Thoughts on AI Safety Camp — LessWrong