4 ms·

I was at an ML conference in 2017 and the keynote speaker asked the audience how many years they thought it would be until ADAS lvl 5 was on the market. The key

I was at an ML conference in 2017 and the keynote speaker asked the audience how many years they thought it would be until ADAS lvl 5 was on the market. The keynote had proudly proclaimed 5 years, and then audibly scoffed when about 90% of the hands went up for "20 or more years". There's such a disconnect between real engineers and cheerleaders.

celeritascelery 7y ago
I have a friend who a mechanical engineering professor doing robotic research. He says that one of the reasons that robotics has not advanced more is because it has primarily been the domain of computer scientists who don’t operate in the real world. Every robot works great in a simulation. Same could be said for cars.
tim333 7y ago
Though there's the Kurzweil argument that people's intuition doesn't work well with exponential growth. The engineers are probably thinking these systems work terribly and have for decades and so will for decades more but meanwhile GPU performance is doubling every year or two and many people are trying to take advantage of that. I've been following this stuff casually since about 1980 when I totally failed to write an AI program and since then processor performance has been steadily increasing, first with clock speed and now more with multi cores. Hans Moravec, a robot guy did a reasonable estimate that to get equivalent hardware to the brain you'd need about 100 teraflops and just recently the first 100 teraflop GPU has come out (https://www.nvidia.com/en-us/data-center/v100/ https://www.nvidia.com/en-us/data-center/v100/) which is kinda historic in a way, maybe a billion times faster than a cheap computer in 1980. Of course this will keep going for a while so the situation will flip from we can't do much because processors are much slower than the brain to the other way around. (update NIDIA may have fudged the numbers a bit but anyway.)

celeritascelery 7y ago

I have a friend who a mechanical engineering professor doing robotic research. He says that one of the reasons that robotics has not advanced more is because it has primarily been the domain of computer scientists who don’t operate in the real world. Every robot works great in a simulation. Same could be said for cars.

tim333 7y ago

Though there's the Kurzweil argument that people's intuition doesn't work well with exponential growth. The engineers are probably thinking these systems work terribly and have for decades and so will for decades more but meanwhile GPU performance is doubling every year or two and many people are trying to take advantage of that. I've been following this stuff casually since about 1980 when I totally failed to write an AI program and since then processor performance has been steadily increasing, first with clock speed and now more with multi cores. Hans Moravec, a robot guy did a reasonable estimate that to get equivalent hardware to the brain you'd need about 100 teraflops and just recently the first 100 teraflop GPU has come out (https://www.nvidia.com/en-us/data-center/v100/ https://www.nvidia.com/en-us/data-center/v100/) which is kinda historic in a way, maybe a billion times faster than a cheap computer in 1980. Of course this will keep going for a while so the situation will flip from we can't do much because processors are much slower than the brain to the other way around. (update NIDIA may have fudged the numbers a bit but anyway.)