A look at the math that OpenAI claims computes faster Fourier transforms

What happened: A blog post examines OpenAI's math repository claim of a faster Fourier transform, arguing it is a galactic algorithm rather than a practical breakthrough. Key numbers: OpenAI's estimated constant C was 10^100 when N=10^12. The post says the Cooley-Tukey FFT would be 10^99 times faster even at n=10^100. Why it matters: The post argues big-O notation hides unfathomably large constants, so the algorithm is only asymptotically faster and runs faster only at impossibly large n. It cites Lipton and Regan that galactic algorithms can still spur research.

A blog post examines OpenAI's math repository claim of a faster Fourier transform, arguing it is a galactic algorithm rather than a practical breakthrough.

The post argues big-O notation hides unfathomably large constants, so the algorithm is only asymptotically faster and runs faster only at impossibly large n. It cites Lipton and Regan that galactic algorithms can still spur research.