Abstract
Large language models (LLMs) increasingly serve as teachers generating training data for smaller students. Previous 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:
- Under execution verification, all teachers solve standard problems near-perfectly after self-correction (99-100%), 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.
- 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).
- 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.
Blogger's Review: This article explores the competition and collaboration between AI teachers, emphasizing the importance of verifiable learning environments and highlighting the limitations of traditional imitation learning. Through proper assessment and collaboration, AI can more effectively enhance students' programming skills, providing new insights into future educational models. The potential applications of this approach in reinforcement learning are particularly worthy of further investigation.