TLDR: Former OpenAI CTO Mira Murati has launched ‘Thinking Machines Lab,’ an AI startup focused on developing AI systems aligned with human values and capable of diverse applications. The company has attracted significant talent from OpenAI and other tech giants, and is actively seeking over $100 million in funding. Their first product, ‘Tinker,’ an API designed to simplify the fine-tuning of AI models, has recently been introduced in private beta.
Mira Murati, the former Chief Technology Officer of OpenAI, has embarked on a new venture, establishing an artificial intelligence startup named ‘Thinking Machines Lab.’ This new entrant into the competitive AI landscape aims to create advanced AI systems that are deeply integrated with human values and capable of a wider array of applications beyond conventional programming and mathematical tasks. The initiative has garnered attention, with reports indicating active fundraising efforts targeting over $100 million in venture capital.
The ‘Thinking Machines Lab’ has successfully attracted top-tier talent, with approximately two-thirds of its team comprising former OpenAI employees. Notable hires include Barret Zoph, a prominent AI researcher who now serves as the startup’s Technology Chief, and John Schulman, an OpenAI co-founder who joined as Chief Scientist. Schulman’s move to focus on ‘AI alignment’ underscores a core mission of Murati’s new company: encoding human values into AI models to enhance their safety and reliability.
In a statement, the startup emphasized its unique approach, noting, ‘While current systems excel at programming and mathematics, we’re building AI that can adapt to the full spectrum of human expertise and enable a broader spectrum of applications.’ The company is also committed to fostering open research on AI alignment by sharing its code, datasets, and model specifications.
Recently, ‘Thinking Machines Lab’ unveiled its inaugural product, ‘Tinker,’ an Application Programming Interface (API) designed to streamline the fine-tuning of both large and small open-weight AI models. This managed service aims to empower researchers and developers by abstracting the complexities of distributed training, thereby facilitating novel experiments and the creation of custom models with minimal code adjustments.
Mira Murati announced the launch of Tinker on X, stating, ‘Today we launched Tinker. Tinker brings frontier tools to researchers, offering clean abstractions for writing experiments and training pipelines while handling distributed training complexity. It enables novel research, custom models, and solid baselines. Excited to see what people build.’ Tinker is currently in private beta, with a waitlist available for interested researchers and developers. The service will initially be free, with usage-based pricing to be implemented in the coming weeks.
Also Read:
- Former OpenAI CTO Mira Murati Unveils AI Startup ‘Thinking Machines Lab’ and Tinker Platform, Securing $2 Billion in Seed Funding
- Reflection AI Secures $2 Billion Series B Funding, Valued at $8 Billion, to Lead Open Frontier AI Development
Several research groups have already piloted the Tinker platform, demonstrating its versatility. Princeton’s Goedel Team utilized it to train mathematical theorem provers, Stanford’s Rotskoff Chemistry group fine-tuned models for chemistry reasoning, Berkeley’s SkyRL group conducted multi-agent reinforcement learning experiments, and Redwood Research applied Tinker to complex AI control tasks.
