Notes from the causal frontier
How Causal Discovery Actually Work
“Causal graph” sounds like jargon, but the idea is almost embarrassingly simple — and once it clicks, a lot of confusing things about data suddenly make…
Read the post →Fluent about physics, not grounded in it: LLMs and physical world models.
Ask a frontier language model a hard physics question today and it will likely nail it. The latest models score near-perfect on university-level classical mechanics, electromagnetism,…
Read the post →World Models Are Having a Moment — Here’s What’s Missing
World models are having a moment. After a decade as a mostly academic idea, they have surged to the center of the AI conversation. Google DeepMind’s…
Read the post →The Most Expensive Question in AI: “What If We’d Done It Differently?”
There is a question that sits underneath nearly every important decision, and it is almost never answered honestly. After the campaign runs, the treatment is given,…
Read the post →Ice Cream Doesn’t Cause Drowning
Here is one of the most reliable statistics you will ever meet: on the days when a town sells the most ice cream, the most people…
Read the post →Why Agentic AI Needs a Model of Consequence
The center of gravity in AI is shifting. For years the goal was a better predictor — a system that, shown the world so far, tells…
Read the post →Data-Efficiency as a First-Class Advantage
Every AI system runs on data, but not every system runs on the same amount of it — and in the real world, that difference decides…
Read the post →Counterfactual aerodynamics: what a causal world model does for CFD
Every aerodynamicist knows the ritual. Tweak the geometry. Rebuild the mesh. Launch the solver. Wait — sometimes minutes, often hours, occasionally overnight — while it churns…
Read the post →Counterfactual aerodynamics: what a causal world model does for CFD
Every aerodynamicist knows the ritual. Tweak the geometry. Rebuild the mesh. Launch the solver. Wait — sometimes minutes, often hours, occasionally overnight — while it churns…
Read the post →The model that trains itself: the flywheel behind a self-enriching world model
Most machine-learning models are frozen the moment they finish training. From that point they only decay — the world drifts, the data ages, and the model…
Read the post →Correlation, intervention, counterfactual: the three levels every AI system lives on
Ask an AI system a question and, underneath the answer, it is operating at one of three levels. The levels form a hierarchy — each one…
Read the post →Causal discovery and its relationship with Reinforcement learning (RL)
A reinforcement learning agent is, whether it knows it or not, running experiments. Every action it takes is an intervention on the world — a deliberate…
Read the post →Prediction Tells You What. Counterfactuals Tell You What If.
A weather forecast tells you it will rain tomorrow. Useful. But it never has to answer the harder question: would it have rained if we hadn’t…
Read the post →Why physics needs a world model
Why physics needs a world model For generations, simulating a physical system has meant one thing: write down the governing equations and solve them. Navier–Stokes for…
Read the post →