Ajeya Cotra: danger can arrive before recursive self-improvement

‘A slightly more capable agent swarm … is more concerned about avoiding detection by humans, might just succeed in getting a foothold’

Enter your email to view