Thinking Machines and interaction models

Thinking Machines just released Interaction Models. This is their first real AI model release1 after a year of work and two billion dollars of capital. What is an “interaction model”? First, it’s not a frontier model. Thinking Machines is not yet competing with OpenAI, Anthropic and Google. Instead, they’re working on the problem of better real-time interaction with models. Some parts of what they’re doing are not new at all, other parts are slightly-questionable benchmark gaming, and still other parts represent a genuine technological advancement. I’ll try to lay it all out. Fully-duplex voice models If you’ve used ChatGPT in audio mode, you know that you can’t talk to it exactly how you’d talk to a human. There’s a big latency gap between when you finish talking and when the model jumps in. The model won’t interrupt you like a human, and doesn’t react to you interrupting it like a human would either. And of course you can’t give the model visual feedback like facial expressions.…