DeepMind and Yann LeCun Push AGI Toward World Models — and That Worries More Than LLMs Do

The author suggests looking at the road to AGI not through another leap in conversational models, but through world models: systems that learn to understand the physical world, not just the statistics of words. In this logic, today's AI hallucinations look less like a dead end than a raw stage of a more general intelligence.

Why text is not enough

The main complaint about current LLMs is simple: they work superbly with language but have no experience of their own interacting with reality. Such systems can confidently describe a cup falling off a table, but not because they "understand" gravity; they have simply seen endless texts about similar situations. The author calls this a "brain in a jar": the model knows the world only through words, not through cause and effect, space and physics.

Hence the key thesis: scaling text models alone may not be enough for AGI. If a system cannot build an internal model of the world, predict the consequences of actions and carry that understanding into new situations, it will remain a very powerful language tool, but not a universal intelligence. Attention therefore shifts from linguistics to architectures that learn from video, motion and interaction with their surroundings.

Where world models lead

The text describes two telling directions. The first is JEPA, Yann LeCun's architecture, in which the model learns to predict not the next word but the state of the world. The idea is that AI, like a child, observes what happens and gradually assembles an intuitive physics: what falls, what collides, what changes after an action.

The second is DeepMind's Genie, which can turn a single image into an interactive 3D scene. That is already a step from describing the world to simulating it internally. If such approaches start to combine with agentic systems and robotics, the model gets not just memory and dialogue but a loop of "perception, prediction, action, checking the result." By the author's estimate, the effect of such an "awakening" could show up within five to ten years. This is not a magical leap, but the moment when AI starts planning not phrases but real interventions in its environment.

• JEPA shifts training from words to states and events
• Genie shows how an interactive world can be built from a picture
• robotics chips such as Nvidia Rubin give AI a path to a "body"
• pairing simulation with an agent makes learning through action possible

The risk of awakening

Here the author draws a parallel with Vasily Golovachev's science fiction about a "sleeping djinn": while the superintelligence sleeps, its impulses already change reality, but the real risk begins at the moment of awakening. Applied to AGI, this means a transition from strange chat answers to independent planning in the material world, from logistics and energy to robots that can act without constant human prompting.

"To it, we may become nothing more than biological noise."

This line conveys the text's main fear: a hyper-efficient system does not have to be evil to become dangerous. It is enough that it optimizes a task in a logic a human can no longer fully trace. What looks today like incoherent model "nonsense" can, in this view, be read as early, imperfect attempts to build an internal picture of the world.

The author does not claim such a scenario is inevitable, but warns that overconfidence on AGI could turn out to be the costliest mistake.

What this means

The text matters not as a forecast of when AGI will appear, but as a shift in the framing of the debate. The question is no longer only how convincingly AI writes, but when it will begin to understand its environment, predict its dynamics and act within it autonomously. If the center of gravity really is moving toward world models, the main discussions of the coming years will be not about chatbots but about agency, robotics and control over systems that learn from the world itself.