OpenAI claims theoretical speedup for Fourier transform

OpenAI published a mathematical result: a deterministic exact discrete Fourier transform with work bounded by O(N·(log N)^(1−10⁻¹³)), asymptotically better than the classic O(N log N). The bound holds in an exact-arithmetic model and promises no real-world speedup; researcher Ben Recht flagged hidden constants and assumptions disconnected from hardware.

• Bound of O(N·(log N)^(1−10⁻¹³)) versus O(N log N) for FFT
• Result holds in an exact complex-arithmetic model with unrestricted coefficients
• Critics cite large hidden constants and no link to real hardware
• No demonstrated reason to replace working FFT libraries