Google DeepMind 提出 Evo2 框架：让智能体的脚手架与模型权重共同进化
Banger paper from Google DeepMind on co-evolving an agent's harness and its weights. (bookmark it) ...
DeepMind 出了个叫 Evo2 的框架，让 agent 自己改自己的脚手架再拿日志训练，GPT-5-mini 光换脚手架就从 24.4% 涨到 48.3%，做 agent 的都该看看。
Google DeepMind 发布论文，提出 Evo2 框架实现智能体脚手架（harness）与模型权重的共同进化。同一个策略承担 Proposer 和 Reflector 两个角色：Proposer 编写脚手架编辑方案，Reflector 诊断每次编辑带来的变化，这些编辑和结果再经 SFT 加交错 GRPO 成为两个角色的训练数据。Evo2 在 SWE-bench Verified 达到 48.3%，在 Polyglot 达到 82.1%，分别高于最强基线的 41.0% 和 62.6%。仅进化脚手架就能把 GPT-5-mini 智能体在 SWE-bench Verified 上的成绩从 24.4% 提到 48.3%，且进化后的脚手架可直接迁移到其他基准。
Banger paper from Google DeepMind on co-evolving an agent's harness and its weights. (bookmark it) ...
Banger paper from Google DeepMind on co-evolving an agent's harness and its weights. (bookmark it) If you already log your harness changes and results, those logs can become training data for the model running inside the harness. This is a nice paper from Google showing how to do this co-evolution. Specifically, they propose Evo2, a framework for weight-harness co-evolution. How it works: One policy plays two roles. As a Proposer, it writes edits to its own harness, and as a Reflector, it diagnoses what each edit changed. Those edits and their outcomes then become training data for both roles, using SFT followed by interleaved GRPO. The result: Evo2 reaches 48.3% on SWE-bench Verified and 82.1% on Polyglot, against 41.0% and 62.6% for the strongest baselines. The harness part alone moves a GPT-5-mini agent from 24.4% to 48.3% on SWE-bench Verified. The evolved harnesses also transfer to other benchmarks without further evolution. Paper: arxiv.org/abs/2610.13025 Chat with Paper: academy.dair.ai/papers/evo-2-e… 💬 23 🔄 13 ❤️ 91 👀 4577 📊 41 ⚡
