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

Large language models’ ever-accelerating rate of improvement raises two particularly important questions for alignment research.