Thinking Machines, an AI startup founded by former chief technology officer at OpenAI Mira Murati, has unveiled what it calls interaction models; AIs that are able to handle human interaction natively rather than through external scaffolding.

The company said its model will allow humans to collaborate with AI the way we naturally collaborate with each other, continually taking in audio, video, and text, and thinking, responding, and acting in real time.

This is necessary, Thinking Machines said, because it more closely mirrors real-world workflows, where collaboration and clarification over a project’s lifespan determine its success. Current AI interfaces push users out not because they are unnecessary, but because there is no room for them, it said.

This is a result of the “turn-based” interface for traditional large language models. Until a user has finished submitting a prompt, the model has no perception of what the user is doing, or how they are doing it. Similarly, while generating a response a model is not able to take on new information. The company likened the process to trying to resolve a crucial disagreement over email, rather than in person.

Models that appear to be interactive, such as commercial speech systems, “bolt on interactivity like a harness”, the company said. Thinking Machines’ model, in contrast, is able to interact in a more human-like way, interjecting in conversation, speaking simultaneously and performing tasks such as web browsing while continuing to involve itself in the conversation.

Thinking Machines said it will launch the model in the next few months with a “limited research preview.”

---