Wow, this is saying something since it's coming from(the),"Director of AI at Tesla, leading the Autopilot Vision team. Previously OpenAI, CS231n, PhD @ Stanford. I like to train large deep neural nets "
AI at Tesla is a scam, so it probably doesn't say what you think it says.
It's a shame that Karpathy actually seems to know what he's talking about. Maybe he originally bought in to the Tesla vision and doesn't have the integrity to admit it's failed. Maybe he's like von Braun, doing great evil because he wants funding. Either way we lost a seemingly talented researcher.
Never the less, if someone that works in AI gets impressed by an AI I will take it seriously. I don't use TikTok but the recommendation engine must be outstanding. I think I'll stay away. I don't want to lose any more time to online apps.
No, nor have I reported Deepak Chopra or Amway or Gwyneth Paltrow. There are blatant scams all around us. A lack of interest from the feds doesn't mean much.
Besides, the SEC has enough on their plate when it comes to Tesla.
>So it does say what he thinks it says.
Doctor Oz is, by all accounts, a brilliant surgeon. Would you follow the medical advice on his show?
Over the years, they repeatedly pushed this narrative to sell cars. Tesla promised on several occasions to have millions of fully autonomous "robotaxis" on the road in 2020, making their owners in excess of $30,000 a year. Currently there are, of course, zero.
Take a look at this video for what Tesla describes as a beta for their "Full Self Driving" system. A beta is, of course, meant to be feature complete software in final testing. Does this look like it's nearly ready to be pushed out to every car on the road?
They continue to work on this project despite knowing it is fraudulent, which makes them complicit.
Besides, I don't see that they're doing solid work. It would be solid work if they ditched pure vision and moved to a system that works. Instead they are putting in heroic efforts on a dead-end technology. That is impressive in the same way that getting Doom running on a TI-84 would be impressive.
I don't know, it depends on what your goal is. If you want to emulate human driving, then pure vision is a viable way to go. After all, humans drive also using pure vision.
But your car has to drive a lot more cautious than with more sensors, like lidar. Can't be 100% sure this white blob is not a truck? Then you have to slow down. With lidar, you can be a bit more robust, but you still need vision to identify objects.
And if your goal is to build a non-human driver, that is one that drives "perfectly" and pushes the speed envelope perfectly, then I think lidar is also only a stop-gap solution. What you'd want is active components in the street and in other cars. In other words, a virtual rail. In that case, you can accelerate and brake as aggressively as the humans inside would tolerate. You could accelerate together with the car in front of you in traffic jams, etc.
Not really. Why are you blatantly drinking that marketing BS cool aid coming from a wack twittering ceo? That so called "pure vision" is backed by our brains trained over decades. This is exactly why you have to be 18+ to be able to drive without any supervision.
Even if we gave them more time, it seems that it still doesn't work and is still completely unsafe for Level 5. Compared to other competitors when tested, this is all Tesla has to show for progress? [0]
A very long way to go for the Level 5 'robo-taxi' readiness claim by Tesla.
Obviously they've made some progress. It would be hard for them to have accomplished nothing at all after all they've invested. It's still nowhere near safe for public roads and there's no reason to think it ever will be.
Even their driver assistance features work worse than the competition.
(his Twitter description)