Thinking Machines: Mira Murati's Interactive AI

⚡

Key Takeaways

1Thinking Machines, founded by Mira Murati, is developing "interaction models" for smoother AI collaboration.

2These models aim to integrate audio, video, and text in real-time, enhancing the continuous perception of AI.

3A limited preview of these models will be available in the coming months, with a broader launch planned for this year.

💡Why it matters — This innovation could transform the way users interact with AI, making exchanges more natural and efficient.

Thinking Machines: Mira Murati's AI Innovation

Thinking Machines, the artificial intelligence company founded by Mira Murati, former CTO of OpenAI, recently announced an innovative project: “interaction models.” These models aim to transform the way users collaborate with AI by enabling interactions as natural as those between humans. The idea is to seamlessly integrate audio, video, and text while providing the ability to think, respond, and act in real-time.

Understanding Interaction Models

Currently, AI models perceive reality through a single channel. This means they wait for the user to finish typing or speaking before reacting. In the meantime, the model does not perceive what the user is doing or how they are doing it. This limitation creates a narrow channel for collaboration between humans and AI, restricting the amount of knowledge, intentions, and judgments that the user can convey to the model, as well as its understanding of the work being done by the AI. It’s somewhat like trying to resolve a significant disagreement over email rather than face-to-face.

Goals and Vision of Thinking Machines

Thinking Machines aims to overcome this bottleneck by making AI interactive in real-time, regardless of the modality used. This means that AI interfaces will be able to adapt to users, rather than forcing users to adapt to the interfaces. This approach could revolutionize the way we interact with artificial intelligence technologies.

Application Examples of the Models

To illustrate the potential of these models, Thinking Machines shared several concrete examples. For instance, the AI could listen for mentions of animals in a story, translate speech in real-time, or alert a person when they are slouching. These applications demonstrate how interaction models can enrich communication and collaboration with AI.

Launch and Availability

Although users cannot yet experience these models, Thinking Machines plans to launch a “limited research preview” in the coming months. A broader launch is anticipated later in the year. Founded in February 2025, the company has already faced significant departures, with some members joining Meta or returning to OpenAI.