OpenAI's former CTO announces new company, poaches numerous ChatGPT core team members

The shoes have dropped—Mira Murati, former CTO of OpenAI, has officially announced her new startup.

The founding team of the new company, Thinking Machines Lab, is exceptionally high-profile, largely composed of former OpenAI members, including but not limited to Barrett Zoph (CTO) and John Schulman (Chief Scientist), both previously rumored to join.

Also appearing in the official lineup is our familiar face, Lilian Weng, who earned her bachelor's degree from Peking University and joined OpenAI in 2017 as a research scientist, contributing to work on GPT-4’s pretraining, reinforcement learning and alignment, and model safety.

Investing is not just about products—it's also about people.

Earlier reports indicated that despite having released no product, Mira’s new venture was already expected to raise over $100 million in initial funding, largely based on reputation alone.

In a post published early this morning on X, Mira explained the founding mission of the new company:

I’ve founded Thinking Machines Lab with an exceptional team of scientists, engineers, and builders. We are doing three things: helping people customize AI systems to their specific needs, building strong foundations for developing more powerful AI systems, and cultivating a culture of open science to help the entire field understand and improve these systems.

Our goal is simple: advance AI by making it widely accessible and easier to understand through solid foundations, open science, and practical applications.

About this team of 29 renowned AI figures, the official website states:

We are scientists, engineers, and builders who have created some of the most widely used AI products, including ChatGPT and Character.ai, open-weight models such as Mistral, and popular open-source projects like PyTorch, OpenAI Gym, Fairseq, and Segment Anything.

Will Thinking Machines Lab become another OpenAI?

According to the official announcement blog, Thinking Machines Lab is an AI research and product company aiming to build a future where everyone has access to the knowledge and tools needed to apply AI to their unique needs and goals.

Despite major breakthroughs in AI technology, the scientific community still has limited understanding of frontier systems, and relevant knowledge remains concentrated within top research labs—this affects public comprehension and use of AI.

Meanwhile, existing AI systems still fall short in personalization and customization.

To address these issues, this group of scientists and engineers—previously involved in developing well-known AI products such as ChatGPT and Character.ai, as well as open-source projects like PyTorch—founded Thinking Machines Lab to enhance AI’s interpretability, customizability, and general usability.

Thinking Machines Lab believes scientific progress depends on open sharing, and plans to collaborate with the research community by publishing technical blogs, papers, and code. These principles all point toward one goal: creating more practical, intelligent, and human-centered AI systems.

Regarding operational philosophy and safety strategy, the blog notes that product deployment drives technological iteration and helps the team focus on solving the most impactful problems.

Thinking Machines Lab will contribute to AI safety in the following ways:

1. Maintaining high safety standards—preventing misuse of released models while maximizing user freedom, 2. Sharing best practices and methodologies for building safe AI systems across the industry, 3. Accelerating external research on alignment by openly sharing code, datasets, and model specifications. Methods developed for current systems, such as effective red-teaming and post-deployment monitoring, are expected to provide valuable insights for future, more powerful systems.

Besides, the name of Mira’s new company, Thinking Machines Lab, carries significant historical weight.

According to Wired magazine, over thirty years ago, American inventor Danny Hillis already envisioned collaborative intelligence between humans and machines.

As a student of AI pioneer Marvin Minsky, Hillis built a supercomputer powered by massively parallel chips—an early precursor to today’s AI computing clusters. It was on this machine that Hillis founded his new company, Thinking Machines.

However, the computer was ahead of its time, and the company ultimately went bankrupt in 1994.

Now, this historically significant name is reborn in a new form. Mira will take up this banner of history and build the next Thinking Machines.