OpenAI is hiring safety researchers to study AI self-improvement risks

OpenAI's Preparedness safety team is hiring researchers with a salary of $295,000 to $445,000. The job is to study what could go wrong when AI can train a stronger version of itself. The ad says they need people with good taste and strategy to think about problems that don't exist yet but might in the future.

OpenAI wants to automate its research work. CEO Sam Altman said last October they plan to run an automated AI research intern using hundreds of thousands of chips by September 2024, and achieve a real automated AI researcher by March 2028. He admitted they might not reach these goals but thinks it's good to share them publicly.

The researcher will focus on defending against data poisoning attacks, building tools to explain how AI models think, and running experiments to understand model safety. Anthropic's study in April showed limited but positive results in using AI to oversee stronger AI models. Anthropic co-founder Jack Clark predicts about 60% chance that AI research will be fully independent of humans by the end of 2028.