Big problem here is that they're probably not going to be able to train the software without massive amounts of real driving conditions. Doing it in a parking lot at the office won't work. Obstacle courses also won't work. You need all the chaos that comes with real driving. And even then it's going to be constant false-positives in terms of braking or swerving. I've used software before and that's all these people are, software developers. If candycrush can't get it right all the time, why the fuck leave it up to an app while driving? Plus the traffic laws are slightly different in every state. Some states you can turn on red, some you can't. You can't even standardize the learning once you've cracked it in NV or CA. You have to train the cars in each state. 10 cars training in each state is 500 cars. That's not even a representative sample of everything drivers encounter. It's a statistics problem that you can't brute force in any reasonable amount of time.
I suspect they'll fudge everything enough to get the rate of accidents down somewhere slightly above normal traffic. Then they'll just expand their legal department to be bigger than their engineering department if it's not already. Every time someone tries to sue they'll blame the driver or pedestrian. This is exactly what happened when cars were invented anyways. We've already been here. It'll be the individual's fault not the fault of the new technology. Then we'll just be okay with that idea and blame pedestrians for being too stupid to not get killed by a murderbot car. Years later the car accident deaths will go up and we'll just argue about it until nobody cares anymore. By then we'll probably be dead anyways.
Will the AI evade potholes? Because that's going to ruin your car pretty quickly if it doesn't. Also imagine letting your AI overtake another car, or a truck, in the night lmao.
They will never get it to work well no matter what they do because they insist on only the optical camera sensor. I work in this industry (but with busses), it is the accepted understanding that there will never be enough data from a camera and it will always be ambiguous and too much noise. Investor and Tech bro want to pretend they can defeat physics but they are wrong. Optical radar and radar are both very common and it is normal to combine as many sensor as possible to not miss anything.
This is one of the big reasons that I really liked the game Watch Dogs Legion. The game has self-driving cars in it, and they make a point of saying that the cars can't operate alone, they essentially need a massive network of surveillance drones on every street, and it turns out those surveillance drones can also, like, do surveillance and shit.
I don't know what the fuck is wrong with SV dudebros who insist on fighting physics, it's so stupid, and this is far from the only case. Colossal waste of resources.
oh come on, all they need to is continuously solve multiple technologically difficult problems in an infinite variety of unpredictable scenarios with a perfect success rate at the risk of injuring or killing dozens of people every minute. how hard could it be? Especially with the typically error-free process of enterprise software development.
Big problem here is that they're probably not going to be able to train the software without massive amounts of real driving conditions. Doing it in a parking lot at the office won't work. Obstacle courses also won't work. You need all the chaos that comes with real driving. And even then it's going to be constant false-positives in terms of braking or swerving. I've used software before and that's all these people are, software developers. If candycrush can't get it right all the time, why the fuck leave it up to an app while driving? Plus the traffic laws are slightly different in every state. Some states you can turn on red, some you can't. You can't even standardize the learning once you've cracked it in NV or CA. You have to train the cars in each state. 10 cars training in each state is 500 cars. That's not even a representative sample of everything drivers encounter. It's a statistics problem that you can't brute force in any reasonable amount of time.
I suspect they'll fudge everything enough to get the rate of accidents down somewhere slightly above normal traffic. Then they'll just expand their legal department to be bigger than their engineering department if it's not already. Every time someone tries to sue they'll blame the driver or pedestrian. This is exactly what happened when cars were invented anyways. We've already been here. It'll be the individual's fault not the fault of the new technology. Then we'll just be okay with that idea and blame pedestrians for being too stupid to not get killed by a murderbot car. Years later the car accident deaths will go up and we'll just argue about it until nobody cares anymore. By then we'll probably be dead anyways.
This shit sucks.
deleted by creator
Will the AI evade potholes? Because that's going to ruin your car pretty quickly if it doesn't. Also imagine letting your AI overtake another car, or a truck, in the night lmao.
They will never get it to work well no matter what they do because they insist on only the optical camera sensor. I work in this industry (but with busses), it is the accepted understanding that there will never be enough data from a camera and it will always be ambiguous and too much noise. Investor and Tech bro want to pretend they can defeat physics but they are wrong. Optical radar and radar are both very common and it is normal to combine as many sensor as possible to not miss anything.
This is one of the big reasons that I really liked the game Watch Dogs Legion. The game has self-driving cars in it, and they make a point of saying that the cars can't operate alone, they essentially need a massive network of surveillance drones on every street, and it turns out those surveillance drones can also, like, do surveillance and shit.
I don't know what the fuck is wrong with SV dudebros who insist on fighting physics, it's so stupid, and this is far from the only case. Colossal waste of resources.
oh come on, all they need to is continuously solve multiple technologically difficult problems in an infinite variety of unpredictable scenarios with a perfect success rate at the risk of injuring or killing dozens of people every minute. how hard could it be? Especially with the typically error-free process of enterprise software development.