Re-reading an article I wrote in 2017, and I'm finding I could have written it yesterday. It's fascinating how the limitations of deep learning have stayed the same since. Same problems, same failure modes, no progress.

→ View original post on X — @fchollet,