Annotated reply to Bengio's "AI Scientists: Safe and Useful AI?" — LessWrong