The guy who posted the yt video was one of those who upgraded their FSD software a coupleof days ago. His is the release version, not beta.
The guy who posted the yt video was one of those who upgraded their FSD software a coupleof days ago. His is the release version, not beta.
They should all be considered pre-release/buggy until they actually work safely with other road and pavement users, this business of using the rest of us as guinea pigs is insane.
It clearly is, and in two other ways, as well as speed variation. The rather excessive distances from the car in front is controlling it's spacing. It also appears to be adjusting its side space from other cars alongside, rather like a fish in a shoal. This is seen in two ways, the wandering complained of when it finds itself alone in a multi lane road with no other nearby vehicles alongside to judge from, so is lost. And the similar lane choice indecision when veering into a changed number of lanes.So is this saying that the auto system is actually taking it's cues from human controlled cars?
I see more than just comprehension, I think there is algorithm conflict and this may also be behind dropping Lidar and Radar, to simplify the range of inputs. That is on the right lines, but fallible humans are so much better at doing this. They are so poor at coping with complexity of inputs that they simplify "on the fly" by acting only on the perceived priorities of the moment, which are constantly changing, while ignoring what is outside of their perception. They've evolved to do this over countless millions of years in all their past animal and humanoid forms with only the successful surviving.What you see as current deficiency is the limited number of algorithms. The problem is similar to language comprehension.
I agree with that much, no problem at all.Most of my miles are done on motorways, level 3 would be sufficient. I have no doubt that fsd will become reality within a decade.
Want to hear the scary thing about them.Beta testing "self-driving" cars on public roads should be illegal.
I watched last night a yt video of a bunch of robots training another robot. The lesson is simple: Pick boxes from some shelf and deposit them on a conveyor belt. The student robot learns by watching other robots do the job. After a short few minutes, it has figured out the purpose of the job and how to do it. The student robot learns so fast that it scares me a bit. The lead project engineer assured viewers that there is no trick, the filming was continuous from start to finish. It is estimated that it takes 30 robots to train a new one with their factory neuralnet.
But on the other hand, also many years ago, I approached a T junction on my bike; no headphones, totally aware. I looked carefully left and looked right, observed a car coming, and rode out straight in front of it. Luckily the driver wasn't an AI robot but was extremely alert and she stopped in time. I'm not sure how that speaks to my ability to handle the complexity of all the calculations, considerations, awareness and multiple actions.Many years ago, whilst driving a clapped out Astra, I approached an offset cross roads. I was on a side road, crossing a main two lane road, going into another side road offset to the right by half a roads width.
As I approached I went down through the gears, judged the speed and distance of oncoming traffic, cogged down another gear and accelarated across the road - all whilst singing along to a cassette.
The reason I remember it so much is that the episode impressed on me the marvelous multi abilities of the human system. The complexity of all the calculations, considerations, awareness and multiple actions that this body I live in was able to make, without breaking a sweat and still sing along. Personally, I have no knowledge of all the formulae and mathmatics it would take to accomplish such a feat. Yet it was done.
I marvel at such unconcious, somehow innate ability and doubt that any machine will ever be so capable .
Simple task.I watched last night a yt video of a bunch of robots training another robot. The lesson is simple: Pick boxes from some shelf and deposit them on a conveyor belt. The student robot learns by watching other robots do the job. After a short few minutes, it has figured out the purpose of the job and how to do it. The student robot learns so fast that it scares me a bit. The lead project engineer assured viewers that there is no trick, the filming was continuous from start to finish. It is estimated that it takes 30 robots to train a new one with their factory neuralnet.