Trending on X: Thinking Machines Unveils Full-Duplex Interaction Models
What Happened
Thinking Machines published its first product research: native "interaction models" that handle continuous audio, video, and text streams in parallel rather than waiting for a user turn. The implementation runs on SGLang and is positioned as a foundation for agents, tutors, and assistants that behave more like coworkers than chatbots. The post drew the largest X engagement of the day.
My Take
This is the first credible attack on the chat-box paradigm since ChatGPT shipped. The reason every AI product feels the same is that every team copied the same input box. If interaction models work, the differentiator stops being model IQ and starts being product surface — latency, interruption handling, whether the thing can shut up when you start typing. Most "AI features" being shipped right now will look as dated as a 2010 IVR phone tree inside eighteen months.
Read Original Source