Summary of Key Points
Lilian Weng, former Vice President of Research and Security at OpenAI and co-founder of the startup Thinking Machines, left her position due to the overwhelming stress associated with being a co-founder. However, just a day later, she returned to OpenAI to lead a research team focused on "recursive self-improvement." This move is not a change of heart but rather a strategic shift from an entrepreneur who takes on all company responsibilities to an expert dedicated to a specific research area. By doing so, she avoids the relentless pressures of entrepreneurship while continuing to advance in the cutting-edge field of AI. For OpenAI, it represents a reinforcement of a critical direction; for Thinking Machines, it marks another significant loss of core talent.
The Pressure of Being a Co-Founder
Co-founders of startups are not just senior researchers but often act as general managers responsible for everything. In her resignation letter, Weng mentioned that her illness frequency and severity over the past seven months were unprecedented. She had to handle research and product development, as well as team recruitment, fundraising, company strategy, external relations, and even organizational culture. For example, she felt guilty even when asking for time off during a team sprint to rest. Attempting to limit her responsibilities proved difficult because of her personality, which made it hard for her to be less involved. This kind of boundless responsibility and uncertainty is more exhausting than the high-pressure research environment in established companies. It’s like running a small restaurant where you have to cook, handle customer service, manage inventory, and take care of staff—your body simply can’t take it all.
Returning to OpenAI: Not a Backlash, but the Right Role
What Weng left was the role of co-founder, not the AI industry itself. She clearly wanted a job with clear responsibilities and a predictable work environment. OpenAI offered her exactly that: leading an internal research team focused on recursive self-improvement, free from fundraising and recruitment concerns, allowing her to concentrate solely on research. Moreover, she is a seasoned veteran at OpenAI, having contributed to the development of GPT-4’s security system and built a large Safety team from scratch. Her expertise is essential for OpenAI’s progress in this area, and their needs align perfectly with hers.
What is Recursive Self-Improvement?
In simple terms, it involves enabling AI to improve its own training processes, creating a self-enhancing cycle. Previously, humans designed the training data and adjusted model parameters; now, AI can generate more effective training data, optimize model structures, and even propose new research methods. For example, it’s like asking students to create their own questions, correct their homework, and summarize learning strategies—the more they learn, the better they become at learning. Weng has written about this in previous blogs (such as "Harness Engineering for Self-Improvement"), so her expertise makes her perfect for this role.
The Different Situations for the Two Companies
- OpenAI’s Victory: Weng is well-versed in OpenAI’s research system and security mechanisms and understands recursive self-improvement, which will help the company systematize its use of AI to accelerate research efforts and reinforce its technological leadership.
- Thinking Machines’ Losses: Out of the original six co-founders, four have left (Weng and Talock joined Meta, while Zov and Meets returned to OpenAI), and 13 out of the initial 42 founding members have departed, leaving only CEO Murati and Chief Scientist Schulman. This significant loss of talent is a major blow for a startup that has just raised $2 billion in funding.
The Lesson Behind this Story
Weng’s decision highlights several important points:
1. Not Everyone Is Suitable for Entrepreneurship: Especially roles like co-founders, which require comprehensive responsibility. If you excel in a specific field rather than management, a position in a mature company might be more suitable.
2. Health Comes First: No matter how passionate you are about your career, you can’t afford to burn out. Weng emphasized that the future we want to build should put people first, which means taking care of yourself.
3. Returning to a Previous Employer Is Not a Failure: As long as your skills match the company’s needs and the role is suitable, it’s wise to return to a familiar environment where you can continue to contribute.
This “24-hour turnaround” is essentially about an AI expert redefining her role. She hasn’t left the AI industry she loves; she has just found a more comfortable way to make a difference. For us all, it’s a reminder to find the right position for ourselves in our careers and avoid being trapped by the dream of entrepreneurship or high-ranking positions. Health and fit are the keys to long-term success.