After her sudden departure from OpenAI, word got out that Mira Murati was looking to fund an AI startup of her own. Reports said the former OpenAI CTO (and brief interim CEO) wanted to raise $100 million for the new venture. While we have no idea how much capital Murati raised, we finally know the name of her AI startup: Thinking Machines Lab.
Murati formally introduced Thinking Machines Lab this week when we got confirmation that plenty of former high-ranking OpenAI execs have joined her effort.
Murati will act as the CEO of Thinking Machines Lab, and the team is currently comprised of about 30 researchers, two-thirds of which are former OpenAI engineers.
Here are some of the notable Thinking Machines Lab employees who have been working on ChatGPT before departing OpenAI:
- Mira Murati (CEO) – Former OpenAI CTO
- Barret Zoph (CTO) – Former OpenAI VP of Research
- John Schulman (Chief Scientist) – Co-founder of OpenAI
- Alexander Kirillov – Expert in multimodal Al from OpenAI
- Lilian Weng – Former VP of Research (safety) at OpenAI
- Luke Metz – Former OpenAI researcher
- Jonathan Lachman – Former OpenAI special projects lead
Thinking Machines Lab also hired former engineers from Google, Anthropic, and Mistral, other big names in AI development.
As for what Thinking Machines Lab is building, it’s still a big mystery. We know it’s an AI startup but have no idea when we’ll get the first products from Thinking Machines Lab.
The startup published a mission statement on its website, describing the sort of AI work it’s going to focus on. Thinking Machines Lab says it will develop “AI that works for everyone,” suggesting its priority isn’t autonomous AGI and superintelligence systems:
Instead of focusing solely on making fully autonomous AI systems, we are excited to build multimodal systems that work with people collaboratively.
Thinking Machines Lab also wants to develop AI that can help it in multiple fields, not just programming and mathematics:
We see enormous potential for AI to help in every field of work. 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 firm will put multimodality at the center of its AI products, not that I can imagine anyone creating AIs that don’t support multiple simultaneous inputs, like text, images and audio. Thinking Machines Lab mentions “advanced multimodal capabilities” without explaining the advanced part:
We see multimodality as critical to enabling more natural and efficient communication, preserving more information, better capturing intent, and supporting deeper integration into real-world environments.
AI safety is another big focus for the startup, with Thinking Machines Lab saying it’ll adopt an empirical and iterative approach to AI safety:
The most effective safety measures come from a combination of proactive research and careful real-world testing. We plan to contribute to AI safety by (1) maintaining a high safety bar–preventing misuse of our released models while maximizing users’ freedom, (2) sharing best practices and recipes for how to build safe AI systems with the industry, and (3) accelerating external research on alignment by sharing code, datasets, and model specs.
The startup also said it will develop frontier AI intelligence without detailing what that might be or how it will compete with existing rivals, whether ChatGPT or other products:
Ultimately, the most advanced models will unlock the most transformative applications and benefits, such as enabling novel scientific discoveries and engineering breakthroughs.
Finally, Thinking Machines Lab also notes that it will make infrastructure a “top priority” for its AI products, which will include doing things right from the start with no shortcuts:
Research productivity is paramount and heavily depends on the reliability, efficiency, and ease of use of infrastructure. We aim to build things correctly for the long haul, to maximize both productivity and security, rather than taking shortcuts.
All that sounds exciting, considering that it’s coming from former high-ranking OpenAI execs who helped ChatGPT become what it is today and then left the company. OpenAI losing top talent at a rapid pace has been a problem we can’t ignore, especially since many of the departing engineers were working on AI safety.
That said, it’ll probably be a while before we see the first products from Thinking Machines Lab, and we have no idea how long we’ll have to wait.
