Lilian Weng Announces Departure from Thinking Machines Lab

Weng Li has left the company.

Just now, Peking University alumna Lilian Weng posted a resignation letter on social media, announcing that tomorrow will be her last day at Thinking Machines Lab.

It has been exactly 20 months since she co-founded the company with Mira Murati, former CTO of OpenAI. Just a few days ago, Thinking Machines released its first open-source large model, Inkling, which Weng Li described in her letter as “a remarkable milestone in the company’s history.”

It was somewhat surprising that the co-founder chose to step away just as the first model was launched.

In Weng Li's resignation letter, the word that appears repeatedly is "health."

She mentioned that over the past seven months, she has been sick more times than at any other point in her life. While the entire company was racing to launch Inkling, she had to take sick leave, all while grappling with guilt over not being able to contribute.

She described the state of resting while battling guilt as unpleasant.

Weng Li also considered narrowing her responsibilities to take on a role with less pressure and uncertainty. But she ultimately realized that this arrangement wasn’t right for her. According to her, as long as she couldn’t devote her full energy to what she was responsible for, she felt she wasn’t doing her job well.

Throughout the letter, expressions such as "sorry" and "letting you down" appear multiple times.

Weng Li, who once built OpenAI’s safety team from scratch and led a team of over 80 people, was most concerned about letting down her colleagues as she departed. Her repeated expressions of apology reveal her guilt and reluctance to leave.

After the resignation letter was posted, many former colleagues and industry peers came by to leave comments.

Mira Murati said that the experience of co-founding Thinking Machines was invaluable, and she was glad that Weng Li prioritized her health. PyTorch core developers such as Soumith Chintala and Saurabh Garg also extended their wishes, with most comments conveying the same message: health comes first, take good care of yourself.

Leave with restraint, and say goodbye with dignity.

Many may not be familiar with Weng Li’s background, so it’s worth briefly introducing her role in the AI industry.

Weng Li was born in Shaanxi, won first prize in the Shaanxi province of the National High School Mathematics Competition in 2004, and entered Peking University in 2005 to study Information Management and Information Systems, during which she received the National Scholarship.

Thereafter, she pursued her doctorate at Indiana University Bloomington, focusing primarily on information diffusion in network science, complex systems, and social networks—a direction distinct from the large model research more widely known today.

In 2018, Weng Li joined OpenAI. Her first major project was the robotic hand Dactyl. The research team spent two years teaching the robotic hand to solve a Rubik’s Cube using a single hand, and Weng Li was one of the key contributors to the project.

After the release of GPT-4, she began building the Safety Systems team from scratch. By the time she left OpenAI, the team had grown to more than 80 people. Some of the safety mechanisms and evaluation systems used in ChatGPT today are related to the work she championed during her tenure.

In August 2024,翁荔 was promoted to Vice President of Research and Safety at OpenAI. Three months later, she left OpenAI. In February 2025, she co-founded Thinking Machines Lab with Mira Murati and others.

In the broader tech community, Weng Li is also widely known as the author of the technical blog Lil'Log.

Lil'Log has systematically organized and explored cutting-edge technologies such as reinforcement learning, Transformers, and agents, and is one of the most widely read personal technical blogs in the AI field, with some of its articles included as reference materials in university courses.

She once wrote that clearly explaining a concept is the best way to test whether you truly understand it. In a sense, this statement encapsulates her influence in the AI community.

Viewing Weng Li's departure within the context of Thinking Machines' trajectory over the past year reveals a more complex backdrop.

Thinking Machines attracted significant attention from its inception.

In July 2025, the company completed a $2 billion seed round, setting a record for the largest seed funding round for a startup at the time. The round was led by a16z, with participation from NVIDIA, AMD, Jane Street, and others, resulting in a post-money valuation of $12 billion.

At the time, the company had not yet released any official products. Rumors suggested that both Apple and Meta had expressed interest in investing or acquiring the company, but Murati declined.

A strong start led to very high expectations, but subsequent product progress and personnel changes did not proceed as smoothly.

In November 2025, Thinking Machines attempted to launch a new funding round at a valuation of $50 billion to $60 billion. By January 2026, the negotiations had not been completed, and the market valuation had reverted to approximately $12 billion.

The core team has also undergone continuous changes.

Co-founder Andrew Tulloch was recruited by Meta. In January 2026, co-founder and CTO Barret Zoph and co-founder Luke Metz left and returned to OpenAI.

Regarding Zoph’s departure, the two parties also publicly disagreed. Murati stated that he was fired for “misconduct,” and OpenAI subsequently responded that it did not agree with Thinking Machines’ account of the matter. Since then, several other researchers have left the company.

External skepticism toward Thinking Machines centers on several key issues: an inflated valuation, slow product releases, substantial computing costs, and an unproven business model. Ongoing turnover among core talent has further intensified market uncertainty.

However, the company still holds substantial resources.

Thinking Machines has entered into a multi-year strategic partnership with NVIDIA, planning to deploy at least one gigawatt of next-generation Vera Rubin systems starting in 2027; the company has also signed a multi-billion-dollar computing agreement with Google Cloud.

On the product side, the model fine-tuning tool Tinker has launched, and the company’s first open-source large model, Inkling, was recently released. The company is finally shifting from fundraising and talent narratives to actual product delivery.

The market has not yet fully turned pessimistic toward Thinking Machines. It faces many challenges, but its resources in funding, talent, and computing power far exceed those of most AI startups.

Weng Li's resignation letter did not mention valuation, funding, or internal personnel changes. The entire letter focused on one reason: long-term stress had exceeded what her body could endure.

She said she still loves AI and continues to read papers and follow research advancements. Next, she hopes to find an environment with a more predictable rhythm and clearer role boundaries, free from the ongoing pressure and uncertainty faced by co-founders.

This year, competition in the AI industry has accelerated to the point of being measured in days.

A model has just been released, and the countdown to the next update has already begun; a capability has barely become an industry standard when a new technological pathway emerges. Startups must simultaneously keep pace with product development, computing power, funding, and talent—everyone fears that falling behind even one step could leave them stranded by the next wave.

On the hottest赛道, speed is no longer just a competitive advantage—it's increasingly becoming an inescapable pressure.

Models can increase parameters, expand context, and continue running with more computational power. Humans, however, have their own limits. Sleep cannot be compressed, the body cannot skip recovery, and accumulated fatigue over time does not vanish automatically with a single product launch.

Twenty months isn’t long, but Weng Li wrote in her letter that the experience “felt like a lifetime.”

She may have realized earlier than many others that standing at the forefront of the AI wave means participating in shaping an era—and bearing the fastest, most intense pressures of that era.

Inkling has become a milestone for Thinking Machines, and the AI industry will continue to move forward. New models will be released, new companies will emerge, and rankings will keep changing.

For Weng Li, what’s more important right now is allowing herself to stop.

Leaving doesn't always mean failure. Sometimes, it takes great courage to acknowledge that your body can no longer endure, to let go of unfinished responsibilities, and to take proper care of yourself.

Reference address attached:

https://x.com/lilianweng/status/2081816923088814421

https://lilianweng.github.io/

This article is from the WeChat official account "APPSO," authored by: Publishing Tomorrow's Product