Model Hacking Beyond Just the Static Audio Format

I'm interested in how we can move beyond just the audio format too. I think like latent space is an interesting thing to hack with and trading around models or latent spaces when they're somehow interoperable would be really cool. Imagine our webs of live streams that are live remixing each other, even on YouTube and Twitch. Like this is, yeah, around the corner, we can make it like an open source toolkit for that. And I've been researching a biologically inspired method called conceptors which trains tiny tiny tiny little grains from an RNN that is random and untrained of the output layer. It learns how to control this chaos.