As a former math student and a PhD in Computational Science, this feels surreal🤯. I remember spending days, weeks, sometimes even months chasing a single proof. Research often felt like solving a giant puzzle, one insight at a time. Now we’re watching our models discover mathematics.

Grateful to have worked on the research and model training for these transcription models with an incredible team. 🚀 As a heavy Codex user, I rely on transcription throughout my day—from coding at my desk to talking to Codex in the car with music playing. What makes it truly useful isn’t just that it works reliably in real-world noisy environments, but that it’s context-aware: understanding domain-specific terminology, colleague names, mixed-language conversations, and my accents. That’s the kind of experience that makes AI feel like it can genuinely fit into everyday life.