The Search Space: What Happens When Physical Reality Becomes Economically Explorable
- Rich Washburn
- 1 day ago
- 6 min read
OpenAI just posted ten mathematical proofs. A model called ASTRA — apparently the class above GPT-5.6 Soul — solved ten longstanding problems that humanity had been stuck on for decades. Some for fifty years. The total cost was $2,000 in API tokens.
That's not a typo. Ten problems that the world's best mathematicians and supercomputers couldn't crack in half a century. Solved for the price of a decent GPU.
And here's the part that should make you stop scrolling: the model isn't brute-forcing these problems. It's not running quadrillion-operation sweeps until something clicks. It's doing something more interesting and more dangerous — it's pulling together bodies of knowledge from different mathematical disciplines and finding connections that no single human has deep enough expertise across all of them to see.
As one mathematician commented on the previous OpenAI breakthrough: if you put all the top experts from different fields in a room and had them share ideas, they probably would have found something like this. But we didn't put them in a room. We put them in separate buildings, separate universities, separate decades. The model just... collapsed the distance. Terrence Tao — one of the greatest living mathematicians — compared it to the industrial revolution. Before the factory, a craftsman made one thing, start to finish, alone. After the factory, a system of people and machines produced things on a conveyor belt. No single person is "the maker." Tao thinks mathematics is about to go through the same shift. New theorems will roll off a conveyor belt, and a new profession will emerge: people who take ugly machine-generated proofs and make them humanly comprehensible. That's the frame. Now let me tell you why this is bigger than math.
Your iPhone Is a Materials Science Achievement
Your iPhone is not really a consumer-electronics product. It is a stack of materials breakthroughs wearing a user interface: Chemically strengthened glass. Semiconductor-grade silicon. Nanometer-scale insulating and conducting layers. Rare-earth magnets. Lithium-ion chemistry. Engineered adhesives. Thermal-management materials. Corrosion-resistant alloys. Microscopic pigments and coatings. Polymers tuned for flexibility, durability and radio transparency.
Even the fact that it can survive sitting in your sweaty pocket while receiving radio signals, dissipating heat and displaying millions of colors is a materials science achievement. Every component, every layer, every material in that device was discovered, characterized, refined and manufactured through a process that is fundamentally about understanding the mathematics, chemistry and physics of matter. The pigments that produce the yellow or green or blue you pick at the store — those aren't arbitrary colors. They're engineered materials.
Material science almost doesn't do itself justice as a term. It covers metals, chemistry, semiconductors, biomaterials, polymers, coatings, composites, crystals, alloys, glasses, ceramics, adhesives, radioactive shielding, optical films, battery electrolytes and about a thousand other categories that each represent decades of human investigation.
And this is the kind of thing that AI is about to spin the dial right off whatever the dial is mounted on.
Medicine Is a Materials Problem
People hear "AI in medicine" and imagine a chatbot diagnosing a rash or a model reading an MRI. That is almost the least interesting version.
The gigantic version is AI helping us reason through the underlying mathematics, chemistry and physics that determine:
Which molecules bind to which proteins.
Which materials the immune system accepts or rejects.
How a drug dissolves and travels through the body.
How to target one type of cell without poisoning everything around it.
How to build implants that are stronger, lighter and less inflammatory.
How to create artificial tissue scaffolds. How to engineer membranes that filter blood. How to deliver medicine across the blood-brain barrier.
How to make sensors that can live inside the body.
How to create new contrast agents, surgical adhesives, prosthetics and replacement organs.
That is materials science too. Biology itself is, at least partly, an obscenely complicated materials system that learned how to reproduce. The question isn't whether AI can help with medicine. The question is what happens when the mathematical reasoning that underlies drug discovery, biomaterial engineering and molecular design acquires a computable price.
When Math Gets a Price Tag
The part of the ASTRA announcement that matters most is not necessarily whether every claimed proof survives peer review. It's the proposed economic transition.
Mathematical and scientific reasoning begins to acquire a computable price.
Instead of saying, "Perhaps a once-in-a-generation mathematician will connect these three distant fields sometime in the next 40 years," you start asking: how much inference compute should we allocate to the problem?
