Predictable AI: Thinking Machines Lab Is Tackling Model Consistency

At IGF, we closely follow the innovations shaping the future of artificial intelligence and its connection to startups. One interesting and intriguing startup is Thinking Machines Lab, founded by Mira Murati. As Mira Murati was the former Chief Technology Officer of OpenAI, her expertise allowed her to create a solution that can change the way we see AI and interact with it. The startup is powered by $2 billion in seed funding, which allows it to develop more. With a team of former OpenAI researchers, the lab decided to solve one difficult issue connected to AI. They wanted to create an AI model that answers consistently.

People who actively interact with AI know that asking the same question twice rarely gives you the same answer. Such AI behavior was accepted as normal for a long time, but Thinking Machines Lab believes this limitation can be overcome. They believe that overcoming that can give people new opportunities. This solution is even more important for startups, researchers, and investors who are in constant search for modern solutions.

That’s why in this article, we’re going to uncover the topic of the Thinking Machines Lab solution and its impact.

Why Reproducible AI Matters

One interesting and important fact that influences the work of AI and its consistency is GPU kernels, in other words, small programs running on Nvidia chips. In the recent post, Thinking Machines Lab conducted research and found out that the problem is in how these GPU kernels are orchestrated during inference. So, the more control over this layer, the more precise and stable the responses.

Such an approach could influence the solutions companies deliver with the help of AI. Reproducible outputs give the possibility to build a simple and easy workflow with the help of AI-powered solutions. This step also influences customer experience and better the quality across applications.

It also has big implications for reinforcement learning (RL). When model responses vary, the feedback used for training becomes noisy, making the process slower and less efficient. More deterministic responses could make RL smoother and more powerful — helping startups fine-tune models faster for their specific use cases.

A Transparent Approach to Frontier AI

Mira Murati has recently announced that Thinking Machines Lab is going to release its first product in several months. Even though there are no details about the solution, reproducibility will be the top theme here.

We’re building multimodal AI that works with how you naturally interact with the world – through conversation, through sight, through the messy way we collaborate. We’re excited that in the next couple of months we’ll be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom models. Soon, we’ll also share our best science to help the research community better understand frontier AI systems,” she posted.

At IGF, we support startups that change the industry and create fresh and modern solutions that can make existing solutions. For us, such solutions are about a new era in AI startups. If you want to find an investor or vice versa, contact us today to bring your ideas to life.