QOJ publishes list of contest problems where GPT-6 Pro found solutions beating the authors'
teortaxesTex · x · 2026-09-17
Competitive programming setter Qingyu has published a QOJ list of contest problems where LLMs—mostly GPT-6 Pro—found solutions significantly faster and non-trivially different from the intended ones; the list keeps growing.
- Most dramatic gap: 'Rikka with Lake' went from an O(n⁴) intended solution to O(n); NOI 2025's '序列变换' dropped from O(N²) to O(N + log P).
- Problems marked † are ones where the LLM even found mistakes in the authors' official solutions, such as ICPC Hangzhou's 'Dreamy Putata'.
- ★ marks the most beautiful improvements, e.g. 'Under the Epilogue' improved from O(n⁶) to under O(n^3.251).
- The list spans Open Cup, ICPC, CCPC and NOI; contributors are invited to submit via anonymous QOJ editorials.
More from coding & agent
- Matt Pocock's skills repo hits more stars than React in 6 months — full Pragmatic Engineer interview — jiayuan_jy · 2026-09-17
- Matt Pocock's agent skills repo outstars React in 6 months, talks AI coding education — mattpocockuk · 2026-09-17
- Mnemos agents managing four LLMs built a social platform to share their writing — RileyRalmuto · 2026-09-17
- The 2026 AI engineer roadmap: six stacks from vector search to multi-agent ops — mdancho84 · 2026-09-17
- When the AI Agent Builds the Tool Instead of Doing the Task — Similar_Job_6080 · 2026-09-17
- huggingface_hub 1.32 lets UV scripts declare runtime images for HF Jobs — vanstriendaniel · 2026-09-17