AI researchers debate how close we are to recursive self-improvement
Dwarkesh Patel sits down with John Schulman of Thinking Machines, Beren Millidge of Zyphra, and Charlie O'Neill of Baseten to steelman the case against rapid recursive self-improvement. They explore why AI progress might asymptote before crossing human-level research, whether long-horizon RL can produce AGI, the persistent sim-to-real gap, and what it would take to discover the next paradigm shift.
You could imagine, as AIs get more and more capable, that they're capable of making progress on simulations which incentivize getting better at not only AI R&D, but at science generally.