I've been in the AI world for like two years now and I am continuously amazed by how many problems are no longer technical problems.

The bigger cost of implementing automation is fine and institutional inertia.

Kind of fun to use inside of Zed.

PR is up github.com/zed-industri...

Even that joy has been taken away.

Based on GitHub actions logs they're now routing more of actions runners through Azure. Hope they finish this transition soon and stabilize.

Widgets are hairy. Code is here: github.com/nteract/elem...

I'm assuming you're passing through comms somehow?

The other interesting tidbit here is that the IPython widgets are wholly implemented in react. No dependencies on backbone or any of the other legacy bits. Pure react, tailwind, and radix UI. nteract-elements.vercel.app/docs/widgets...

This is rad, I'm loving it.

My theory here is the major labs have done very little or no RLHF on visual explanations and diagrams? Seems like a blind spot. Image generation alone or text generation alone are very well optimised but the combination seems uniquely hard for current models.

Looking good on `main`. So happy to see the REPL in Zed get some love. This is not sped up. Pure Rust runtime kernel via @posit.co (Ark) and Zed's totally native UI.

Any idea how to get out of the orange wheel cycle? I don't have any logs in the server about failures here. 404s in the web console though on `/api/events/subscribe` though.

Oooh, this is a good advancement since I last tried jupyter-collaboration.