From OpenAI to Thinking Machines: A Different Vision for AI
Mira Murati, former CTO of OpenAI and one of the leaders behind ChatGPT, is now betting on a different frontier: human-speed AI. Her new venture, Thinking Machines Lab, has unveiled what it calls “interaction models,” systems designed not just to answer prompts but to participate in continuous conversations. Instead of the familiar turn-based pattern—user speaks, AI waits, then replies—these models listen, watch, and respond at the same time. Murati frames the problem as one of interactivity, arguing that how we work with AI should not be an afterthought. The company’s vision is less about squeezing out extra benchmarks and more about closing the gap between human thought and machine processing, so collaboration with AI feels closer to working with another person than issuing commands to a tool.
What Human-Speed AI Actually Looks Like
Thinking Machines’ flagship model, TML-Interaction-Small, responds in about 0.40 seconds, a pace deliberately chosen to mimic natural conversation. Rather than processing a full sentence and then replying, the system slices interaction into roughly 200-millisecond chunks. One component tracks the flow of dialogue while another tackles more complex reasoning in the background—much like how a person can keep talking while planning what to say next. The model is built to handle audio, video, and text simultaneously, enabling it to watch, listen, and speak at once. Demos show the AI counting exercise reps from video, translating speech in real time, and even noticing posture changes such as slouching, all while maintaining conversation. This human-speed AI is less about beating benchmarks and more about matching the rhythms of human attention and response.
Beyond Speed: Rethinking AI as a Collaborative Peer
Most AI collaboration tools are optimized for throughput: you feed in a prompt, wait, and receive a block of output. Murati’s team is challenging that paradigm by targeting the “bandwidth bottleneck” between humans and AI—the friction created by turn-based exchanges. By allowing the system to listen while it talks and see while it responds, Thinking Machines AI aspires to function more like a peer in a meeting than a background assistant. In practice, this could mean an AI that joins a video call, tracks slides, listens to speakers, and interjects with relevant data or clarifying questions at natural moments. Such conversational AI design places emphasis on timing, context, and shared situational awareness. The goal is not to automate humans out of the loop but to make the loop itself more fluid, responsive, and cognitively aligned.
Implications for Knowledge Work and Future Workflows
For knowledge workers, human-speed AI could reshape daily workflows. Imagine drafting a report while an AI partner tracks your references, flags inconsistencies, and responds to spoken questions without pausing your train of thought. In creative work, an AI could react mid-brainstorm—suggesting examples, surfacing prior work, or annotating a shared document as ideas emerge. In operations, it might monitor live video feeds, conversations, and documents simultaneously, stepping in with reminders or alerts as needed. This represents a philosophical shift from AI as a tool you periodically consult to AI as a persistent collaborator that shares your tempo. Thinking Machines’ technology remains in research preview, with limited access planned for partners before a broader release, but its design signals a future where AI doesn’t merely answer faster—it learns to think with us, at our speed.