arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2607.08255cs.AI
收藏

竞争然后合作:前沿人工智能教师构建可验证课程以提升编码学生超越模仿

Compete Then Collaborate: Frontier AI Teachers Build a Verifiable Curriculum to Improve a Coding Student Beyond Imitation

Genesis Cortex AI Inc.(创世纪皮层人工智能公司)

Miseong Shawn Kim

AI总结:

研究探讨大型语言模型作教师时的教学问题,提出竞争然后合作框架,让四个前沿人工智能教师先对决排名,再合作构建课程提升学生。发现执行验证下教师解决标准问题情况,模仿不一定提升学生,合作课程能提升学生,强调合作构建可验证环境的价值。

摘要:

大型语言模型越来越多地充当为较小模型生成训练数据的教师。先前的多教师知识蒸馏方法在合并输出时未确定哪个前沿模型教学效果最佳,通常依赖偏向自身输出的语言模型评判。我们引入了一个竞争然后合作的框架,其中四个前沿人工智能教师(Claude、Codex - GPT、Grok、Gemini)由基于执行的评判(单元测试和标准输入输出检查)进行公平控制的两两对决排名,然后合作构建一个可验证课程给学生(Qwen2.5 - Coder)。我们报告了三个发现。(1)在执行验证下,由于饱和效应,所有教师在自我纠正后几乎完美地解决标准问题(99 - 100%),但更具挑战性的竞争问题将它们区分开来(Gemini 77% > Claude 69% = Codex 69% > Grok 50%);然而,稳健的学生端结果并不依赖于教师排名。(2)对已验证解决方案的模仿(监督微调)不会提升,甚至可能降低70亿和320亿参数规模下已有能力的学生(例如,在MBPP测试中从76.7%降至72.7%,在竞争问题中从5.9%降至2.9%)。(3)使用相同的合作课程作为具有可验证奖励的强化学习(RLVR)环境可提升学生(在竞争问题上从5.9%提升至峰值8.8%,相对增益 +49%),扭转了监督微调的趋势。人工智能教师合作的价值不在于汇集答案进行模仿,而在于共同构建一个学生通过实践学习的可验证环境。我们发布了一个可重现的本地管道(NVIDIA GB10)以及框架补丁,用于在前沿堆栈上运行GRPO。

英文摘要:

Large language models increasingly serve as teachers generating training data for smaller students. Prior multi-teacher knowledge distillation methods merge outputs without determining which frontier model teaches best, often relying on an LLM judge biased toward its own outputs. We introduce a compete-then-collaborate framework where four frontier AI teachers (Claude, Codex-GPT, Grok, Gemini) are ranked head-to-head by an execution-based judge (unit tests and stdin-stdout checks) with fairness controls, and then collaborate to build a verifiable curriculum for a student (Qwen2.5-Coder). We report three findings. (1) Under execution verification, all teachers solve standard problems near-perfectly after self-correction (99-100%) due to a saturation effect, but harder competition problems separate them (Gemini 77% > Claude 69% = Codex 69% > Grok 50%); however, the robust student-side results do not depend on teacher ranking. (2) Imitation (SFT) on verified solutions does not improve, and can degrade, an already-competent student at 7B and 32B (e.g., from 76.7% to 72.7% on MBPP-test, and 5.9% to 2.9% on competition problems). (3) Using the same collaborative curriculum as a reinforcement learning with verifiable rewards (RLVR) environment improves the student (from 5.9% to 8.8% peak on competition problems, a +49% relative gain), reversing SFT's direction. The value of AI-teacher collaboration lies not in pooling answers to imitate, but in jointly constructing a verifiable environment where the student learns by doing. We release a reproducible on-prem pipeline (NVIDIA GB10) with framework patches for running GRPO on a bleeding-edge stack.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