Most founders are wired to optimize. You find a problem, you build a solution, then you tweak it, iterate, and push for marginal gains. But what if the problem you're trying to optimize around is built on a faulty premise? What if the constraint you accepted as absolute is actually a mirage?

That's the radical insight from OpenAI's Alex Lupsasca, who discussed how GPT models tackled a year-long puzzle in theoretical physics. The AI didn't just solve a complex problem; it overturned a deeply held assumption, revealing a far simpler truth.

Key Takeaways

  • OpenAI's GPT models resolved a physics puzzle that stumped human experts for over a year: defining single-minus gluon tree amplitudes. These specific interactions were long assumed to be impossible.
  • The AI didn't just confirm existing theories; it challenged the fundamental assumption that these gluon interactions were "zero" (i.e., forbidden), conjecturing they were indeed non-zero.
  • GPT derived a concise, "linear" formula for these amplitudes, a dramatically simpler solution compared to the "factorial growth" (super exponential) typically anticipated for such complex particle interactions.
  • This discovery reshaped understanding of strong force interactions, demonstrating AI's capacity to question and rewrite established scientific principles rather than just process them.

The Gluon Problem: Zero or Not?

For ambitious builders, theoretical physics might seem far removed from daily sprints and market traction. Yet, the story of how AI cracked the gluon puzzle offers a masterclass in challenging core assumptions. At the heart of the matter lay single-minus gluon tree amplitudes, a specific interaction within the strong force, one of nature's four fundamental forces. These amplitudes describe how gluons—the particles that bind quarks together—interact. For a long time, the scientific consensus was clear: these particular interactions were forbidden. As Alex Lupsasca explained, “It's been known for a long time that actually in that case, the amplitude is just zero. Which means the interaction is forbidden and cannot happen.”

Human experts had spent a year trying to unravel this. They were stuck, perhaps because they were working within the accepted framework that these interactions were, by definition, zero. They sought to understand why they were zero, or how to work around that constraint. This is a common pitfall: when a problem is deemed impossible, we stop looking for solutions and instead try to rationalize the impossibility.

AI's Linear Logic Unlocks a Fundamental Mystery

OpenAI's GPT models approached the problem with a clean slate. Instead of accepting the 'zero' premise, they probed what might happen if these amplitudes were, in fact, non-zero. The AI conjectured a solution, and then, remarkably, proved it. “So, then suddenly these really simple amplitudes previously thought to be zero, if they're not zero, we should compute them and they should be something really nice and simple and special,” Lupsasca said. The elegance of the AI's solution wasn't just in its accuracy, but in its profound simplicity.

Physics problems often lead to formulas with "factorial growth," meaning the complexity explodes as the number of variables increases. But GPT's discovery was different. "The amazing thing is that the formula that it proposed, instead of having this factorial growth which is super exponential... Here it's actually linear," Lupsasca elaborated. "So if you double the number of particles, you only double the number of terms. It's the nicest possible behavior you could imagine." The AI didn't find a more complex way to justify 'zero'; it found a simpler, linear truth that revealed these interactions do happen. As Lupsasca put it, "The title is single minus gluon tree amplitudes are non-zero. So these special interactions between gluons... which were previously thought to never occur, actually these interactions can happen."

What to Do With This

Identify one core assumption in your product, market, or operations that you've held for years – perhaps a feature deemed "impossible" or a market segment "unreachable." Instead of optimizing around that accepted constraint, dedicate a 2-hour sprint next week. Brainstorm how things would change if that assumption were completely false, and outline the simplest, most linear path to test that counter-intuitive reality.