Tesla shows off the AI supercomputer training what it hopes will one day be an actual self-driving car If you're wondering what it takes to develop a self-driving car, know that Tesla is using a 1.8-exaFLOP AI supercomputer packed with 5,760 GPUs that train neural networks it hopes one day will power autonomous vehicles. The machine was described by the automaker's senior director of AI, Andrej Karpathy, during an online …
I find it fascinating that rarely you see people who actually develop the product and do the work to become billionaires or even millionaires. Salaried employees never enjoy the value they produce unless they belong to the select few who get the shares in the company. That also seems to be guarded with small prints.
Should salaries become tied to company revenue? (Because profits disappear for tax purposes).
> While everyone would like a salary that goes up when the company is doing well, I doubt many people would appreciate receiving less salary when the company is doing badly.
The revenue of many companies is disproportionate to the employee salaries. If company makes billions but workers get a minimum wage, this would mean those workers could be getting 10x more.
There should be safeguards like the salary shouldn't fall below national minimum wage.
Bonuses currently are discretionary. It could look like a base salary plus a mandatory bonus based on revenue though.
People who work in successful startups generally do become millionaires... And people who work at Alphabet certainly become millionaires if they stay long enough. The entry-level compensation for engineers is 6-digits, and a third of that is in shares.
Tesla engineers also certainly receive shares, and the share price got multiplied by ten in the past two years.
There are people who achieve great results without being rewarded, but not those this article is about.
Isn't that another way through which company avoids paying tax? What if an employee got paid a salary that enabled them to buy shares themselves if they wished so and without having to be caught by a small print?
If you compare these 6 digit salaries to what kind of value the employees generate, it is laughable.
If a company paid properly, they wouldn't build flashy offices in prime locations nor would have CEOs flying in private jets. Imagine how much employees could do with that kind of money - it is the value they generated that has not been shared with them.
People in IT can't grasp how they are being exploited.
People are not paid the value they generate, they are paid how much they accept to work for. These employees are generating value, but they could easily be replaced if they quit their job. If your job is so easy that a lot of people can do it, you don't get to demand huge rewards.
And again, the employees of these companies are paid more than 99% of people, the engineers in particular.
In an ideal market economy, everybody might be paid the amount they are willing to work for and that would accurately reflect the value of their work. E.g., if creating an office suite would be extraordinarily valuable, lots of skilled people would start programming until the competition would drive down the price.
In the real world, companies go to great length to extract 'rent', i.e., collect money based on locking the customer into a system without choice. You pay rent for your apartment because you need a place to live. You pay rent for MS office, because you need it to read/write/edit documents that your employer (or others) request in the MS format. You pay 30% extra for your iphone apps because two big companies locked down the market with proprietary phone operating systems. You pay extra money to lawyers, doctors, and other professional groups that protect their status via licensing / accreditation procedures. You pay rent to many companies and people that create an artificial scarcity and make you pay for the scarce resource. Prime company example was Enron, but most modern companies try do the same thing if they get the chance. And the same mechanism is at work, when board members decide that only a member of their exclusive circle can possibly steer the company (as opposed to, e.g., pulling up a talented engineer from their company).
So your salary, even if you are forced to accept it, doesn't really reflect your fair value. The person cleaning the toilets at HP is surely smart enough to run HP into the ground -- but is payed a tiny fraction of the CEO's salary. And don't tell me that more people are willing to clean toilets instead of flying the corporate jet around the world. So your income reflects a complex environment where some manage to extract rent and others pay extra.
It's the same across Europe isn't it? Salaries in northern Europe are at least half as much as they are in the US for similar work. (And even worse the further south you go. A quick search shows that a senior (5+ years experience) C++ software engineering position in Madrid, Spain, for example, has a range today of €18k-€30k.)
It's a catch-22 though. People who refuse to work for the going rate have trouble finding jobs. Employers will pay as little as they can, but then those same employers are complaining all the time that they can't find engineers.
This post has been deleted by its author
This post has been deleted by its author
Self driving cars are coming, they're one major breakthrough away. Those that think otherwise are the ones thinking internal combustion is still the way to go, despite all the indicators to the contrary
PS The UK head of the organisation I work for described electric vehicles as 'a fad that will go away soon' just before he ordered his top of the range executive plugin hybrid company vehicle. Seems he didn't believe it either
Self-driving cars are already working in ideal conditions. Ideal conditions being no rain or snow, and no badly signalled construction work.
