I've spent several years studying machine learning security with the goal of making ML reliable before it is used in more and more important contexts. Unfortunately, ML capabilities and adoption are growing much faster than ML robustness.

→ View original post on X — @goodfellow_ian,