James Hoffend comments on AI Safety has a scaling problem