Thinking Machines Lab, the AI startup formed last year by former OpenAI CTO Mira Murati, announced a concept called interaction models on Monday. Essentially, it’s AI designed to interrupt you.
Currently, all AI models operate in the same manner. You speak, the AI listens. It responds, and you listen. Thinking Machines aims to alter this by developing a model that processes input and generates responses simultaneously, making it more like a phone conversation rather than a text exchange.
The technical term for this process is “full duplex,” and the company asserts that its model, TML-Interaction-Small, responds in 0.40 seconds, aligning with the speed of natural human dialog and considerably quicker than similar models from OpenAI and Google.
However, this is only a research preview, not a commercial product. The company isn’t making it publicly available yet. A “limited research preview” is expected in the coming months, with a broader release planned for later this year.
What can be concluded? It’s unclear. The benchmarks are remarkable, and the fundamental concept — that interactivity should be integrated into a model rather than added as an add-on — is intriguing. Whether the practical experience meets the technical promises remains to be seen until users have access to it.