May KAN will be the next AI disruption step

KANs (external link) revolutionize traditional architecture by swapping out weight parameters for univariate function parameters. Utilizing multiple variables like B-splines, KANs create smooth, adjustable curves. Nodes sum these outputs without nonlinear transformations, enhancing accuracy. The secret lies in adaptive univariate functions, which minimize approximation error and achieve higher accuracy with fewer parameters by intricately modeling data relationships.