You misspelled moronic.
Operation Hardcore on full display @ Tesla over the past day:
I sincerely hope that Apple and Google remove Twitter from the app store, and Musk builds a new phone. The sooner this guy is (relatively) broke and no longer in control of so much wealth the better.
Sounds like not enough code reviews going on at Tesla! 
Maybe the Tesla software engineers who should have been conducting design walkthoughs and building testing protocols at Tesla were too busy grading lines of code at Twitter, now re-invented as a junior technical college (“make big bucks computer programming at Twitter”, as seen on the back of a matchbook).
I was watching yet another of those fascinating FSD videos, and the driver said something interesting, which was that the most recent update has fixed some problems he knew about, but then had “uncorrected“ some other ones that had been previously fixed. Which I suspect is what happens when you pile layers and layers of fixtures onto existing code. Code reviews indeed.
Here’s another minidoc putting the lie to pretty much every claim that Musk has made about himself. He really has no skill in anything except self promotion and pushing out actual innovators after they’ve done the heavy lifting.
It’s what happens in a “shoot-from-the-hip” culture that doesn’t understand regression testing, nor, apparently, the priorities involved in engineering reliable software.
Pushing the other eggs out of the nest is standard behavior for a cuckoo. I wonder if there’s a connection here.
Given that a lot of the “code” that these are operating under isn’t actually human readable and is largely created by machine learning, there isn’t much to “review”.
Although it is tempting to ascribe FSD issues with a lack of development rigor in Tesla software V&V processes, the truth is even more fundamentally problematic. Tesla (and most if not all autonomous driving systems in development) use machine learning algorithms (so called ‘deep learning’) to develop the complex set of rules and interactions that a truly autonomous system must have to operate on a public road with other drivers and changing obstacles. This is really the only viable approach because a discrete “database of rules” approach would quickly become unwieldy and impossible to apply to the highly dynamic environment of a public road. However, it means that engineers cannot simply “review the code” or indeed, have any deep insight into what the algorithms are doing other than actually observing the reaction of the system to realistic obstacles and hazards.
Some (boring, stodgy) companies are incrementally developing such learning algorithms under configuration control, and then testing them on closed tracks against progressively more complex scenarios to evaluate the maturity and reliability of such systems. The Tesla approach?:
One aspect of Tesla’s machine learning program that sets it apart from its competitors is the fact that the entire Tesla fleet operates as one wirelessly connected network. Therefore, when one vehicle learns something new from additional data inputs, every Tesla vehicle can instantly share in its improvements. In this sense, Tesla vehicles are similar to smart connected gadgets such as the thermostat Nest.
I’m sure that sounds great to someone but what it really means is that some kind of error that can make the pilot act in an unstable manner can rapidly propagate through the system, as could a deliberate malware attack. If your Nest thermostat is suddenly hijacked, you might wake up in an overly warm house. If your Tesla FSD is unreliable, you might find yourself hurtling into a truck or barricade at 100 km/hr. I might take the risk on the thermostat but I’d want to be damned sure that any autonomous piloting system is not only thoroughly demonstrated to be at least as reliable as a human driver in al conditions but is also well secured and not prone to suddenly turning my trip into a shitty Will Smith movie.
Stranger
Ok, so I was making a pun on a Will Smith movie and then… holy shit, so many of his crappy film titles just work by themselves. Suicide Squad anyone? Hitch? I, Robot?
When your machine learning algorithms have thousands of outputs based on thousands of different inputs, it is really easy for machine learning in one area to turn into machine unlearning in a different area.
If this text were literally true, why would FSD be released in staged updates?
As recently as four years ago, really my last involvement with any of this, yes they were using machine learning, but they were still very carefully deciding what inputs to reflect, and at least claiming that they had controls for each update, rather than any kind of continuous and universal learning happening. The only kind of machine learning that I recall actually happening on the fly back then, on the vehicle, with some stuff they were doing with depth perception for the cameras.
I have real trouble, and I’ve always had real trouble, sifting between their marketing bullshit, and what they actually do.
I’m open to any kind of education on this, I just wish I had a clearer picture of how the whole thing works.
Exactly, and granted that FSD is an enormously complex problem, but regardless of how the rules are acquired, the system can and should be subject to rigorous testing protocols – including thorough regression testing – for every release. This takes patience, discipline, and relevant technical expertise, none of which Elmo is known for.
Elon’s fans/opportunistic investors have built him a tribute. It is the silliest thing I have ever seen:
“You shall not make gods of silver to be with me, nor shall you make for yourselves gods of gold.” — Exodus 20:23
I mean, I’m not a religious man but it seems like this passage was written specifically in prophecy of this moment in time.
Stranger
I mean, first thing I thought of as well. Probably many others did, too. Where’s the head slap emoji when I need it? Reality truly is stranger than fiction.
That…thing is supposed to be flattering? I mean, if I had unlimited funds, this is precisely how I might dunk on the guy. And how come there’s no video of the goat fire-farting?
Apparently it cost something like $850,000 to make. My mind has boggled.
Stranger