Professor argues SWE lacks math-style open problems—yet AI models still crush 'solved' tasks
DimitrisPapail · x · 2026-10-07
Professor Dimitris Papailiopoulos argues that software engineering doesn't have open problems in the same sense as math, where an open problem means all human brain-tokens spent on it weren't enough, yet AI comes crushing through. He questions whether CS/SWE has analogous "many smart people tried for long and failed" problems—and whether such a characterization is even needed, given current models can already perform long-horizon, complex research tasks. A reply cites seeing models do autoresearch on compressing classifiers for thoroughly beaten problems like MNIST, achieving 2-3x better than state of the art—requiring ruthless iteration, tons of tools and environments, and weeks of long context, beyond what any expert could do.
More from AGI Musings
- $1.8B CZ Biohub–DeepMind–Meta effort aims to build AI models of living cells — kimmonismus · 2026-10-07
- AI Ads for AI Apps for an Audience Being Replaced by AI: 'The Circularity Is Here' — sull · 2026-10-07
- Ex-Google's Lucas Beyer: Prof's 'Fudge Factor' Admission Was an Eye-Opener and Life Saver — giffmana · 2026-10-07
- $50 and GPT-4.1 made 50,000 fake stats — ChatGPT cited them 72,000 times a month — metehan777 · 2026-10-07
- While others prophesy math's death, a professor considers hiring a math PhD for the first time — dyamins · 2026-10-07
- Terence Tao: AI could support the whole research pipeline, but the opposite is happening — GaryMarcus · 2026-10-07