The bit that caught my attention is that 6 of the 8 founding scientists are gone. I bet the former and current heads of AutoPilot are shaking their heads now. Sterling Anderson -> Chris Lattner -> Stuart Bowers -> Milan Kovac. Note how there was a progression from experienced highly respected scientists and engineers until Elon finally put in place someone who owes his career to Elon.
In the real world, the people who get stuff done get replaced with Yes Men who go "yes, no problem, we can hit that deadline" and end up delivering trash.
Tesla's self driving division will really have to include a beheading life insurance policy for me to buy their cars after the shit that came out on Twitter about how things are done in their "move fast, break things, kill people" organization.
Wow...I always thought software development in the automotive field was superior in terms of care, precision etc to that of an average software company. Now I can without doubt tell the stack and the codebase in the organization that I work is far more robust and carefully done.
I'm incredibly careful with all my projects when it comes to security or avoiding bugs in general. It's only been after a decade or so working in my field that I took on projects where I had to write authorization/authentication code, or where I had to harden and manage a server.
Every single time that I took on work as a contractor for "Serious" BigCo, I was shocked by how much worse they were handling all of it. Literally no exception. For some of the chaotic startups I worked for, I wasn't too surprised, but this was the norm everywhere.
For example, as a contractor I'd be able to take entire continents' high-traffic websites offline for a particular media company, probably costing them a noticeable amount of money. Not only could I do this, but I could still do so more than a year after I had left the company.
I suppose having the resources to open up a can of lawyers makes a big difference though. So I'm still very careful personally.
There have been a total of six deaths related to Autopilot misuse over 3 billion Autopilot driven miles by the fleet. Humans are terrible at risk management.
> There have been a total of six deaths related to Autopilot misuse over 3 billion Autopilot driven miles by the fleet. Humans are terrible at risk management.
Or, humans rightly assessed that despite marketing to the contrary autopilot only really works on the easiest part of driving (highway miles) and still managed to kill 6 people.
The things the ceo proclaims on stage and in the press are marketing -what it’s sold as.
A click wrap legal disclaimer after you already purchased the thing is in no way what something is sold as. It literally can’t be - because you have already bought it.
Based on Tesla’s valuation and their autonomy license revenue (publicly available in the financials), I don’t believe your assertion is borne out. Customers and investors are confident in their autonomy aspirations, based on the data available.
> These numbers are quite meaningless without the counterfactuals how many deaths were there by comparison?
Parent probably should have quoted the figures, but they're frequently cited (so somewhat assumed, I guess) in such discussions, and very easy to find on the net.
From [0] I see:
"1.13 deaths per 100 million miles traveled."
Naive scaling to 3 billion, that'd be ~34 deaths in non-Tesla Autopilot driving.
Parent is correct that humans are poor at risk management.
Another common refrain when talking about auto-driving technology is that it doesn't need to be perfect -- just better than humans. And that's a much lower bar.
That stuff, about 6% of deaths occur on the freeway versus 20% of miles are on the freeway. The majority of autopilot is gonna be in freeway or similar roads so let's call that 10 deaths for every Tesla death.
Anecdotally tesla put a lot of effort into making a car with great crash protection, so if we are evaluating only the autopilot then comparing tesla's death rate to a 1990s fiat or a car people buy to drive badly like an impreza doesn't seem right.
With that said, leading causes of freeway accidents are, falling asleep, bad lane following, and drivers failing to realise the traffic ahead has stopped. These are all about human inability to stay focused. So we shouldn't assume that in a simplified environment like a freeway, autopilot is going to lose up humans.
> That stuff, about 6% of deaths occur on the freeway versus 20% of miles are on the freeway. The majority of autopilot is gonna be in freeway or similar roads so let's call that 10 deaths for every Tesla death.
Raw numbers are 34 / 3b in non-Autopilot, and 6 / 3b in Autopilot - so that's 5.7:1 ratio.
I'm not 100% clear on how you're arriving at 10:1 ?
