Thinking Machines Lab, an AI startup founded last year by former OpenAI CTO Mira Murati, announced on Monday what it essentially calls an interaction model that sounds like AI that can disturb you.
Every AI model you have used today works the same way. When you speak, it listens. It responds. You listen. Thinking Machines is trying to change this situation by building models that process user input and generate responses simultaneously. So it’s more like a phone call than a text connection.
The technical term for this is “full duplex,” and the company claims its model, TML-Interaction-Small, responds in 0.40 seconds. This is roughly the speed of a natural human conversation and much faster than similar models from OpenAI and Google.
However, this is a preview of the research, not the product. The company has not yet disclosed this to the public. A “limited research preview” will be released in the coming months, with a wider release planned for later this year.
So what should we make? I’m not sure. The benchmarks are impressive, and the basic idea that interactivity should be inherent to the model rather than added is certainly interesting. Whether real-world experience lives up to the technical claims won’t be known until people can actually use it.
