In TrajSynth, created for 32-channel diffusion, I explore the use of artificial intelligence as a generative tool - not for producing entire sound materials directly, but for sonifying the dynamic numerical processes that AI methods inherently generate. These constantly evolving numbers, often ignored in mainstream AI-music platforms, carry a latent energy that I find highly effective for sound creation. Rather than relying on ready-made platforms, I focus on tapping into the internal workings of AI – extracting numerical data from within neural networks or training loops - and translating tha...