New preprint: Improving scalable oversight with co-trained monitors

My new preprint, Improving scalable oversight with co-trained monitors, studies how co-training monitors alongside workers can make AI oversight more robust to monitor evasion.