AI reasoning gains and the case for stronger biodefense

A post questions the timing of warnings about AI-related biological risk, arguing that rapid capability gains were foreseeable when OpenAI released o1. The author says reinforcement learning and models “thinking” before answering favor progress on tasks where iteration is cheap and correct answers are easy to verify. They argue that biological threat vectors have not changed and advocate stronger biodefense regardless of AI, noting that COVID-19 preceded ChatGPT.