations.
All that said, there is a simple disproval of your point: if the Tesla software could reliably and safelly do all that in all required situations, then it would’ve officially been certified as LVL3 and it’s not.
Certifications are how the actual experts assure they’ve officially tested it and it checks out. Non-expert opinions of it or how close they are to it, on the other hand, are just a reflection of wishful thinking and ignorance (ignorance of for example the 90/10 problem in software development).
I don’t see anyone breaking the law in those videos. There’s a driver sitting behind the wheel ready to take over immediately if needed. They’re using the system as it’s intented to be used and the car drives itself as if it was driven by a human and when it fails it’s usually either by getting stuck somewhere or trying to drive somewhere it’s no allowed to like construction zones. There’s a dude on YouTube doing ridesharing with Tesla using FSD and he’s been keeping track of it’s performance and V12 completes 90% of the trips from start to finish with zero human interference. Again - something Drive Pilot physically isn’t capable of doing. It just can’t. That’s not an opinion, it’s a fact.
It doesn’t take a genious to figure out why it’s easier to be confident enough on your self-driving system that you’re willing to certify it as level 3 when the total lenght of road network it can be used on is only like 15k miles long consisting of only hand-picked highways when compared to one that can drive on literally any road in the entire country of United States.
The 10/90 rule is about how the last 10% of the way to a final product are 90% of the development time.
To people outside it might look like “it’s nearly there” whilst the domain experts know that’s not how software development works.
Your entire point from the start is that they can become a software company and sell they FSD system.
My point as an actual very senior software development domain expert with an actual Engineering degree is that whilst you can get to the “works 90% of the time” with hacks, you can’t get to the 99.999% (or whatever the number of nines required for certification) with hacks and judging from all I’ve read and heard about the Tesla software from usability and user interaction point of view (down to how the process for deploying it to user vehicle is sub-standard compared to just about any Tech-related industry but Startups) they certainly don’t come out as following the necessary engineering standards or even having the proper development process to have implemented their FSD properly and with a clear route to the finishing line, but rather that theirs is the same kind of hacky “we’ll fix it when it breaks” way of work as shitty-shit early stage startups.
How long they’ve been at the “nearly there” stage just reinforces my impression: I’ve seen plenty of badly done complex projects get stuck exactly like that, including over the years some of mine.
To me it all adds up to their FSD having a vibe of “stuck down a development dead-end”, though that’s just my gut feeling based on professional experience.
Your entire original point is indirectly predicated on a level of engineering professionalism at Tesla that everything visible from the outside indicates is severely lacking and supported by the kind “works most of the time” evidence that does not show they have the technical capability or even the software foundations to bridge the missing “10% of the way but 90% of the work” chasm.
There really is no amount of “nearly there” examples that can show Tesla has the Software Development and Engineering chops to reach LVL3 whilst Mercedes clearly has.
Then again given the conditions for LVL3 with it, it might turn out that the Mercedes one is also a software development dead-end when it comes to making a general use LVL3 FSD, just a different one.
It’s not at all uncommon for there to be several “nearly there but can’t get there” dead ends in the development of “never done before” tech before somebody has a break-through.
It’s just that from all I’ve seen coming out from Tesla, to genuinelly expect that they’ll be the ones with a break-through is very very optimistic: people who can’t even take in account the usage requirements whilst on the road of things like direction indicator toggles aren’t exactly likely to pivot to become successful providers of a complete certified engineering solution for the rest of an engineering heavy industry - even different teams are still hired by the same leadership with the same “value criteria” when it comes to selecting developers and building software development teams so the likelihood that the FSD team are pros whilst the rest are not is very low.
Of current available self-driving systems FSD is by far the most advanced I’ve seen. Just let me see a video of a vehicle doing what Tesla can do but do it better and I’m ready to change my mind.
Luckily this is the type of argument where we only need to wait and see.