[NEW MODEL] SupraLabs just released supra-title-FFT-preview, 115K samples, almost 10x our first chat title dataset