Just now, Weng Li, an alumnus of Peking University, officially announced her resignation. The best "alignment" in the AI era is to take good care of yourself.

Lilian Weng is leaving her job. 

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

Exactly 20 months have passed since she co-founded the company with former OpenAI CTO Mira Murati. A few days ago, Thinking Machines just released its first open-source large model Inkling, which Weng called "an incredible milestone in the company's history" in her letter. 

The co-founder choosing to leave right after the debut of the first model is more or less unexpected. 

A resignation letter filled with "apologies" 

A recurring word in Lilian Weng's resignation letter is "health".

She mentioned that in the past 7 months, she has fallen ill more times than in any other period of her life. While the entire company was sprinting for the release of Inkling, she had to take sick leave to rest, while bearing the guilt of being unable to participate in the work. 

She described that the state of resting while fighting against the sense of guilt was not pleasant at all. 

Weng also considered scaling back her responsibilities and taking on a role with less pressure and uncertainty. But she eventually found that such an arrangement did not suit her. According to her, as long as she cannot put all her energy into the work she is responsible for, she will feel that she is not doing the job well.

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

Weng, who once built the OpenAI safety team from scratch and led a team of more than 80 people, was most worried about letting her colleagues down when saying goodbye. The repeated apologies also reflect her guilt and reluctance to leave. 

After the resignation letter was released, many former colleagues and peers came to leave messages. 

Mira Murati said that the experience of co-founding Thinking Machines was very precious, and she was glad that Weng was willing to put her health first. Soumith Chintala, Saurabh Garg and other core PyTorch developers also sent their well wishes, and most of the messages convey the same meaning: Health matters more, take a good rest.

Her departure is restrained, and her farewell is sufficiently decent. 

Many people may not be familiar with Lilian Weng's experience, so it is necessary to briefly introduce her position in the AI industry. 

Weng was born in Shaanxi Province. She won the first prize of the Shaanxi Division of the National High School Mathematics Competition in 2004, and was admitted to the School of Information Management and Information System of Peking University in 2005, where she received the National Scholarship during her undergraduate years. 

After that, she went to Indiana University Bloomington to pursue her doctorate, with her research focus mainly on network science, complex systems and information dissemination in social networks, which is not exactly the same as the large model research that is more well-known today. 

In 2018, Lilian Weng joined OpenAI. The first important project she participated in was the robotic hand Dactyl. The research team spent two years getting the robotic hand to learn to solve a Rubik's cube with one hand, and Weng was one of the core contributors to the project. 

After GPT-4 was released, she began to build 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 by ChatGPT today are also related to the work she promoted during her tenure. 

In August 2024, Weng 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 has another widely known identity: the author of the tech blog Lil'Log.

Lil'Log has long systematically sorted out cutting-edge technologies such as reinforcement learning, Transformer, and Agent. It is one of the most widely circulated personal tech blogs in the AI field, and some of its articles are also listed as reference materials for university courses. 

She once wrote that explaining a concept clearly is the best way to test whether you really understand it. To some extent, this sentence also sums up her influence in the AI community. 

Putting Weng's departure into the development trajectory of Thinking Machines over the past year, the context will appear more complex. 

Under high valuation, both products and teams are under pressure 

Thinking Machines has attracted much attention since its establishment. 

In July 2025, the company completed a seed round of financing of about 2 billion US dollars, setting a record for the seed round financing scale of startups at that time. This round was led by a16z, with participation from institutions such as NVIDIA, AMD, and Jane Street, and the post-money valuation reached 12 billion US dollars. 

At that time, the company had not released any official products. There were rumors that both Apple and Meta had expressed interest in investing in or acquiring the company, but Murati rejected them.

The high starting point brought extremely high expectations, but the subsequent product progress and personnel changes did not go as smoothly. 

In November 2025, Thinking Machines tried to launch a new round of financing with a valuation of 50 to 60 billion US dollars. By January 2026, the relevant negotiations failed to be completed, and the market valuation returned to about 12 billion US dollars. 

The core team has also experienced 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 the company and returned to OpenAI. 

There was even a public disagreement between the two sides over Zoph's departure. Murati said he was fired for "misconduct", and OpenAI later responded that it did not agree with Thinking Machines' account of the incident. After that, many more researchers left the company one after another. 

External doubts about Thinking Machines mainly focus on several aspects: overvaluation, slow product release, huge computing power costs, and unproven business models. The continuous turnover of core talents has further amplified the market's sense of uncertainty.

However, the company still has considerable resources in its hands. 

Thinking Machines has reached a multi-year strategic cooperation with NVIDIA, planning to deploy next-generation Vera Rubin systems with a scale of at least 1 gigawatt starting from 2027; the company has also signed a multi-billion-dollar computing power agreement with Google Cloud. 

In terms of products, the model fine-tuning tool Tinker has been launched, and the first open-source large model Inkling was also recently released. The company has finally begun to shift its focus from financing and talent stories to real product delivery. 

As a result, the market's judgment on Thinking Machines has not fully turned pessimistic. It faces many problems, and its capital, talent and computing power resources also far exceed those of most AI startups. 

Humans cannot scale infinitely like models 

Lilian Weng's resignation letter did not talk about valuation, financing or internal personnel changes. The whole letter revolves around one reason: the long-term pressure has exceeded the limit her body can withstand.

She said she still loves AI, is still reading papers and tracking research progress. Going forward, she hopes to find an environment with a more predictable pace and clearer boundaries of responsibilities, so that she no longer has to bear the continuous pressure and uncertainty that co-founders face.

Since the beginning of this year, competition in the AI industry has been so fierce that it is calculated by "days". 

As soon as a model is released, the countdown to the next update begins; as soon as a capability becomes the industry standard, a new technical route will soon emerge. Startups also have to catch up with products, computing power, financing and talents at the same time. Everyone is worried that if they slow down one step, they will be left behind by the next wave. 

In the hottest track, speed is no longer just a competitive advantage, but has gradually become an inescapable pressure.

Models can add parameters, expand context, and run continuously on more computing power. But humans have their own limits. Sleep cannot be compressed, the body cannot skip recovery, and long-term accumulated fatigue will not disappear automatically just because a product is released. 

20 months is not a long time, but Weng wrote in her letter that this experience "feels like a lifetime". 

Perhaps earlier than many others, she realized that being at the forefront of the AI wave means participating in shaping an era, and also means bearing the fastest and most intense pressure of this 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 rise, and positions on the leaderboard will keep changing. 

For Lilian Weng, what is more important right now is to allow herself to stop.

Leaving does not always mean failure. Sometimes, it also takes a lot of courage to admit that your body can no longer hold on, put down the unfinished responsibilities, and take good care of yourself.

Reference links are attached:

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

https://lilianweng.github.io/

This article is from the WeChat official account "APPSO", author: The one releasing tomorrow's products, published with authorization from 36Kr.