Tinker Demystified: Mira Murati's Single-Line Code Revolution for Customizing Frontier AI Models
In the world of artificial intelligence, a major name is making new waves. Mira Murati, the former CTO of OpenAI, has launched the first product from her new AI startup, Thinking Machines Lab.
The product is called Tinker, and it's a flexible Application Programming Interface (API) designed to simplify the process of fine-tuning AI models.Tinker's main goal is to empower researchers and developers by giving them control over algorithms and data, while the platform itself handles the immense complexity of distributed training. This allows users to run novel experiments and build custom models without getting bogged down in the technical backend. Murati’s lab highlights the API’s flexibility, noting that a user can switch from a lightweight model to a massive mixture-of-experts system simply by changing a single line of Python code.The mission behind Tinker is to enable more people to conduct research on cutting-edge models and tailor them to their specific needs.The platform is currently in a private beta phase, and a waitlist is open for interested researchers and developers.
It has already been utilized by several prominent research groups, including teams at Princeton for training mathematical theorem provers and at Stanford for fine-tuning models on chemistry reasoning.
Berkeley’s SkyRL group has also used it for multi-agent experiments.To further support developers, Thinking Machines Lab is also releasing the Tinker Cookbook, an open-source library featuring modern post-training methods built on the Tinker API. The service will be free to start, with a plan to introduce usage-based pricing in the coming weeks. This launch represents a significant step toward making advanced AI tools more accessible to the broader research community.
·
13:11
|
In the world of artificial intelligence, a major name is making new waves. Mira Murati, the former CTO of OpenAI, has launched the first product from her new AI startup, Thinking Machines Lab.
The product is called Tinker, and it's a flexible Application Programming Interface (API) designed to simplify the process of fine-tuning AI models.Tinker's main goal is to empower researchers and developers by giving them control over algorithms and data, while the platform itself handles the immense complexity of distributed training. This allows users to run novel experiments and build custom models without getting bogged down in the technical backend. Murati’s lab highlights the API’s flexibility, noting that a user can switch from a lightweight model to a massive mixture-of-experts system simply by changing a single line of Python code.The mission behind Tinker is to enable more people to conduct research on cutting-edge models and tailor them to their specific needs.The platform is currently in a private beta phase, and a waitlist is open for interested researchers and developers.
It has already been utilized by several prominent research groups, including teams at Princeton for training mathematical theorem provers and at Stanford for fine-tuning models on chemistry reasoning.
Berkeley’s SkyRL group has also used it for multi-agent experiments.To further support developers, Thinking Machines Lab is also releasing the Tinker Cookbook, an open-source library featuring modern post-training methods built on the Tinker API. The service will be free to start, with a plan to introduce usage-based pricing in the coming weeks. This launch represents a significant step toward making advanced AI tools more accessible to the broader research community.
Subscribe
Listen to The Daily AI Chat using one of many popular podcasting apps or directories.