The ASTRA model found these solutions not by doing something alien to human mathematics, but by doing something humans couldn't do fast enough — combining existing mathematical knowledge across disciplines in ways that no single specialist could hold simultaneously. It's existing math, remixed in ways we hadn't thought of.
That is the dial being ripped off the panel. Because material discovery has historically been constrained by some combination of: What humans could theorize. What humans could synthesize. What humans could test. How quickly knowledge moved between disciplines.
AI attacks the first and fourth immediately. Robotics, automated laboratories and simulation begin attacking the second and third.
The Full Loop
So now imagine the complete cycle: AI proposes a material. Simulation rejects 99.999% of the bad candidates. A robotic lab synthesizes the survivors. Automated instruments characterize them. The results go back into the model. The model proposes the next generation. Not once a year. Not after a grant cycle. Not after five papers, three conferences and two retiring department chairs finally introduce one another. Continuously.
That loop doesn't just produce a better phone screen. It produces better batteries, grid storage, desalination membranes, superconductors, catalysts, semiconductors, radiation shielding, carbon capture, aircraft, spacecraft, medical implants, drug-delivery systems and manufacturing processes.
Pick Any Problem
Pick almost any supposedly separate global problem and eventually you discover that some piece of it is waiting on a material that doesn't exist yet, is too expensive, is too unstable, can't be manufactured at scale or hasn't been discovered.
Energy is a materials problem. The bottleneck isn't policy. It's battery density, catalyst efficiency, and semiconductor performance.
Compute is a materials problem. The bottleneck isn't software. It's lithography, thermal dissipation, and dielectric scaling.
Climate adaptation is a materials problem. Carbon capture, desalination, reflective coatings, and drought-resistant agriculture all come down to material performance.
Medicine is a materials problem. Drug delivery, tissue engineering, implants, diagnostics, and therapeutics all depend on molecular and material design.
Transportation is a materials problem. Lighter alloys, better composites, more efficient batteries, and stronger structural materials.
Defense is a materials problem. Armor, sensors, propulsion, electronic warfare, and radiation-hardened electronics.
Space is absolutely a materials problem. Heat shields, radiation shielding, structural materials that survive thermal cycling, and propulsion chemistry.
The list doesn't end. Because the physical world doesn't end. Every physical constraint we face is, at some level, a materials constraint. And every materials constraint is, at some level, a mathematical and physical reasoning problem that we couldn't afford to solve at scale until now.
The Other Side of the Wall
The YouTube video that kicked this off asked a simple question: what happens when we can just buy more understanding of math?
The search space of physical reality becomes economically explorable.
We've never lived in a civilization where that was true. Throughout all of human history, discovering a new material required a human mind to theorize it, human hands to synthesize it, and human instruments to characterize it. The cycle time was measured in years. Decades. Sometimes careers. When the cycle time drops to days — or hours — because the reasoning is computable and the synthesis is automated, the constraint doesn't just loosen. It changes kind. It goes from "can we discover this?" to "how much do we want to spend discovering this?"
Nick Bostrom, when asked about a world where AI does most economically valuable work, said he'd rather have a cure delivered immediately by AI than spend 500 years figuring it out ourselves for the glory. But he also said maybe we should leave some mysteries lovingly preserved in jars — some deep-sea crevice where a sea monster might still be lurking — so humans still have something to discover. That's a nice thought. But the sea is very large, and the AI is very fast, and the jars might not be big enough.
The question isn't whether AI will accelerate scientific discovery. It already has. The question is what happens to a civilization when the rate of discovery stops being bottlenecked by the number of human minds working on the problem and starts being bottlenecked by the amount of compute you're willing to point at it.
ASTRA solved ten problems for $2,000. The next model will solve harder problems for less. The model after that will solve problems we can't currently formulate. The dial is already off. We're just watching it spin.
Rich Washburn is a technologist and strategist working at the intersection of AI, infrastructure, and capital. He is Managing Partner and Chief AI Officer at Eliakim Capital and CIO of Data Power Supply.