Had fun this weekend with YOLO and may check out Google Conversational Image Segmentation to augment. Gotta filter the crowd out and then see if I can identify screens, pick and roll, etc.

That’s what Boom supersonic will basically do.

It sucks that music production VSTs pretty universally persist in binary formats. Kind kills vibe coding.

What if you could jam with an AI instead of asking it to spit out a finished song? I built a full house track with Codex + OP-XY and open sourced it all. youtube.com/watch?v=xP7jnW…

Keep thinking.

This is amazing. I've got Codex hacking on Supercollider SynthDefs, and LISTENING to what it produces. It identifies problems and sets about fixing them.

It’s baffling to me that stem separation is not more of a solved problem. There is a clear path to make millions of hours of pro DAW/VST/fx mixes with full stems as a training set.

ChatGPT is disappointingly shallow in its little morning briefs. I love how it can be steered and how it follows up on previous threads, but it’s way too shallow.

This post is about to go viral. Watch. It's well worthy too. Thanks OpenAI for sharing on their Dev Day state of the union broadcast.

Woo! My music coding repo was featured in the OpenAI Developer State of The Union talk. You can check it out at youtube.com/watch?v=xP7jnW…

Made a big break through getting Codex to listen to itself and iterate on sound design. It's thinking of building its own tools to iterate faster.

Tim O’Reilly’s Web 2.0 essay and Tim Berners Lee’s insistence web was read/writable don’t get enough credit for helping create AI. User Generated Content turned out to be key to making AI in our own image, for better or worse.

Firing up a World Series ticket scraper to figure out the market for tickets.