I felt the same, and also thought that the article didn't really do enough to explain why "write better reward functions" isn't enough, especially because in ga

I felt the same, and also thought that the article didn't really do enough to explain why "write better reward functions" isn't enough, especially because in gaming, if anyone set the same goals these researchers did, humans would eventually learn the same patterns.