The risks from current AI capabilities that are diffusing throughout society are more pressing than potential future risks from Recursive Self-Improvement