Mira Murati’s AI startup reveals its first product
▼ Summary
– Current AI models process user input and generate responses in a single, sequential thread, pausing perception during each step.
– This creates a narrow communication channel that limits how much human knowledge, intent, and judgment can be shared with the model.
– The model’s perception freezes while generating output, preventing it from receiving new information until it finishes or is interrupted.
– Thinking Machines proposes solving this bandwidth bottleneck by making AI interactive in real time across any modality.
– This approach would allow AI interfaces to meet humans where they are, instead of forcing humans to adapt to AI’s limitations.
Today’s artificial intelligence systems experience reality in a single thread. While a user is still typing or speaking, the model passively waits, blind to what the person is doing or how they are doing it. Likewise, while the model generates a response, its perception freezes,it receives no new input until it finishes or is interrupted. This creates a narrow channel for human-AI collaboration, severely limiting how much of a person’s knowledge, intent, and judgment can actually reach the model, and how much of the model’s output can be properly understood. Imagine trying to resolve a critical disagreement through email rather than face-to-face conversation.
At Thinking Machines, the startup founded by former OpenAI CTO Mira Murati, the team believes this bandwidth bottleneck can be solved. Their vision is to make AI interactive in real time across any modality, enabling interfaces that meet humans where they are, rather than forcing people to twist their behavior to fit rigid AI systems. Today, the company unveiled its first product, marking a significant step toward that goal.
(Source: The Verge)
