How Does Automating AI Research Create Superintelligence?

Leopold Aschenbrenner says automating AI research takes the path from AGI to AI vastly smarter than humans within a year or a couple of years. A couple of years of lead could be utterly decisive in military competition.

AGI shows up as a drop-in remote worker that can replace remote jobs like Aschenbrenner's or Dwarkesh Patel's. By default, turning the crank one or two more times produces a system smarter than humans. Aschenbrenner says one of the first jobs automated is the AI researcher engineer, and automating AI research is what makes the pace very fast. Algorithmic progress is already half an order of magnitude a year. GPU fleets in the tens of millions for inference, or more, can run 100 million human equivalents of automated AI researchers. He says you can maybe do a decade of ML research progress in a year, a 10x speedup, and make the jump to AI vastly smarter than humans within a year or a couple of years. The acceleration then broadens into research and development across other fields. At the extreme, a billion superintelligent researchers, engineers, and technicians are on the cluster, and a billion AI researchers smarter than the smartest human researchers figure out robotics.

Aschenbrenner poses the question of when the CCP and the American national security establishment realize superintelligence is going to be absolutely decisive for national power. He says the result is fairly unlike any other technology, because a couple of years of lead could be utterly decisive in military competition.