Thoughts on the conservative assumptions in AI control — AI Alignment Forum