Lilian Weng, who built OpenAI's safety team, is rejoining the lab to lead recursive self-improvement research — the ...