Thinking Machines Lab's Interaction Models Keep Humans in the AI Loop

Mira Murati, former CTO of OpenAI, has launched Thinking Machines Lab with a contrarian vision for superintelligence: keep humans firmly in the loop. This week the startup previewed what it calls 'interaction models' — AI systems trained to communicate through camera and microphone that natively understand messy human conversation, including pauses, interruptions, and shifts in tone. Unlike most voice modes that simply transcribe speech and feed it into a chatbot-style pipeline, these models interpret continuous communication and adapt on the fly when someone clarifies or changes the subject. The company showed demo videos but has not released the models publicly.

Thinking Machines has raised billions in funding and released one product so far: Tinker, an API launched in October 2025 that allows researchers and engineers to fine-tune open source frontier models with custom data. Founding team member Alexander Kirillov, an expert in multimodal AI, says the interaction models constantly perceive what a user is doing and can reply, search, or use tools — something no current model does natively. Murati frames this as the first bet on human collaboration, aiming to amplify people's preferences and values rather than automate them away.