Automated Alignment Researchers: Using large language models to scale scalable oversight