“have a look at this paper where some floks used the normal generally available Gemini Pro 2.5 to solve 5/6 of the 2025 IMO problems”
Winning Gold at IMO 2025 with a Model-Agnostic Verification-and-Refinement Pipeline
靠验证与精炼提示流水线,让通用大模型解出 IMO 2025 五题
作者设计了一套与模型无关的「验证—精炼」提示流水线,并在 IMO 2025 上评测,对发布早于比赛的模型避免了数据污染。配上 Gemini 2.5 Pro、Grok-4 或 GPT-5 中任意一个,流水线都能正确解出 6 题中的 5 题(约 85.7%),而基线在 32 个候选中取最优仅为 31.6%、21.4% 和 38.1%。结论是推理能力的提升不只靠更强的基座模型,也靠善用模型的方法。
评论者用它说明:只靠提示与编排,公开可用的通用模型(如 Gemini 2.5 Pro)不借助外部工具也能解出大部分 IMO 2025 题目,常用来反驳对 LLM 数学能力的质疑。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“This paper [1] shows that gemini pro 2.5 without data-contamination and some minimal prompting (llm orchestration) can solve problems 1-5 on imo 2025.”
“IMO is not breakthrough, if you craft proper prompts you can excel imo with 2.5 Pro. Paper: Winning Gold at IMO 2025 with a Model-Agnostic Verification-and-Refinement Pipeline.”
“[2] Winning Gold at IMO 2025 with a Model-Agnostic Verification-and-Refinement Pipeline Appendix A”
“Reference: Winning Gold at IMO 2025 with a Model-Agnostic Verification-and-Refinement Pipeline”
“I think you meant to link to”
看全部 9 条 →
“It's reasonable to question it, but there's a fun Chinese paper where they attempt to create a system of prompts that can be used with commercial LLMs to solve hard maths problems.”
“a) no, gemini 2.5 was shown to "win" gold w/o tools. - Winning Gold at IMO 2025 with a Model-Agnostic Verification-and-Refinement Pipeline”
“How can anyone choose to remain so ignorant in the face of irrefutable evidence that they're wrong?”
- Gemini 2.5 Pro Capable of Winning Gold at IMO 2025 ↗__rito__ · 2025-07-22 · 4 分
- Gemini 2.5 Pro Capable of Winning Gold at IMO 2025 with Prompting ↗thorum · 2025-07-24 · 4 分
- Gemini 2.5 Pro Capable of Winning Gold at IMO 2025 ↗chaosprint · 2025-07-25 · 1 分