Summary of Key Points
Lilian Weng, a prominent figure in the AI community, recently announced her departure from Thinking Machines Lab, which she co-founded with OpenAI's former CTO. The reason for her resignation is the long-term high-pressure work environment, which has taken a toll on her health. At a time when the company just released its first open-source large model, Inkling, her departure came as a surprise. However, the recurring themes of "health" and "guilt" in her resignation letter highlight the fierce competitive pressures in the AI industry. Thinking Machines Lab, a star startup that received a $2 billion seed round of financing and is valued at $12 billion, is facing challenges such as difficulties in fundraising and the loss of key talents. Weng's departure serves as a reminder that, even in an era where AI models can be continuously improved, humans have physical limitations, and taking care of oneself is of utmost importance.
Detailed Analysis
1. Who is Lilian Weng?
Lilian Weng is no ordinary professional; she is a leading expert in the field of AI security. After graduating from Peking University with a bachelor's degree, she pursued a Ph.D. and joined OpenAI in 2018. She was involved in projects that enabled robots to solve puzzles like Rubik's cubes and later built OpenAI's security team from scratch (some of the security mechanisms behind ChatGPT were developed by her team). Her team has grown to over 80 members by the time she left. Weng is also known for her technical blog, "Lil'Log," where she explains complex AI concepts like reinforcement learning and transformers in an easy-to-understand way, making her articles a valuable reference for many university courses. In short, she is an expert who can conduct cutting-edge research while also effectively communicating complex technology to a wide audience.
2. The Resignation Is Not a Shifting of Responsibility; Her Health Could No Longer Take It
Her resignation letter is filled with apologies. She mentioned that she has been sick more frequently in the past seven months than at any other time in her life. During the company's product development phases, she had to take sick leave, feeling guilty for letting everyone down. Despite trying to reduce her workload, she felt that she couldn't perform well unless she dedicated herself fully. Her perfectionism, combined with the constant pressure, ultimately led to her health breaking down. Her colleagues understand her situation; Mira Murati, OpenAI's former CTO, said, "I'm glad you put your health first," and PyTorch's core developer also encouraged her to rest. It's clear that she wasn't unwilling to continue working; she simply ran out of energy.
3. The Glories and Challenges of Thinking Machines Lab
Thinking Machines Lab started with impressive credentials: it was founded in 2025 and raised a $2 billion seed round within seven months, setting a record for startups at that time, with a valuation of $12 billion. Investors such as NVIDIA and AMD contributed to its success. However, the company faces several challenges:
- Financing Issues: It aimed to raise additional funds at a valuation of $5-6 billion in late 2025 but failed to secure the deal, resulting in its valuation returning to $12 billion.
- Talent Loss: Co-founder Andrew Tulloch was poached by Meta, and CTO Barret Zoph left the company amid public disagreements (with OpenAI disputing his "improper behavior"). Many other researchers have also departed.
- External Doubts: The company faces skepticism regarding its high valuation, slow product releases, and high computational costs, as well as an unproven business model. Despite these issues, it still has significant resources, including a 1-gigawatt computing agreement with NVIDIA and contracts for billions of computing power from Google Cloud. Its recently released Inkling model is considered a milestone, so the market isn't completely pessimistic about its prospects.
4. How Competitive Is the AI Industry?
The competition in the AI industry is incredibly fast: as soon as one model is released, the next round of updates begins, and what's considered standard today can be obsolete tomorrow. Startups must balance product development, computing power acquisition, fundraising, and talent recruitment all at once. The pressure is so intense that even models can be upgraded with more parameters or additional computing power, but humans cannot work indefinitely—lack of sleep and fatigue don't disappear on their own. Weng described her 20 months of entrepreneurship as feeling like "a lifetime" due to the sheer intensity of the demands. Her departure serves as a warning that while AI models can be scaled infinitely, human limitations are real.
5. Leaving Is Not a Failure; Aligning with Oneself Is Crucial
The news headline mentions that "taking care of oneself is the best way to align with AI expectations" in the AI era. Weng used this concept to emphasize the importance of balancing work and health. Her resignation doesn't mean she gives up on AI; she continues to read research and stay updated, but she wants a slower pace and clearer responsibilities. This shows that in a highly competitive environment, acknowledging one's limits and taking time to rest is a sign of courage. Leaving a company doesn't equate to failure; it may actually be the most responsible decision.
Final Conclusion
No matter how popular the AI industry becomes, don't risk your health—models can be restarted, but human well-being cannot. Lilian Weng's departure serves as a reminder to everyone involved in this field: take care of yourself first if you want to go further.