SAN FRANCISCO, CALIFORNIA – DECEMBER 03: Mira Murati, Technologist speaks onstage during “New Beginnings: A Conversation with Mira Murati” at WIRED’s The Big Interview 2024 at The Midway SF on December 03, 2024 in San Francisco, California. (Photo by Jon Kopaloff\/Getty Images for WIRED) | Image Credits:Jon Kopaloff\/Getty Images for WIRED <\/p>\n
There\u2019s been great interest in what Mira Murati\u2019s Thinking Machines Lab is building with its $2 billion in seed funding<\/a> and the all-star team of former OpenAI researchers who have joined the lab. In a blog post<\/a> published on Wednesday, Murati\u2019s research lab gave the world its first look into one of its projects: creating AI models with reproducible responses.<\/p>\n The research blog post, titled \u201cDefeating Nondeterminism in LLM Inference,\u201d tries to unpack the root cause of what introduces randomness in AI model responses. For example, ask ChatGPT the same question a few times over, and you\u2019re likely to get a wide range of answers. This has largely been accepted in the AI community as a fact \u2014 today\u2019s AI models are considered to be non-deterministic systems\u2014 but Thinking Machines Lab sees this as a solvable problem.<\/p>\n The post, authored by Thinking Machines Lab researcher Horace He, argues that the root cause of AI models\u2019 randomness is the way GPU kernels \u2014 the small programs that run inside of Nvidia\u2019s computer chips \u2014 are stitched together in inference processing (everything that happens after you press enter in ChatGPT). He suggests that by carefully controlling this layer of orchestration, it\u2019s possible to make AI models more deterministic.<\/p>\n Beyond creating more reliable responses for enterprises and scientists, He notes that getting AI models to generate reproducible responses could also improve reinforcement learning (RL) training. RL is the process of rewarding AI models for correct answers, but if the answers are all slightly different, then the data gets a bit noisy. Creating more consistent AI model responses could make the whole RL process \u201csmoother,\u201d according to He. Thinking Machines Lab has told investors that it plans to use RL to customize AI models for businesses<\/a>, The Information previously reported.<\/p>\n Murati, OpenAI\u2019s former chief technology officer, said in July that Thinking Machines Lab\u2019s first product will be unveiled in the coming months<\/a>, and that it will be \u201cuseful for researchers and startups developing custom models.\u201d It\u2019s still unclear what that product is, or whether it will use techniques from this research to generate more reproducible responses.<\/p>\n Story Continues <\/p>\n