The former CTO of OpenAI presented a prototype of an AI that is always "present".
Summary
Thinking Machines, founded by former OpenAI CTO Mira Murati, unveiled the Interaction Model, a prototype the company describes as an "always-present" multimodal AI designed for face-to-face style interaction. The team frames presence around three communication conditions—copresence, contemporality, and simultaneity—and argues current systems meet at best the third condition. The Interaction Model uses encoder-free early fusion, unified multimodal training (audio dMel embeddings, hMLP for video, and text embeddings) and a 200ms micro-turn internal clock so the model continuously processes and generates tokens to maintain real-time presence. The article contrasts this approach with OpenAI's GPT-Realtime-2 and ByteDance's Seeduplex, noting those improve simultaneity but still lack full copresence and contemporality.