It seems that generating hypotheses is no longer the main problem. The harder problem may be continuously collecting real-world data, building accurate world models, and using them to guide experiments. Is this where automated research is heading?