The current computer vision technology would need a breakthrough indeed to handle bad weather; as it is, snow causes too much noise for the computer to understand what is going on, and even after many advances there may well be cases where the computer will refuse to drive — like a human would in a blizzard.
And the self-driving capabilities would need to get better by an order of magnitude to know how to act when construction work has made it impossible to use the road in a normal way; they currently completely lack the ability to say "screw the rules; I can drive on the other side of the road for ten seconds if I need it". And even then, they would fail to understand what to do given a complicated enough setup — again, much like a human might need instructions, be it to go another way.
An extra would be understanding the weird setups that locals understand but has visitors utterly confused, such as magic roundabouts and the like. These can be solved by adding more lines of code, but I'm sure the engineers would prefer not spending one year to find a software solution to the latest creative bit of urban planning.
"when construction work has made it impossible to use the road in a normal way"
There was a great example of that posted in a comments section on a similar story here a little while ago. The self-driving taxi had a total brain-fart over some cones in the road blocking the lane it wanted to use.
Locally, there are some roadworks which are light controlled leaving a single lane only, alternating direction of traffic flow. It's be interesting to see how much hesitancy an autonomous car has in that situation and how much rage it causes in the drivers behind it hoping to get through on this green light and not the next cycle or the one after that. Especially if the self-drive is travelling in the direction whereby the only open lane is on the wrong side of the road.
I don’t drive well In snow and I’m not alone.
I’m actually quite happy to drift a rally car on loose surface, but when it fresh snows on public road shared with other people, I stay off the road until it’s cleared. There are a lot of people who should consider doing the same.
Just being Canadian doesn't prove a thing about your ability.
I've seen Highway Thru Hell and Heavy Rescue 401. Canada seems to have it's fair share of people who don't know what to do in snow, even when it's fairly light :-))))
I would be very happy if fully self driving cars were imminent (I can drive but really dislike it & use public transport whenever possible) but, certainly for UK driving, it's going to be after my lifetime before an AI car can cope with what UK roads & weather throw at it with zero human intervention required.
Disclosure: Live in a rural area so all the fun of cattle crossing, single track roads (with 2 way traffic, so the "negotiations" with other road users as to who reverses into a passing place), horses (& even horses leading carts) not uncommon (and needing a real wide and slow overtaking scenario to avoid spooking them)), crossroads with no lights / roundabouts (so again, "negotiations" with other road users).
The "negotiations" bit will be interesting - person to person sometimes it can be done via lights, horn, small movements of the vehicle but often massively aided by a facial expression / hand gesture (not an option open to an AI unless start adding external animated screens so AI can do a "beckoning on hand gesture" or whatever.
"The "negotiations" bit will be interesting - person to person sometimes it can be done via lights, horn, small movements of the vehicle but often massively aided by a facial expression / hand gesture (not an option open to an AI unless start adding external animated screens so AI can do a "beckoning on hand gesture" or whatever."
I've seen many references to cars being able to "talk" to each other, mainly in relation to sending out notices of slow/congested traffic etc. But I don't recall any mention of standards so that cars from different manufactures can "talk" to each other. Fun days ahead!
One way is to test it in shadow mode on your fleet of vehicles. Doing this over time identifies more and more of the weird edge cases that can be added to the training set for the network whenever the real driver and the shadow driver have a difference of opinion. The dataset is constantly growing and evolving.
So you've taken your driving test - how do we know how you'll respond in fifteen years time to a situation that wasn't in your test?
We don't. But at least with self driving vehicles every vehicle will be able to learn from the outcome, whether good or bad.
When a car can pass a driving test (things like manoeuvres are already basically trivial) then it can be let loose on the road - it will still be learning, and improving, but it will rapidly be doing millions of times the mileage that a typical learner (because you're still learning after you passed the test) will be able to.
At the moment Tesla is already learning on the road - the autopilot is recording and deciding, even when the car is in manual control... So the ability to beam back "Hey, I'd have done this differently" absolutely will be improving the situation.
The Register - Independent news and views for the tech community. Part of Situation Publishing
Biting the hand that feeds IT © 1998–2021
