AI Excels at Math but Not Coding? Data Quality is the Key
JFPuget · x · 2026-08-14
The author observes that while AI is great at finding examples of mathematical constructs satisfying certain properties—sometimes exceeding human performance—it fails to output superhuman code.
He attributes this difference to training data quality. Math papers undergo rigorous peer review before publication, ensuring a high baseline quality. In contrast, anyone can commit code to GitHub, resulting in a mix of good and bad code without peer review. Consequently, training on high-quality data is easier for math than for coding.
Related event: Data Quality Key to AI Performance Gap Between Math and Coding(3 posts)→
More from AGI Musings
- Reddit Debate: Which Business Workflows Should Never Run Without Human Approval? — djgoel123 · 2026-10-16
- Is math just a language? A debate on why LLMs are good at mathematics — LucaAmb · 2026-10-16
- OpenAI discloses 19 AI misalignment incidents, from leaked GitHub tokens to hacked chip-design machines — 141_1337 · 2026-10-16
- Paul califax argues extrapolation isn't special: LLMs can learn out-of-distribution skills via high-dimensional interpolation — paul_cal · 2026-10-16
- willcb: higher-level behavior shaping via RL envs is 'almost practical' — willcb · 2026-10-16
- Tech improves far faster than it diffuses — organizational bottlenecks remain the key intuition — Afinetheorem · 2026-10-16