There is significant interest in what Mira Murati’s Thinking Machines Lab is building with its two billion dollars in seed funding and the all-star team of former OpenAI researchers who have joined the lab. In a blog post published on Wednesday, Murati’s research lab provided the world with its first look into one of its projects, which focuses on creating AI models with reproducible responses.
The research blog post, titled Defeating Nondeterminism in LLM Inference, attempts to unpack the root cause of what introduces randomness in AI model responses. For instance, if you ask ChatGPT the same question multiple times, you are likely to get a wide range of answers. This has largely been accepted in the AI community as a fact, with today’s AI models considered non-deterministic systems. However, Thinking Machines Lab sees this as a solvable problem.
The post, authored by Thinking Machines Lab researcher Horace He, argues that the root cause of AI model randomness is the way GPU kernels are stitched together during inference processing. GPU kernels are the small programs that run inside of Nvidia’s computer chips. He suggests that by carefully controlling this layer of orchestration, it is possible to make AI models more deterministic.
Beyond creating more reliable responses for enterprises and scientists, He notes that achieving reproducible AI model responses could also improve reinforcement learning training. Reinforcement learning is the process of rewarding AI models for correct answers, but if the answers are all slightly different, the data becomes noisy. Creating more consistent responses could make the entire reinforcement learning process smoother. Thinking Machines Lab has told investors it plans to use reinforcement learning to customize AI models for businesses.
Murati, OpenAI’s former chief technology officer, said in July that Thinking Machines Lab’s first product will be unveiled in the coming months and that it will be useful for researchers and startups developing custom models. It remains unclear what that product is or whether it will use techniques from this research to generate more reproducible responses.
Thinking Machines Lab has also stated that it plans to frequently publish blog posts, code, and other information about its research to benefit the public and improve its own research culture. This post is the first in the company’s new blog series called Connectionism and appears to be part of that effort. OpenAI also made a commitment to open research when it was founded but has become more closed off as it has grown larger. It remains to be seen if Murati’s research lab will stay true to its claim.
The research blog offers a rare glimpse inside one of Silicon Valley’s most secretive AI startups. While it does not exactly reveal where the technology is going, it indicates that Thinking Machines Lab is tackling some of the largest questions on the frontier of AI research. The real test is whether the lab can solve these problems and create products from its research to justify its twelve billion dollar valuation.