Here's the question keeping every AI researcher up at night: are we actually close to the moment where AI can improve itself, with zero human intervention?
In a new episode, John Schulman, Beren Millidge, and Charlie O'Neill sat down to talk honestly about what's really happening at the frontier of AI research.
The concept is called Recursive Self-Improvement, meaning a model that can enhance its own training process, or even design a smarter version of itself, without you stepping in.
Here's an analogy: think of a snowball rolling down a mountain. At first it's slow and easy to stop, but once it starts growing and rolling on its own, its speed outpaces your ability to control it.
That's exactly the philosophical trade-off here. If this moment happens, progress accelerates dramatically, which is fantastic news for innovation, but human oversight and predictability drop sharply at the same time.
There's no clear answer yet, and even the researchers disagree on the timeline.
So tell me, do you think we're two years away or twenty years away from this moment? And how do you think it changes what your job as an engineer looks like in the coming years?
🔗 Original Source & Reference: https://www.dwarkesh.com/p/john-beren-charlie
Published automatically via FeedMind AI Content Pipeline.
Top comments (0)