/biz/ - Business & Finance

>>22639768
Tesla's problem is precisely that it refuses to rely on even proven ideas, let alone new ideas. It uses old things and ignores good practice such as redundancy.
Roadrunner walls are how you trick tesla's, as an example, except you paint some adversarial noise (but if the algorithm updates between the time you paint and the time the car sees the image, it already stops working). The point, however, was that even humans get tricked by this shit. This isn't "hacking". This isn't even "a flaw in the algorithm". It's an optical illusion. Pic related for another class of examples that can lead to the same sort of issues. The point is, it is impossible, from the data, to tell which one is correct, because there is not enough data. It's like asking someone (or something) to draw the 3D shape that represents a picture of an object. Said object is a square on the picture. Well, is it a pyramid seen from the bottom? A box? Just the square? You cannot possibly know that, because the data plain isn't there. If one human draws a pyramid and another draws a cube, is one of them retarded? If you stick a bait on a hook in front of a fish, is the fish's brain deficient for eating it? (No, it was tricked. Only reasoning could have defended against this, not better visual acuity). Did the driver who was on the highway while the nignogs were retardedly "protesting" criminally deficient for trying his best to honk and stop but not being able to stop quickly enough because it was not possible to see the protestors from that far away? If I take a bird, kill it, stuff it, and use it as a kite, is it "visual hacking" because you'll not see the difference between that trumped up bird and a real bird unless you look at it for a long-enough time to notice it's not obeying normal physics?