OpenAI algorithm for the FFT is less a breakthrough and more a malicious exploit
A critique argues OpenAI's faster FFT claim is a galactic algorithm, slower than Cooley-Tukey at realistic sizes.
Argmin reviews an OpenAI mathematics paper that claims an asymptotically faster Fourier transform than Cooley-Tukey. The author says big-O notation hides constants around 10^100, so the method only wins for impossibly large n, and calls it a galactic algorithm. Even at n of 10^12, estimated constants were about 10^100, leaving the classical FFT vastly faster in practice. The post treats the claim as an artifact of loose asymptotics rather than a usable signal-processing breakthrough.
- OpenAI's math paper claims a slightly better asymptotic FFT than Cooley-Tukey.
- Hidden constants were estimated near 10^100 when n is 10^12.
- The method would beat Cooley-Tukey only at impossibly large sizes.
- Author labels it a galactic algorithm, not a practical breakthrough.
From the original article
8 min read at argmin.net
Hi there, argmin readers! Today’s post looks at part of OpenAI’s math repository that claims to compute faster Fourier transforms. It’s more technical than usual, but I tried to add some tutorial material for people who need a refresher course on signal processing or computational complexity.
By mochomocha · argmin.net
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.