What happened
Thinking Machines Lab, supported by $2 billion in seed funding and staffed with former OpenAI researchers, published its first detailed research on eliminating randomness in AI model responses. The lab identified that nondeterminism in large language models (LLMs) arises from how GPU kernels are orchestrated during inference processing. Their research proposes methods to make AI model outputs consistent and reproducible.
Why it matters
- Consistent AI responses improve reliability for enterprise applications requiring accuracy, such as finance, medicine, and legal review.
- Eliminating randomness enhances reinforcement learning by reducing noise in training data, potentially improving AI model training efficiency.
- The research lays a technical foundation for transforming AI systems from experimental tools into dependable production tools.
Key facts
- 1. Thinking Machines Lab received $2 billion in seed funding and was established in 2023.
- 2. The research was published in September 2025 in a blog post titled "Defeating Nondeterminism in LLM Inference."
- 3. Experiments with a 235-billion-parameter model showed 100% output consistency after 1,000 repeated tests under identical input conditions.
