An Anthropic researcher just gave us a peek at self-improving AI

Researchers have demonstrated that AI systems can improve their own behavior without human intervention. This raises important questions about AI safety and control that matter to policymakers and the general public thinking about AI risks.