“The world is everything, that is the case,” Ludwig Wittgenstein wrote in Tractatus Logico-Philosophicus in 1921. A century later, artificial intelligence may finally be taking him at his word. That is the opening move in Jaspreet Bindra’s Economic Times column, The Next Great Leap — and it sets up his central argument: AI world models, not ever-larger language models, are the next real inflection point in AI. The founder of AI&Beyond makes the case that we have spent four years mistaking eloquence for intelligence, and that the frontier has quietly moved from words to worlds.
The Trigger: One Photograph, a Whole World
Bindra anchors his argument in a demo that went viral earlier this month. World Labs — founded by AI pioneer Fei-Fei Li — released Atlas. Give it a single photograph, and you can move a virtual camera around it, look at the scene from the side, from above, from behind, even at parts the original camera never saw. From one flat image, Atlas imagines a coherent three-dimensional world. Feed it a few more images and imagination transforms into reconstruction. World Labs calls Atlas an “omni world model”, trained natively across text, images, video and 3D.
For Bindra, that demo is not a party trick. It is evidence that the centre of gravity in AI is shifting.
Why Bigger LLMs Are Not the Next Frontier
Since ChatGPT burst onto the world four years ago, AI has been synonymous with LLMs. Every month brings more parameters, more training tokens, more billions of dollars. Bindra’s point is not that these models are weak — it is that scaling them is no longer where the leap lies.
His reasoning is disarmingly simple. The world is not a paragraph of words. It exists in a glorious three-dimensional cornucopia of objects, sounds, sights and senses. LLMs acquired extraordinary capability by consuming enormous quantities of text and images. But humans do not learn this way. A child does not read Wikipedia before understanding that a cup falls when pushed off a table, or that a ball rolling behind a sofa still exists. As Bindra puts it, AI’s “pre-training data” is the textual and visual internet; our pre-training dataset is reality itself.
This is where AI world models enter. Long before children acquire sophisticated language, they have absorbed an astonishing amount of common sense about objects, distance, gravity, other people, and cause and effect.
The Bandwidth Argument
Bindra adds a second, underrated point: bandwidth. Typing communicates words. Speaking adds tone, rhythm, emphasis and sound — and that is still the best way most of us work with AI today. But vision adds infinitely more: objects, depth, movement, context. It makes the informational richness of an experience explode.
Which is why, he argues, children do not need to be trained on the “entire internet”. They learn remarkable things from comparatively little lived experience, because each experience carries vastly more structure than a sentence does. That single line is the strongest compression of the whole world-models thesis.
Yann LeCun’s Warning and Fei-Fei Li’s “Spatial Intelligence”
Bindra situates his view within a broader shift among AI’s foundational researchers. Yann LeCun has argued for years that human-level intelligence cannot emerge simply by feeding machines ever more human-written text. Current systems, LeCun says, are extremely good at manipulating language — which can fool us into “equating eloquence with intelligence.” His alternative is AI that learns how the physical world works, predicts what will happen after an action, and uses that prediction to plan.
Fei-Fei Li has staked her next chapter on a related idea: spatial intelligence. Where language lets machines understand and generate words, spatial intelligence lets them reason about objects, places, geometry, movement and interactions in the physical world. She calls it the transition “from words to worlds.”
What an AI World Model Actually Does
At its simplest, a world model is an internal representation of an environment that allows AI to understand its state, imagine how it might change, and predict the consequences of actions.
Bindra frames the difference in one elegant comparison. An LLM asks, loosely: “What token comes next?” A world model asks something closer to: “If this is the world now, and I do this, what will the world look like next?”
It sounds like a small shift. Its complexity and implications are enormous.
Gaussian Splats: The New Building Block
Here Bindra introduces the most wonderfully named concept in the piece. If a token is a small unit in a stream of language, a Gaussian splat is a tiny, fuzzy, semi-transparent blob positioned in 3D space. Each carries information about where it sits, its shape, orientation, opacity and appearance. Millions of them blend to form a photorealistic three-dimensional scene.
“Gaussian” comes from Carl Friedrich Gauss and the bell-shaped mathematical distribution bearing his name. “Splat” refers to how these fuzzy blobs are projected — splatted — onto the screen to render the scene. Simplistically put: tokens help machines talk about reality; splats help machines build it and render it. Atlas can generate explicit 3D worlds as Gaussian splats, including geometry no camera originally captured.
Where This Goes Next
Bindra’s future scenarios are concrete rather than utopian. A robot needs a world model to understand a kitchen, not merely recognise the word “kitchen”. Autonomous vehicles need to predict how pedestrians and traffic will move. Factories could create living digital twins and test changes before touching the production line. Architects could walk through buildings before constructing them; students could step into ancient Rome instead of reading a paragraph about it. Personal AI could understand where things are in our homes, not merely what our calendars say. At global scale, the same conceptual direction matters for weather, climate, materials and biology.
He is careful about the obstacles. They are formidable: physical causality is far harder than visual realism, world models need far richer data, better representations of physics, and vastly more grounding in reality. But the direction, he insists, is important.
His closing line is the one to remember: in the beginning was the Word. AI learnt that first. Now it is reaching for the World.