Thinking Machines Releases First Technical Share: Building Real-Time Multimodal Collaborative AI
Thinking Machines (founded by former OpenAI CTO Mira Murati) stated that humans can simultaneously speak, listen, observe, think, and collaborate in conversations, and their designed AI will interact with people in the same way.
The company shared its core technology roadmap, early experimental results, and a video demonstration of the model's real-time operation, highlighting its real-time multimodal interaction capabilities (speech + vision + parallel thinking).
ABAB AI Insight
Thinking Machines' core concept is to break through traditional "question-and-answer" AI and create intelligent agents that truly possess real-time multi-threaded collaboration. Combined with the previously released 400ms ultra-low dialogue latency, this means their model not only "responds quickly" but can also perform visual understanding, contextual reasoning, and proactive collaboration while the user is speaking, significantly approaching the state of natural human communication.
On the capital front, the company's $2 billion seed round (valued at $12 billion) is betting on this direction of "human-like interaction," targeting trillion-dollar scenarios that require high naturalness in voice interaction, such as telephone customer service, sales, medical consultations, and remote collaboration.
Structural judgment: Essentially a technological replacement. Thinking Machines aims to replace traditional single-threaded voice assistants with real-time multimodal AI, evolving the mechanism from "response tool" to "parallel collaboration partner," significantly reducing interaction friction and pushing capital and applications towards the next generation of voice agents with genuine human dialogue perception capabilities.
ABAB News · Cognitive Law
The best AI is not the one that answers the smartest, but the one that listens, sees, and thinks like a human simultaneously.
400ms is not just a parameter, but a watershed for whether it can be considered "human."
Whoever first makes AI a true parallel collaboration partner will take the next generation of human-computer interaction entry.