Beyond Chatbot Hype: Why Mira Murati's New Bet Highlights Current AI's Dead End

If there is one person who knows the secrets, limitations, and behind-the-scenes of ChatGPT, it is Mira Murati. As the former Chief Technology Officer of OpenAI, she led the transition of Large Language Models from laboratory experiments to global mass phenomena. Precisely for this reason, when she speaks publicly to say that the current state of Artificial Intelligence is fundamentally 'primitive', the entire tech industry would do well to stop and listen.

Having left OpenAI, Murati wasted no time: she founded Thinking Machines Lab, a new research-focused entity that has already closed billion-dollar rounds (with a shock valuation of 12 billion dollars led by Andreessen Horowitz) and aims to rewrite the rules of human-machine interaction.

The starting point of her thesis is brutal in its simplicity: today's models — including industry giants like GPT-4, Claude, or Gemini — while processing information, are effectively blind and deaf.

The Condemnation of Static Models: The 'Turn-Taking' Problem

We live in the illusion that clean graphical interfaces and fast text responses equate to fluid interaction. The engineering reality is very different. Current chatbots are based on the rigid concept of turn-taking: the user writes a prompt, the machine pauses, 'thinks', processes a static output, and spits it out.

During this processing phase, the system is totally disconnected from the context:

It does not perceive vocal interruptions naturally.

It is unable to decipher body language or the user's micro-expressions through the camera.

It completely ignores the physical or environmental context of the room where the user is located.

In short: it does not collaborate, it simply executes an advanced statistical calculation based on past inputs. This makes today's systems excellent enhanced search engines or writing assistants, but keeps them dramatically away from being true operational partners.

The 'Tandem Bike' Philosophy of Thinking Machines

The goal of Thinking Machines Lab is not simply to throw more billions into computing power to scale the parameters of a text model (a brute-force strategy that is already showing the first signs of diminishing returns). The goal is architectural.

Murati recently proposed an effective metaphor, defining the next generation of AI as a tandem bicycle. On a tandem, both cyclists have their hands on the handlebars and co-steer in real time. When the climb gets tough, whoever is stronger pedals more, but control and direction remains a synergistic and continuous action. Translated into code, this means developing native multimodal systems where voice, text, and video are not 'extra features' tacked onto a language model, but parallel and constant input channels. A continuous presence that listens, observes, and course-corrects while the user works, studies, or programs.

The Market is Changing: The End of 'Best Answers'

If we analyze this scenario alongside recent trends — such as Anthropic's report on internal code automation or the race for local-first hardware like NVIDIA's RTX Spark — a very clear picture emerges: the race for 'who gives the smartest text answer' is almost over. The new war is being fought over contextual integration.

The next technological revolution will not be determined by a chatbot you ask questions to in a browser, but by software infrastructures capable of integrating into the messy, dynamic way humans collaborate in reality.

For startups and developers who continue to build businesses based on pure prompt manipulation (so-called wrappers), the birth of entities like Thinking Machines Lab is yet another alarm bell. Value is shifting from content generation to workflow fluidity. And those who do not understand that AI must stop being a static tool to become a dynamic collaborator are destined to be left behind.