Thinking Machines Lab hopes to make AI models more consistent
Source: CoinWorld
Time: 2025-09-11 05:38:17
Thinking Machines Lab, led by Mira Murati, published its first research blog post titled “Updating nondeterminism in LLM reasoning,” exploring the root causes of the randomness of AI models in response. Researcher Horace He attributes this to the orchestration of GPU cores during inference and points out that controlling this layer can enhance determinism. Repeatable responses can improve reinforcement learning and enterprise AI customization. The lab plans to regularly disclose research results through its “Connectionism” blog series. Murati hints about a product for researchers and startups coming soon, but details have not been disclosed. The $12 billion startup is designed to meet core AI challenges while maintaining transparency.