Preventing Language Models from hiding their reasoning — AI Alignment Forum