Google’s co-founder suggested that threatening an AI improves performance. Researchers discovered that it sometimes works.
