ScienceBuddy：通过迭代优化提示词与模型提升科学Agent准确率
Don't pick between better prompts and a better model: improving both in turns li…
Agent开发者必读，这篇论文给出了可落地的“提示词-模型”交替优化工作流，实测效果显著，值得收藏参考。
Don't pick between better prompts and a better model: improving both in turns lifted a science agent from 42.2% to 73.3% accuracy.
不要在选择更好的提示词和更好的模型之间二选一：两者交替改进使科学代理的准确率从 42.2% 提升至 73.3%。
Researchers correct AI agents all the time, but fixing an answer in chat doesn't make the agent better at the next task.
研究人员一直在纠正 AI 代理，但在聊天中修正答案并不能让代理在下一个任务中表现得更好。
Correcting an AI agent in chat, the fix usually dies with the conversation.
在聊天中纠正 AI 代理时，修正通常随对话结束而消失。
ScienceBuddy, an AI assistant for scientists, turns that feedback into scored test tasks. Then it takes turns: rewrite the agent's prompts and skills, retrain the model, and repeat.
ScienceBuddy 是一款面向科学家的 AI 助手，它将反馈转化为带评分的测试任务。然后它交替进行：重写代理的提示词和技能、重新训练模型，并重复此过程。
With a small 4B model on biology tasks, prompt and skill changes alone raised accuracy from 31.1% to 51.1%. Retraining alone also helped, raising the share of problems solved within 4 tries from 48.3% to 67.8%.
在生物学任务上使用一个小型 4B 模型，仅通过提示词和技能的更改就将准确率从 31.1% 提高到了 51.1%。仅重新训练也带来了帮助，将 4 次尝试内解决问题的比例从 48.3% 提升到了 67.8%。
If you build agents, save every user correction as a test, and keep upgrading your prompts and your model in turns.
如果你正在构建代理，请将每次用户纠正保存为测试，并交替升级你的提示词和模型。
– arxiv. org/abs/2609.17523v1
– arxiv.org/abs/2609.17523v1
Title: "ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents"
标题：《ScienceBuddy：交互式科学代理的递归式自我改进》
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力