Thinking Machines Lab Aims for Reproducible AI | JQJO

Mira Murati's Thinking Machines Lab, backed by $2 billion in seed funding, aims to create AI models with reproducible responses. Their research focuses on eliminating randomness in AI model outputs, a problem addressed in a recent blog post. The lab believes controlling GPU kernel orchestration during inference processing can achieve more deterministic AI responses, improving reinforcement learning and creating more reliable models for businesses and researchers. Their first product launch is expected in the coming months, and the lab plans to openly share its research findings.