AI 开始自己改进 AI：一个 35B 模型把自家训练成绩从 39% 拉到 71%
AI 改进 AI 不再是口号，这篇直接把它做成了可运行的闭环。研究者搭了一套完整系统：先给 AI 一个能验证的“健身房”（任务环境），让它反复练习四个基本动作——起草、改进、调试、交叉——再用强化学习把这些动作练成肌肉记忆，最后让它在 12 小时内自主搜索、组合这些动作去解决机器学习工程任务。结果：同一个模型，用上这套自我进化机制后，在标准评测上的成绩从 39.39% 涨到 60.61%，再配上更强的搜索策略能到 71.21%，超过了 GPT-5.5 加 Codex 的组合，逼近 GPT-5.6 和 2.8T 参数的 Kimi K3。更关键的是，这套方法在没见过的任务上也有效：换上新模型，匹配率从 50% 提到 70%；换上新搜索框架，从 20% 提到 50%。它不是你明天能用的工具，但它第一次证明：AI 可以像程序员迭代代码一样迭代自己，而且这条路真的走得通。
📄 原文摘要(英文)
Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifiable task environments with execution feedback (OpenMLE-Gym), operator learning (OpenMLE-RL), and long-horizon search (OpenMLE-Evo). On this stack we post-train Frontis-MA1 (35B) as a meta-evolution agent for MLE, aligning post-training and inference around four atomic program-evolution operators (Draft, Improve, Debug, Crossover): the same operators are trained via execution-grounded SFT and RL on data deduplicated against all evaluation benchmarks, then composed into long-horizon search, coupling learning and evolution in a single loop. On MLE-Bench Lite under a 12-hour per-task budget on one RTX 4090 capped at 12 GB VRAM, Frontis-MA1 (35B) improves Medal Average from 39.39% to 60.61% over its base model with OpenMLE-Evo, and reaches 71.21% with OpenMLE-Evo-Max (benchmark-independent experience priors and asynchronous search), exceeding GPT-5.5 + Codex and approaching GPT-5.6 Sol and the 2.8T Kimi K3. On held-out NatureBench Lite, both components transfer: with the framework fixed, swapping in the trained model raises Match-SOTA from 50% to 70%; with the model fixed, swapping in OpenMLE-Evo raises it from 20% to 50%. We release the model weights and the full OpenMLE stack to enable reproducible research on executable AI4AI toward RSI. Code: https://github.com/FrontisAI/OpenRSI