Thinking Machines wants to build an AI that actually listens while it talks
TL;DR
Every AI model so far follows the same turn-taking pattern: you talk, it listens, then it responds while you wait. Thinking Machines wants to break that loop by building a model that processes your input and generates a response at the same time. The result feels less like a text chain and more like a phone call. If it works, voice assistants and AI calls could become noticeably more natural.
Nauti's Take
Promising direction: Thinking Machines is dropping the rigid turn-taking model and building AI that listens and responds at the same time — voice interfaces and AI calls could feel noticeably more natural. The catch is substance: Murati has shown a concept, not benchmarks or a release date, and latency plus hallucination questions stay wide open.
Anyone building serious voice AI should track the approach, but waiting for real tests before integrating is the smarter call.
Sources