Former Openai CTO Mira Murati has announced her new startup. Naturally, it focuses on AI.
A startup called Thinking Machines Lab, which came out of stealth today, said, “We're going to build tools to work with AI.” [people’s] To create AI systems that are “more widely understood, customizable, and generally competent” than those currently available.
Murati leads Thinking Machine Lab as CEO. Openai co-founder John Schulman is the company's chief scientist and former Chief Research Officer of Openai, Barret Zoph is CTO.
In a blog post shared with TechCrunch, Thinking Machines Lab wrote that while AI features are progressing dramatically, there is still a “key gap.”
“The scientific community's understanding of frontier AI systems is lagging behind its ability to advance rapidly,” reads the blog post. “The knowledge about how these systems are trained is concentrated in top research labs, limiting both the public discourse about AI and the ability to effectively use AI. And Despite this possibility, these systems are still difficult for people to customize to their specific needs and values.”
Thinking Machines Lab plans to focus on building “multimodal” systems “working with people,” according to a blog post, and “adaptively adapts to the full range of human expertise and more. It allows for a wide spectrum.
“[W]E is building models on the frontiers of domain functionality, such as science and programming,” the blog post states. “In the end, the most advanced models unlock the most transformative applications and benefits, including enabling new scientific discoveries and engineering breakthroughs.”
AI safety is another core tenet of the work of Thinking Machines Lab. The company will support external research on alignment by preventing misuse of released models, sharing best practices and recipes on how to build safe AI systems in the industry, and sharing code, datasets and model specifications. He said he plans to contribute to safety. .
“We'll focus on understanding how our systems create authentic value in the real world,” Thinking Machines Lab wrote in a blog post. “The most important breakthrough often comes from optimizing existing metrics as well as rethinking your goals.”
I began thinking about Machine Labs with a remarkable team of scientists, engineers and builders. We are building three things:
– Help people to work with AI systems to meet their specific needs
– Developing a powerful foundation for building more capable AI systems
– Raising…
– Miramurati (@miramurati) February 18, 2025
Murati left the company for six years in October last year. At the time, she said she was away to “exploring herself.”
Murati came to Openai in 2018 as Vice President of AI and Partnerships. After being promoted to CTO in 2022, she led the company's work on Codex, a code generation system featuring CHATGPT, AI DALL-E from Text to Images, and an early version of GitHub's Copilot programming assistant.
Mirati was temporarily interim CEO of the Open after the fire of CEO Sam Altman. Altman described her as a close ally.
For months, rumors have flew over Murati, who will hire prominent AI researchers and staff for AI ventures. The Thinking Machines Lab blog lists 29 employees from Openai, Character AI and Google Deepmind among other top companies.
Thinking Machines Lab is actively hiring machine learning scientists and engineers, as well as research program managers, according to a company's post.
At one point, it was said that Murati was in discussions to raise more than $100 million from an unknown VC company. The blog did not confirm or deny this.
Before Openai, Murati spent three years at Tesla as senior product manager for the model X, the automaker's crossover SUV. Meanwhile, Tesla released an early version of Autopilot, the AI-Evabled driver assistance software. She was also the vice president of products and engineering for Leap Motion, a startup building hand and finger tracking motion sensor for PCs.
Murati is on the growth list of startups that have launched former Openai executives, including rivals such as Ilya Sutskever's Safe Superintelligence and Anthropic.