Elegance and intelligence are anticorrelated

Open original

Sean Goedecke argues that very smart people often write inelegant code or solve problems by brute force, because raw working memory lets them skip the clever abstractions everyone else needs. He extends the idea to LLMs, suggesting their tolerance for long, messy proof chains could make human creativity more valuable.

  • Von Neumann solved the fly-and-cyclists puzzle instantly by summing the infinite series, skipping the elegant shortcut most people look for.
  • For software engineers, the key trait is working memory, and engineers with large working memory can tolerate sloppy codebases because they can hold the whole system in their heads.
  • LLMs can parse hundreds of proof steps as easily as ten, which may make elegant abstractions less necessary for them, and that could weaken their ability to generate new conjectures.

Sean Goedecke argues that raw intelligence and elegance can pull in opposite directions. Smart people often skip the clever trick because they can brute-force the problem. He opens with John von Neumann, who was asked how far a fly flies between two cyclists approaching each other at constant speed. The elegant answer is to notice the cyclists meet after an hour, so the fly flies for an hour. Von Neumann got the answer instantly, and it turned out he had simply summed the infinite series in his head. Goedecke's point is that when you can do the hard calculation effortlessly, the elegant shortcut stops being worth finding. He applies this to software engineering, where the key trait is working memory, meaning how many moving parts you can hold at once. Good design is largely about reducing those parts through abstraction and interfaces. Engineers with huge working memory may find that unnecessary, and some of the best of them are comfortable in messy codebases because they can trace a ten or twelve step execution flow at a glance. Goedecke notes the pattern isn't universal, since some strong engineers insist on neatness, and he admits it can be maddening to work with people whose abstractions feel fiddly to you. The same logic, he suggests, may explain why AI-generated mathematical proofs look bloated to human mathematicians. For a capable LLM, a few hundred steps may be no harder to parse than ten. That matters because elegant abstractions are often what spark new conjectures, so inelegant AI proofs could worsen the risk of solving problems without generating new ones. Goedecke doubts LLMs will escape the need for real creativity, since even von Neumann invented a lot of new mathematics. His tentative conclusion is that human creativity, shaped by limited working memory and irrational, jury-rigged reasoning, may be one of our remaining edges over machines.

Ask about this article