I'm always worried my classes are "too easy" or "stuff everybody knows already anyways" so it's great to see when people find the content directly relevant to their day-to-day work.

I have some thoughts around this idea that I need to do some research on. Maybe I'm wrong, but I have a hunch I'm not. Maybe I'll have a paper on this topic in six months or so. Not making any promises however.

Different people have different concerns. I agree it would be great to have an open-source AI with 100% open-source data that the world can contribute to. However, to the extent that AIs are not overtrained and don't memorize images it's not clear to me that the training set is that relevant.

In defense of AI art. www.fxhash.xyz/article/in-d...

In defense of AI art

An impassioned argument in favor of AI image creation as an art form in its own right.

Over on X: Some thoughts about AI in higher education. x.com/ClausWilke/s...

This is amazing: Diffusion models have internal depth and foreground/background maps. This means these models build an internal 3D representation of the scene and then render it. Paper is from a few months ago, but I only encountered it today. yc015.github.io/scene-repres...

Good boys

Sometimes a simple prompt of a few words can yield great results. I love these.

Over on X: Wrote a blog-post about the value of a PhD in terms of personal growth and development of career skills. x.com/ClausWilke/s...

Yeah but the bigger problem of Mastodon is the user interface is atrocious.