> Anecdotally tesla put a lot of effort into making a car with great crash protection, so if we are evaluating only the autopilot then comparing tesla's death rate to a 1990s fiat or a car people buy to drive badly like an impreza doesn't seem right.
I guess we'd have to review numbers of 1990's fiats, and indeed every other car on the road, the crash safety profile, and compare -- but this seems onerous.
The fact that the Tesla vehicles are sold as a whole package - autopilot engineered around capabilities of the car, including withstanding crash scenarios - could in good faith be compared with every other car (without analysing them all in detail) using vanilla statistics on crash events, yes?
It's stupid to use a benchmark on a fancy premium vehicle versus the entire market.
Joe Schmuck who didn't replace his tires and was driving home after a 12 hour shift when his brakes failed and he died counts as much as a guy in a luxury vehicle that drives under a semi.
Thanks - this kind of friendly and helpful advice is the only way I'll learn.
> Joe Schmuck who didn't replace his tires and was driving home after a 12 hour shift when his brakes failed and he died counts as much as a guy in a luxury vehicle that drives under a semi.
I can't work out what point you're trying to make here, other than perhaps implying that the statistics I cited, from the source I cited, don't accurately represent your opinion about some subset of the driving population.
GP wanted, but could not look for, comparative figures. These were provided. If you want to argue the legitimacy of them, let's do that - show me (other) numbers.
The actuarial data (lifetime income potential, etc) that compensation claims are based off of that determine the worth of human life says no.
Joe Schmuck should get the same safety features. To do so, you must drive down per unit costs through scale. I bought our Teslas because expensive cars need to get sold to pay for cheaper cars.
10k miles on Autopilot over here. Made me go from hating driving to loving it. I’ve been seeing just as much hate for Neuralink as Tesla had early on, which has me excited. Let the public put pressure on them, then when they come out with something even a little impressive, it’ll blow everyone’s minds.
Unfortunately with the way things are, outside of public research, the only way to fund a project of this nature is with large amounts of hype.
Seems like great thing. The fact is that science may be moving too slow and given proper motivations could move faster. If it can not move any faster then you quickly figure that out that you've gotten you answer, and you can circle back in a few year. But if so you've effectively moved the needle past what people thought was possible and you have a moat the size of the pacific ocean.
Seems like a better testing suite is in order. How many times since then has that happened, and have we gotten better at figuring out issues before they become this big of an issue.
I ask this because we've had everything from chernobyl to the Columbia rocket explosion. We didn't hamstring those things. We got better at anticipating them.
Remember, it doesn't matter if you've survived 100 Chernobyls, 50 Tsunamis or 20 Covids, all it takes is one bad one to wipe you out of the gene pool(kill). The worse is called worse precisely because it did not happen until then. Yes, testing helps but only when you know what to test, a hastily designed system where the workers are put under pressure and made to ssh into a fleet(See: https://twitter.com/atomicthumbs/status/1032939617404645376) to fix bugs, is not a mere matter of testing. It is a matter of culture, autonomy, prudence, care, and add other things that would help build a well-designed system from ground up.
The cry of every neuroscientist! The trick is how to make an ex vivo brain simalog at all.
Currently we use the brains of other animals. It gets you somewhere. I mean, you can deal with how the heart moves the brain about, how grooming and eating move the brain, you can even get seizures and ischemic events working with the tech. But, those last few percent is where all of it matters. Kinda like viewing an eclipse: the difference between 99.99% and 100% totality is everything.
And that's kinda the problem. Figuring out all the edge cases is really hard and time consuming. And making a faux-brain simalog is probably harder than just going slowly.
The issue isn't size of the testing suite, it's covering all the edge cases.
Some rather common edge cases go ignored. Unless they've updated Waymo in the last couple of months, they still fail to distinguish when they have a reserved destination lane when making right turns, as exhibited when going from eastbound Central to southbound Castro every damn day (before Castro was closed).
