Recursive AI self-improvement enabling AI to autonomously advance frontier models is expected between 2027 and 2028.

Recursive self-improvement (RSI) is now possible because the agent harness runs on the same medium, code, that the agent produces and can modify at runtime, unlike weight-based training.

“Whereas now the harness as it runs, the agent is producing and writing code and can change its own code as it runs.”

Ryan Greenblatt's median expectation is that once AI R&D is automated, you get four or five years of AI progress in a single year.

“Maybe my median expectation is something like four or five years of AI progress in a single year.”

A light form of AGI (RSI) will be achieved by the end of 2027.

“We're 6 monthsish or towards the end of the year to be completely done with code. Like it's a solved problem and then we'll probably hit some form of, you know, light RSI by end of next year.”

Dario Amodei said that if the AI exponential continues, it cannot possibly be more than a few years before AI is better than humans at essentially everything.

“If the exponential continues, then it cannot possibly be more than a few years before AI is better than humans at essentially everything.”

There is a 50% chance of superintelligence by 2029.

“My sort of median estimate 50% chance is currently in 2029.”

Scaling laws have held for almost 10 orders of magnitude and there is no reason they should not continue to hold.

“It's held for you know almost 10 orders of magnitude but there's no reason it should not keep holding.”

Brian Greene is skeptical that AI-driven recursive self-improvement will arrive by 2029, directly contradicting timelines proposed by Dario Amodei, Sam Altman, and Demis Hassabis.

“I'm skeptical that by 2029 this will really be the case.”

Ryan Greenblatt expects full automation of AI R&D around 2030 or 2031.

“I would say that I expect full automation of AI R&D perhaps somewhere around 2031, 2030.”

A fast-takeoff overnight intelligence explosion is unlikely because models rely so heavily on large-scale test-time compute to achieve peak intelligence, creating a time bottleneck.

“I think fast takeoff is relative things are moving very fast but I think there is this hypothesis that you could have basically an overnight intelligence explosion where the models discover some kind of breakthrough through to make themselves smarter and then that leads to more breakthroughs that make themselves even smarter immediately and you have basically in an instance the models just you know becoming very superhuman across the board in moments and I don't think we're headed to that world largely because of the fact that the models rely so much on large scale test time compute in order to achieve their greatest intelligence.”

Compute scaling will slow down in the late 2020s due to fab capacity constraints.

“Compute scaling will start to slow down probably in the late 20s because just like at some point you use up basically almost there.”

It is not realistic in the short to medium term that large-scale pre-training jobs will be autonomously kicked off by an ML intern; humans remain in the driver's seat due to cost and opportunity cost.

“I find it doesn't seem like super realistic in the short to medium term that you're going to just like be letting you know, large-scale pre-training jobs be kicked off by the ML intern.”