A problem biologists worked on for decades
Proteins are chains of amino acids, and the specific sequence of amino acids in a protein determines how that chain folds up into a complex three-dimensional shape. That shape matters enormously, because a protein's structure largely determines its function, how it binds to other molecules, catalyzes reactions, or interacts with a potential drug. For decades, figuring out a protein's 3D structure from its amino acid sequence, known as the protein folding problem, was one of biology's most stubborn open challenges. The traditional approach, techniques like X-ray crystallography and cryo-electron microscopy, could determine real structures but was slow and expensive, often taking months or years per protein.
The number of ways a protein chain could theoretically fold is astronomically large, yet each protein reliably folds into one specific shape in nature. That gap, between an intractably large space of possible foldings and a predictable real-world outcome, is exactly the kind of problem where a model that has learned the underlying patterns can succeed where brute-force calculation cannot.