Mira Murati’s Thinking Machines Solves the Mystery of LLM Nondeterminism

Mira Murati’s new company, Thinking Machines, solved a major AI problem: nondeterministic LLM outputs. By creating batch-invariant kernels for key operations, they ensure identical responses at temperature zero, improving reproducibility for research, audits, and safety-critical AI. This breakthrough may redefine industry standards for deterministic, trustworthy AI.