Lilian Weng Exits Thinking Machines, Returns to OpenAI to Lead Recursive Self-Improvement Research

Skynet Chance (+0.08%): OpenAI explicitly staffing a top-level team around recursive self-improvement moves a canonical loss-of-control mechanism from theory into an institutional research program, raising the odds of capability gains outpacing alignment. That the lead is a former VP of AI Safety Research is a partial mitigant but does not offset the direction of the work.

Skynet Date (-1 days): A dedicated effort to have AI accelerate its own research loop is precisely the feedback dynamic that compresses timelines to hard-to-control systems. Concentrating senior talent on this at a leading frontier lab pulls any such scenario earlier.

AGI Progress (+0.04%): Formalizing recursive self-improvement as a top-level research priority at OpenAI signals that automating AI research itself is now treated as a tractable near-term path to AGI rather than a speculative endpoint. The news reveals strategy and staffing rather than a demonstrated capability, which caps its magnitude.

AGI Date (-1 days): If AI systems meaningfully accelerate their own research, progress compounds rather than scaling linearly with human researcher headcount, which would shorten AGI timelines. Talent consolidating from a rival startup back into OpenAI further concentrates capability at the frontier.