He said something along the lines of "I have yet to see an application where evolutionary techniques outperform gradient descent approaches." His talk was about

He said something along the lines of "I have yet to see an application where evolutionary techniques outperform gradient descent approaches." His talk was about parameter optimization for costly experiments (e.g., manufacturing and testing a vehicle), so his emphasis was on optimization that minimizes the number of function applications. But he seemed to be an expert with a lot of experience.