Ryan Greenblatt – Human level AIs might build runaway superintelligences by 2032

Automating AI research could advance model capabilities faster than the ability to judge their safety. In his interview with Dwarkesh Patel, Ryan Greenblatt, chief scientist at the AI safety group Redwood Research, argues that the most verifiable parts of AI research improve quickly, while recognizing risks from a new training method takes subtler judgment. An automated research process could therefore reward measurable progress while overlooking problems it introduces.

Patel challenges whether some apparent failures reflect limited capability rather than misalignment. Greenblatt also allows that AIs could reach a beneficial cycle in which they make later systems more aligned. He presents this as one possible scenario, not as a date by which runaway improvement must occur.

“Maybe my median expectation is something like four or five years of AI progress in a single year. This requires really overcoming a huge amount of diminishing returns in research and basically doing the equivalent of the progress we would have gotten after a really large compute scale-out.”

Ryan Greenblatt

“I would say that I expect full automation of AI R&D perhaps somewhere around 2031, 2030. Getting to the “beats all humans on the job” milestone, maybe my median expectation is around 2033.”

Ryan Greenblatt

“When I look at AIs right now, it’s already the case that they can pretty competently match humans who are mediocre at ML research at doing ML research. It’s just that being mediocre at ML research is not that helpful.”

Ryan Greenblatt

“I think ML is a very shallow domain relative to math. In math, there was much more of a thing where you find some true deep abstraction, and if you really understand that thing, which is hard to understand, then you get somewhere. Whereas I feel like the things that are the equivalent of that in ML are really dumb bullshit.”

Ryan Greenblatt

“My understanding, based on how algorithmic progress works, is that we’d be able to train a model that’s as good as the best model we had perhaps around three years ago.”

Ryan Greenblatt