Search Results
6/23/2025, 7:53:05 AM
>>16705422
>more efficient at learning by many magnitudes
Computers of course are already better than us at basic arithmetic and playing chess, but the difficulty is in teaching them to think like a human being. This requires the multiplication of matrices, it is more than flipping 1s and 0s but it is ultimately reduced to flipping 1s and 0s.
It is mind-boggling to consider the calculus of higher dimensions. We have to reduce it to these nodes because it goes well beyond 2D and 3D conceptualization, two or three dimensions. We are talking thousands of dimensions and calculating the derivstions of a "slice" of each dimension one at a time to make tiny adjustments. It requires immense computing power. GPUs are best suited for it, as are supercomputers. A desktop CPU will take hours to run a model that can predict images only about 100 pixels wide.
>more efficient at learning by many magnitudes
Computers of course are already better than us at basic arithmetic and playing chess, but the difficulty is in teaching them to think like a human being. This requires the multiplication of matrices, it is more than flipping 1s and 0s but it is ultimately reduced to flipping 1s and 0s.
It is mind-boggling to consider the calculus of higher dimensions. We have to reduce it to these nodes because it goes well beyond 2D and 3D conceptualization, two or three dimensions. We are talking thousands of dimensions and calculating the derivstions of a "slice" of each dimension one at a time to make tiny adjustments. It requires immense computing power. GPUs are best suited for it, as are supercomputers. A desktop CPU will take hours to run a model that can predict images only about 100 pixels wide.
Page 1