Lilian Weng Rejoins OpenAI to Lead Recursive Self-Improvement AI Team
Just two days after leaving her co-founder role at Thinking Machines, former OpenAI safety lead Lilian Weng has rejoined OpenAI. She will lead a new team dedicated to developing recursive self-improvement (RSI) models. This move aligns with OpenAI's long-term strategy to build an autonomous "AI researcher" that can automate the design, training, and evaluation of next-generation models. Weng's leadership in RSI could accelerate the timeline toward superintelligence and highly autonomous AI agents. Weng's approach to RSI, as outlined in her recent blog post, emphasizes engineering the "harness" (scaffolding, workflows, and context management) rather than just rewriting model weights. OpenAI's internal timeline aims for an autonomous AI research intern by September 2026 and a fully automated multi-agent research system by March 2028.
## BACKGROUND
Recursive self-improvement (RSI) is a process where an AI system uses its own capabilities to analyze and improve its own software, potentially triggering an intelligence explosion. Lilian Weng previously spent seven years at OpenAI, contributing to its robotics team and building its Safety Systems team from scratch before briefly co-founding Thinking Machines, which recently launched the Inkling model.