Descriptions:
Matthew Berman sits down with Tibo, a product infrastructure leader at OpenAI who previously worked on language model research at Google DeepMind, to trace how lessons from Google’s internal LM Chat project — shelved roughly a year before ChatGPT launched — shaped the culture and product discipline Tibo now helps maintain at OpenAI.
The conversation covers OpenAI’s bottom-up, high-shipping-velocity culture and the deliberate balance it strikes between radical experimentation and product coherence (citing the ChatGPT iOS app as a benchmark for quality). Tibo also reflects on the competitive divergence between Google DeepMind, which was structured for research rather than product delivery, and OpenAI, where research and product teams co-design from day one.
The latter half of the interview turns to the future of human-AI interaction. Tibo argues that large language models succeeded precisely because natural language is a fundamentally human medium, and that the ideal AI interface should adapt entirely to humans rather than requiring users to adapt to it. He points to rapidly accelerating adoption of ChatGPT’s voice mode as early evidence that users default to the most natural modality available, and envisions a near-future where ambient, multimodal AI — aware of whiteboards, gestures, and conversational context — becomes a seamless extension of everyday human work and communication.
📺 Source: Matthew Berman · Published August 24, 2026
🏷️ Format: Interview
