[amelius]

But ML relies on GPUs with lots of parallelism, so e.g. O(N^3) becomes O(N^2) for N smaller than the number of cores (roughly speaking).