Mira Murati, the former CTO of OpenAI, has announced the establishment of a new artificial intelligence startup called Mind Machines Lab. The company will focus on developing multimodal models capable of processing various types of data.
Four months after Reuters first reported on the project, Mind Machines Lab is entering an increasingly competitive AI market. According to earlier reports, Murati was in discussions to raise over $100 million for her new venture. In its launch blog post, the company did not address these funding rumors but confirmed another previously reported detail: the initial team includes Barret Zoph, a former head of research execution at OpenAI.
Murati and Zoph left OpenAI in September of last year, shortly followed by John Schulman, one of the co-founders of ChatGPT, who also resigned. Murati now serves as CEO of Mind Machines Lab, Zoph as CTO, and Schulman as Chief Research Officer.
The startup aims to train multimodal models that can handle text and multimedia files like images. As CTO of OpenAI, Murati oversaw the development of ChatGPT and DALL-E, a series of image generation models. She also played a pivotal role in facilitating OpenAI’s partnership with Microsoft, which provided significant cloud infrastructure support for AI research.
In its launch blog post, the company emphasized "infrastructure quality as a top priority." It highlighted the importance of research productivity, which heavily depends on reliable, efficient, and user-friendly infrastructure. The company aims to build things correctly for long-term development.
The company also shared additional details about its development roadmap. Unlike OpenAI's o1, which focuses on programming and math tasks, Mind Machines Lab’s models will be designed to adapt to the full spectrum of human expertise and support broader applications. It remains unclear whether these applications will include consumer use cases.
Customizability will be another key focus of Mind Machines Lab’s engineering efforts. The company plans to allow customers to easily tailor AI models according to their specific needs. Currently, customization is achieved through prompts and fine-tuning neural networks on proprietary training datasets.
Reports from last year suggested that Mind Machines Lab’s products would be based on proprietary models. However, the company appears to plan on open-sourcing at least some components of its AI technology stack.
"We believe that collaborating with a broader community of researchers and builders will most effectively advance humanity's understanding of AI," the company stated on its website. "We intend to regularly publish technical blog posts, papers, and code."
The open-source initiative will cover areas such as AI safety. The company plans to publicly release code and other technical assets to prevent harmful AI outputs. It intends to test its algorithms' security using established methods, including red-teaming, which involves simulating cyberattacks to identify vulnerabilities.
Mind Machines Lab currently employs 29 people. Besides former OpenAI employees, the founding team includes researchers from Google LLC, Meta Platforms Inc., Mistral, and other major players in the AI ecosystem.
