Last September Mira Murati unexpectedly left her job as chief technology officer of OpenAI, saying, “I want to create the time and space to do my own exploration.” The rumor in Silicon Valley was that she was stepping down to start her own company. Today she announced that indeed she is the CEO of a new public benefit corporation, called Thinking Machines Lab. Its mission is to develop top-notch AI with an eye towards making it useful and accessible.
Murati believes there’s a serious gap between rapidly advancing AI and the public’s understanding of the technology. Even sophisticated scientists don’t have a firm grasp on AI’s capabilities and limitations. Thinking Machines Lab plans to fill that gap by building in accessibility from the start. It also promises to share its work by publishing technical notes, papers and actual code.
Underpinning this strategy is Murati’s belief that we are still in the early stages of AI, and the competition is far from closed. Though it occurred after Murati began planning her lab, the emergence of DeepSeek—which claimed to build advanced reasoning models for a fraction of the usual cost—vindicates her thinking that newcomers can compete with more efficient models.
Thinking Machines Lab will, however, compete on the high end of large language models. “Ultimately the most advanced models will unlock the most transformative applications and benefits, such as enabling novel scientific discoveries and engineering breakthroughs,” the company writes in a blog post on Tuesday. Though the term “AGI” isn’t used, Thinking Machines Lab believes that upscaling the capabilities of its models to the highest level is important to filling the gap it’s identified. Building those models, even with the efficiencies of the DeepSeek era, will be costly. Though Thinking Machines Lab hasn’t shared its funding partners yet, it’s confident that it will raise the necessary millions.
Murati’s pitch has attracted an impressive team of researchers and scientists, many of whom have OpenAI on their resumes. Those include former VP of research Barret Zoph (who is now CTO at Thinking Machines Lab), multimodal research head Alexander Kirillov, head of special projects John Lachman, and top researcher Luke Metz, who left Open AI several months earlier. The lab’s chief scientist will be John Schulman, a key ChatGPT inventor who left OpenAI for Anthropic only last summer. Others come from competitors like Google and Mistral AI.
The team moved into an office in San Francisco late last year and has already started work on a number of projects. Though it’s not clear what its products will look like, Thinking Machines Lab indicates that they won’t be copycats of ChatGPT or Claude, but AI models that optimize collaboration between humans and AI— which Murati sees as the current bottleneck in the field.
American inventor Danny Hillis dreamed of this partnership between people and machines over 30 years ago. A protege of AI pioneer Marvin Minsky, Hillis built a super computer with powerful chips running in parallel—a forerunner to the clusters that run AI today. He called it Thinking Machines. Ahead of its time, Thinking Machines declared bankruptcy in 1994. Now a variation of its name, and perhaps its legacy, belongs to Murati.
