_zoltan_

I appreciate the time you took to reply to me.

Thank you!

so the workers understand cross worker topology? Do they always just see individual tiles? what if there is a dependency - eg I'm streaming data and want to compute something later on an earlier computation? How would that get distributed?

sounds very cool - is this basically "C" support for cuTile? :)

In 2 years once it fails and Intel shuts it down completely you mean :)

That's not entirely true: my work might get zero traction because nobody will know about it.

hmm, I'm sure with a model that is not restricted, one might be able to extract these locally.

Cold, no sun AND no alcohol? Man that must suck. :)

As Terry Tao said, LLMs are not outsmarting us, they are out remembering us.

I'm fairly sure your understanding is not fully accurate.

But reading the doc if it's maintained llmwiki style is much cheaper than reading the code on a large codebase.

vast.ai?

reaching the goal is what matters. I never understood people who don't get satisfaction from getting there. I don't care that AI writes the code.

why does it matter that it's a vibe coded website?

I actually look at the tests first. But for this to work, you have to have good hygiene. If the test makes sense, fails without the PR but passed with the PR, I'm almost happy.

billions? who is running BILLIONS of agents?

tens, hundreds, maybe a couple thousand at a time? absolutely.

I optimize a lot of CUDA and it got really, really good at it.