Particularly stiking example of biases in language models: The german short-bio generated by Safari says Virgina Woolf was a male author.

I made a machine learning puzzle: Find out what‘s going on in the different models and data sets to free the ghost of the latent space. Check it out here: github.com/mattminder/l...

Maybe we shoud start the ML equivalent to the demoscene where we train absurdly large models to generate a single small video sequence