Thinking Machines Lab Full-Duplex AI
Key Questions
What is Thinking Machines Lab?
Thinking Machines Lab is Mira Murati's startup, formerly OpenAI CTO, focusing on advanced AI interaction models. It ships full-duplex AI for real-time voice, video, and text. The lab aims for human-like conversations.
What are the key features of Thinking Machines' interaction models?
They support real-time voice/video/text with 0.4s latency using a 276B MoE model and continuous input/output. This enables full-duplex AI that listens while talking. It beats GPT and Gemini benchmarks.
How does Thinking Machines' AI differ from OpenAI's voice approach?
It argues interactivity is what OpenAI gets wrong about voice, emphasizing near-realtime previews. Full-duplex allows natural convo shifts for agents/products. Demos show fluid, human-like interactions.
What latency does Thinking Machines' AI achieve?
The AI responds in 0.4 seconds, enabling seamless real-time conversations. This low latency supports multimodal inputs like voice and video. It marks a shift from turn-based chat.
What benchmarks does Thinking Machines' model outperform?
It beats GPT and Gemini on relevant benchmarks for interaction quality. The 276B MoE model powers natural, continuous dialogues. This positions it as a leader in full-duplex AI.
Mira Murati's startup ships interaction models for real-time voice/video/text (0.4s latency, 276B MoE, continuous input/output); beats GPT/Gemini benchmarks; natural convo shift for agents/products.