Self driving system vendors can claim that fewer deaths will occur with their system vs the average driver, but that mostly reflects how dangerous the "average"/distracted driver is. Those numbers are little comfort to the loved ones of someone killed in an "edge case" that a competent driver handles without injury.
Not just startups, any corporation too, or retail, hospital, etc. People have another thing coming if they think this is not standard life at most workplaces.
There will always be a tit for tat between labor and employers. It’s a perpetual conflicted relationship that needs constant maintenance and compromise to achieve sane productivity.
Well, the cars look great, the rockets look great. Now we wait and see if it works for the mind-machine interface. I don't think HN is capable of generating good predictions on this.
Maybe in 50 years workers will be forced to impant this thing. Imagine your boss screaming at you literally in your brain because neuralink is telling him you feel lazy!
Nah. I don't buy it. I don't think that future will ever come to pass in any democratic country, even a sham democracy like the US. The evil forces haven't even managed to get an encryption-backdoor law passed, and most policymakers don't even understand that issue.
What you're describing is so far beyond any social issue of today in terms of egregious and horrifying threats to society that it would just get voted out of existence so hard you'd barely know it was ever a risk.
Nobody will ever let thought intrusion come to pass.
On the flip side, everyone is now walking around with an internet connected camera, microphone, and location tracker in their pocket. Most of us are livestreaming their location to cloud services.
I don't think this will come to pass as a mandate from on high, but it might come to pass as a practical device that everyone needs to function in modern society.
Most tech companies are in-line with some form of abuse. Whether it's creating campaigns against reasonable salary increases to installing highly intuitive applications to track and monitor your activity as an employee.
More simply put, things we'd give people felonies for tech companies (and others) get a pass.
I could believe this dystopian future, especially as political forces continue to form stern camps whose only substantive difference is an opposing view to the other.
> Nobody will ever let thought intrusion come to pass.
Human rights will take steps backwards if it's necessary to keep an economic advantage.
If thought intrusion has any productivity value, a less scrupulous nations will take that advantage, and deregulation by more scrupulous nations will be re-framed as gaining necessary competitive advantage, and will be demanded by unemployed citizens and business minds alike.
we already have cameras everywhere, you cant participate in society without signing up for and along with someone else massive system of bullshit. We cant even review utility bills without installing an app or visiting a website and starting an account. 'That future' we are in the middle of it.
> The evil forces haven't even managed to get an encryption-backdoor law passed
They've managed to do a few other pretty evil things recently, though. And they're going to keep trying until they get it through. They only have to win once, and then no government is going to voluntarily give up the extra power and control.
Or you're in an authoritarian regime where not having an implant may result in longer lines, less financing options, no plane access, harder train access, and direct pressure on/from friends.
I'm not worried about those things happening in Canada tomorrow though.
Even if that’s the case (it’s not necessary at all), net change is a terrible metric, because the negatives are externalized often onto those that are already oppressed/marginalized
Well, unlike sci-fi universes, this universe has sci-fi books so people are adequately terrified of that possibility¹. I don't believe this will come to pass.
Imagine being in Gitmo and having one of these implanted to both administer pain, and tell your handlers if you're still feeling like hiding something.
There's diminishing returns along all those axes, suddenly throwing billions of dollars at a problem doesn't automatically yield a fast, high quality result.
Part of the culture of "move fast and break things" is that sometimes the things you think are "broken" are going to end up not being included in the final product, or will be better optimised later, once you understand the process better.
It's like spending time developing a sprue-cutting process is going to be wasted when in the next iteration you refine the manufacturing process to no longer leave sprues connecting moulded parts.
Tesla, SpaceX and Neuralink are very different companies, different kinds of products, different markets with different dynamics. So it's hard to compare.
Elon Musk leaves a trail of destruction with the early employees of his companies but the results are tremendous and have huge implications for humanity. I'll give Musk the benefit of the doubt at this point.
