In the late 90s some friends at CMU created Psychodeli — a groovy spiral engine that shipped in After Dark. The whole thing started from one formula: C = sin(pitch · r + sin(θ)). Twenty-five years later I set out to reimagine it with AI support — not vibe coding, but coding the vibe. 1,662 commits later, it's a real-time audio-reactive visualizer running 8,600 lines of GLSL.
I'm a cognitive psychologist not mathematician. But we swapped simple trig for Riemann zeta functions and modular forms, replaced Euclidean distance with L^p norms, and somewhere in there the code grew Turing patterns — leopard spots from pure trigonometry.
The AI journey was epic: 58 abandoned features, 39 reverts, and a throwaway rate that peaked at 18% in August. The essay documents the whole thing — with git stats revealing the 11/25 unlock of new gemini and claude models.
Is this the state of the art in AI music listening?
3 comments
[ 2.1 ms ] story [ 16.7 ms ] threadI'm a cognitive psychologist not mathematician. But we swapped simple trig for Riemann zeta functions and modular forms, replaced Euclidean distance with L^p norms, and somewhere in there the code grew Turing patterns — leopard spots from pure trigonometry.
The AI journey was epic: 58 abandoned features, 39 reverts, and a throwaway rate that peaked at 18% in August. The essay documents the whole thing — with git stats revealing the 11/25 unlock of new gemini and claude models.
Is this the state of the art in AI music listening?