To say a bit more: my read is that all these ”hacking incidents” are the result of lots of reinforcement learning through verifiable results where they were doing so much that they weren’t paying any attention to what was getting reinforced, and weren’t paying any attention to the tests
