Remix.run Logo
userbinator a day ago

Music files (releasing in order of popularity)

Increasing or decreasing? IMHO increasing would make more sense, as the most popular music is already mirrored in countless other places. It's the rare stuff that is most in need of preservation.

I wonder how much of the content there is AI-generated. Honestly, even as someone who was initially skeptical, I've found some of it to be rather good --- not knowing that it was AI-generated at first. Now if they could only reverse-engineer the prompt and only store the model, that would be an extremely efficient form of "compression".

reassess_blind 19 hours ago | parent [-]

Same model and same prompt won’t necessarily create the same result, unless I misunderstand how these audio models work.

squigz 18 hours ago | parent [-]

It's possible to generate the same images and text from LMs by tweaking the settings, right? Are audio models different?

Philpax 15 hours ago | parent [-]

Yes and no - yes, in theory, but in practice, non-determinism can be introduced at different points along the stack. See Thinking Machines' post on LLM non-determinism: https://thinkingmachines.ai/blog/defeating-nondeterminism-in...