I keep thinking it's a reminder "here's what you CUDA done if you knew how to" :D

Diffusion in its other variants (Perona-Malik, Edge-Enhanching) was used and was well known in CV right? and they were used quite a bit. Modern interest in diffusion seems to have progressed rapidly post the success of DALL-E [the latter part is what I think].

I wonder if its possible to setup something like backblaze + mount it on colab [something like fsspec maybe?]

Even as someone in the field, I was quite thrilled with how it performed [especially in the many Indian languages I used while interacting].

Definitely recommend anyone using mac terminal/iTerm to give it a shot, its very user-friendly [and power-user friendly].

Looking sharp, and yes Twitter was a challenging place to be the past few days._.

I am so glad that I'm not the only one XD I feel the same way about systems too.

Didn't the early work from Deep Mind on atari directly work on image space? (More like RL adopting deep learning techniques in CV). I also remember this: github.com/wuhuikai/TF-...

Yes, and the original sin of computer science was its haughty conception as a subfield of mathematics rather than an engineering discipline.

If there's an error message, is it really unexpected?

What would you consider as evals that semi-reflect practical utility right now?