I’m not sure that the high churn rate at the high end is really in Musk’s interest, but it MIGHT be in society’s interest. Instead of sucking up a bunch of highly skilled and driven people indefinitely, the few who are able to work well in that environment stay and are able to get stuff done while the others leave and spread the knowledge elsewhere, perhaps starting up competing efforts that either complement Musk’s efforts or give Musk a run for his money. This is a strong argument against Non-Competes.
It probably is in Musk's interest too. Musk has earned the benefit of the doubt in trusting his decision as to "should Very Smart Person X be employed by Neuralink" or not. There's a lot more to making a company successful than being competent and as much as I find Musk personally distasteful there is nobody with his track record at building transformative private sector companies.
It’s not Elon musk that’s doing the work, it’s the actual workers. Sure, maybe Musk has the capital and audacity to try these things, but that does not mean he’s the one putting in the skilled work required for this
No, it's like saying the owners of aerospace engineering companies don't deserve credit for the designs the aerospace engineers working at them produce.
It's not a statement I agree 100% with, but it's definitely the case these days that individual workers are not given enough credit. It's also the case that the owners, executives, and other high-ranking management types are given far too much credit for successes and far too little blame for the failures.
Without any of the founders and Elon Musk there would be no Tesla Model Y. Doubt there are very many engineers about whom the same could be said. In totality the individual contributors are more important but Elon Musk is the person without whom the enterprise is dead on arrival.
I have personally come to a conclusion that people like Elon will always have disgruntled employees around him, and the bar for expressing dissent with Elon should be high and responsibility should be on them to express the dissent and not for us to read between the lines.
Unless every employee who has quit documents convincing literature they don’t get any of my attention.
Please don't snark or post unsubstantive HN comments. Maybe you don't owe CEO cult figures better, but you owe this community better if you're participating in it.
I think it's likely that OP was genuine and not snarky, Mr.Dang.
As in, since Musk is known to not be 'punctual' as he himself has said in his presentations, since he is known to not have these two in sync 1. Announcement of products with production dates 2. Actual product market release dates, OP may have meant he is surprised Musk seem to be rushing with Neuralink and not delaying as seem to be the case with his other companies.
Given Tesla's track record, I find that statement hard to agree with. I have no qualms necessarily with most nuts and bolts part of it, but the entire self driving division reeks of rushed compared to what the project actually requires.
Isn't Tesla's FSD computer best compared to Google's TPU? Both use integer operations instead of floating point and are optimized for matrix operations.
It's a great product and genuinely impressive, but you're comparing specialist vs generalist hardware. Half an order of magnitude improvements are table stakes, otherwise it's crazy not to just buy off the shelf.
My 2070 Super does 9 TFLOPs but it's paired with an 3900X which barely does 1... is my graphics card an order of magnitude better than my CPU?
Ok the freeway, my Model 3’s autopilot is amazing. It’s not 100% on surface streets, but it’s impressive given all the edge cases and the scale of the problem.
It’s clearly marked as in development. What’s impressive to me is just how good the 80% solution they have is, even without the additional FSD package.
Is a custom chip really that big of an accomplishment for a company the size and scale of Tesla? I really don’t know. Maybe someone here has some insight into semiconductor design and fabrication and can provide some insight into just how hard it would be to build a chip similar to what Tesla has built.
Did/do you work there? Is it a worthwhile place to work for a senior eng w/o much corporate experience? (15yr Bootstrap SMB, BSci academia, early stage startups, corporate retail service). (Anyone? Thanks.)
> So you're expecting 70x more people to die from the vaccine than from the virus itself.
I have no idea where this number comes from?
Do you think I said the vaccine will kill 1% of the population?
But then do you believe the virus will kill 1%/70? Which is .01% of the population? That's a order of magnitude less worse than the flu. That's probably like a common cold death rate.
The vaccine won't save 1%.
It's just a way of attacking something that's on track to kill 1% of the population. That's not a hard 1%, I think it's .4% but the new data this month is coming back with .9% and I'm also combining economic deaths.
But certainly I think the risk/reward for the coronavirus vaccine is more than what Elon is doing.
