This is really really useful and something I've been looking to build in the roadmap for my notebook IDE (https://trykrypton.com/)

Any reason why you also decided to go with .py files as part of your detachable kernel architecture?

Thanks! I wanted to drop the dependency on .ipynb for this version, so I went with .py files instead. The detachable kernel itself isn't tied to .py, so I'm considering a dedicated extension like .pynb (Python notebook) to distinguish notebooks from regular Python scripts in the future.

that sounds like an absolute nightmare haha

An engineering nightmare. But as a passenger, it might be nice. Can take time getting on to the container and settled, and then the whole container gets loaded.

I think the idea is a natural next step in GPU's as a service. If you can create the unifying API layer for GPU access I think that would be immensely useful.

do you have proof? Taking these guys at face value is not wise

what did you end up building?

Interesting reply

Hackernews needs a mute keywords feature. Clawd/molt-slop is mass AI psychosis on steroids.

If only there was some sort of thing that would help you build that for yourself.

Where’s the asteroid at

Same place as my latency budget: disappearing fast.

When we need him the most (a world overrun in llms and AI slop) it seems like he's vanished...

bot reply