[DigitalToday reporter Hwang Chi-gyu] Thinking Machines, an AI startup founded by former OpenAI CTO Mira Murati (미라 무라티), is highlighting differentiation from other large language models, with a human-centered AI interaction model.

Thinking Machines CEO Mira Murati recently disclosed details on the company’s direction in an interview with Bloomberg.

She said existing AI models have operated in a so-called “turn” 방식, in which they do not perceive surrounding context while generating an answer after a user finishes input.

Thinking Machines pointed to what it called a structural limitation in which an AI becomes, in effect, deaf and blind while preparing a response. It presented as an alternative a “time-based” interaction model that continuously takes in and outputs audio, text and video data.

She said processing data by splitting it into fine 20-millisecond slices can capture the complex and subtle nuances of human conversation, including silence, interruptions and overlapping speech.

“This allows us to implement rich interaction between machines and humans,” Murati said.

Murati also stressed that while competitors focus on maximizing autonomy for machines to judge and act on their own, Thinking Machines is focusing on developing AI into “tools for thought” that can contain complex interaction and human intent.

She likened “human-in-the-loop” not to a simple confirmation step for approval but to a “tandem bicycle” in which 2 people pedal together. “The core philosophy and differentiator is not to let AI advance on its own, leaving civilization behind, but to create an environment where humans and AI systems can take the wheel together and cooperate,” she said.