OpenAI proposes an AI safety technique called iterated amplification. This method allows for specifying complex behaviors and goals by breaking down tasks into simpler sub-tasks, rather than relying on labeled data or reward functions. The idea is still in its early stages, with only simple experiments conducted so far. It's presented as a potential scalable approach to AI safety. --- Why it matters: This matters because it could provide a new way to tackle complex AI goals and behaviors, potentially improving the safety of advanced AI systems. Source: https://openai.com/index/learning-complex-goals-with-iterated-amplification