Stuart_Armstrong comments on When Goodharting is optimal: linear vs diminishing returns, unlikely vs likely, and other factors