Lilian Weng posts 'hard and sad decision' note about Thinking Machines

Lilian Weng, co-founder of Thinking Machines Lab, posted a cryptic message on July 27 calling her decision 'hard and sad,' without specifying its nature, following the July 15 release of the company's first general-purpose model, Inkling. Weng's role at the AI startup, valued at $12 billion after a $2 billion seed round led by Andreessen Horowitz, remains unconfirmed as the company has seen departures of co-founders Barret Zoph and Luke Metz.

Lilian Weng, a co-founder of Thinking Machines Lab https://thinkingmachines.ai/?ref=runtimewire , posted on July 27 that she had made a "hard and sad decision" and had shared a message with colleagues at Thinking Machines, which she called Thinky. Lilian Weng on X https://x.com/lilianweng/status/2081816923088814421?ref=runtimewire The public text does not identify the decision. Its image attachment could not be independently verified, leaving Weng's current role and status at Thinking Machines unconfirmed. Her post also repeated the line "The future worth building is human," the title of a Thinking Machines manifesto https://thinkingmachines.ai/blog/the-future-worth-building-is-human/?ref=runtimewire published July 10. A public note after Inkling's release Weng's post followed Thinking Machines' July 15 release of Inkling https://thinkingmachines.ai/news/introducing-inkling/?ref=runtimewire , its first general-purpose model. Inkling accepts text, images and audio and generates text. Thinking Machines released the model with open weights https://huggingface.co/thinkingmachines/inkling?ref=runtimewire , allowing outside researchers and developers to fine-tune and integrate it. Thinking Machines says Inkling has 975 billion total parameters, 41 billion active parameters and a context window of up to 1 million tokens. The model card https://thinkingmachines.ai/model-card/inkling/?ref=runtimewire lists an Apache 2.0 license. Thinking Machines says stronger overall models are available and presents Inkling as a customizable base model with multimodal capabilities and variable inference effort. Inkling followed Tinker https://thinkingmachines.ai/tinker/?ref=runtimewire , the managed fine-tuning service Thinking Machines introduced on October 1, 2025. Tinker handles training infrastructure for developers and researchers fine-tuning open-weight models with their own data and algorithms. Together, Tinker and Inkling give Thinking Machines both a base model and the infrastructure customers can use to modify it. The researcher behind Lil'Log Before Thinking Machines, Weng worked at OpenAI in applied AI research, robotics and safety. WIRED has described her as a former OpenAI vice president who led work in safety and robotics. Weng is also the author of Lil'Log https://lilianweng.github.io/?ref=runtimewire , a technical blog focused on machine learning and AI systems, making Weng a recognizable voice outside the labs where she has worked. Weng was part of the founding roster that Thinking Machines assembled around founder and CEO Mira Murati. In July 2025, Thinking Machines closed a $2 billion seed round led by Andreessen Horowitz, with Nvidia, Accel, Cisco, AMD, ServiceNow and Jane Street among the participants. The deal valued Thinking Machines at a reported $12 billion, according to TechCrunch https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/?ref=runtimewire and WIRED https://www.wired.com/story/thinking-machines-lab-mira-murati-funding/?ref=runtimewire . Thinking Machines had yet to release Tinker or Inkling when the round closed. Investors were backing Murati, Weng, John Schulman and a roster of researchers who had built products and research programs at OpenAI and other AI organizations. Leadership questions at a heavily financed lab Thinking Machines has experienced turnover among its early technical leaders. WIRED reported https://www.wired.com/story/thinking-machines-lab-cofounders-leave-for-openai/?ref=runtimewire in January 2026 that co-founders Barret Zoph and Luke Metz were leaving to rejoin OpenAI. Andrew Tulloch separately left for Meta, Reuters reported via Yahoo Finance https://finance.yahoo.com/news/thinking-machines-lab-co-founder-192739653.html?ref=runtimewire . Thinking Machines has also increased its computing commitments. Axios reported https://www.axios.com/2026/03/10/nvidia-thinking-machines-mira-murati?ref=runtimewire on March 10 that Thinking Machines committed to a gigawatt of Nvidia compute, alongside an undisclosed strategic investment from Nvidia. Weng's closing line connects her personal note to Thinking Machines' stated mission. Thinking Machines' manifesto argues that AI should extend human will and judgment and remain shaped by the people using it. The readable portion of Weng's post does not explain how that principle relates to her decision. Until the attachment or another primary source is available in readable form, Weng's employment status and the substance of the decision she referenced remain unverified.