菜单

关于 🐙 GitHub
arXiv 提交日期: 2025-12-05
📄 Abstract - AI & Human Co-Improvement for Safer Co-Superintelligence

Self-improvement is a goal currently exciting the field of AI, but is fraught with danger, and may take time to fully achieve. We advocate that a more achievable and better goal for humanity is to maximize co-improvement: collaboration between human researchers and AIs to achieve co-superintelligence. That is, specifically targeting improving AI systems' ability to work with human researchers to conduct AI research together, from ideation to experimentation, in order to both accelerate AI research and to generally endow both AIs and humans with safer superintelligence through their symbiosis. Focusing on including human research improvement in the loop will both get us there faster, and more safely.

顶级标签: agents theory model training
详细标签: human-ai collaboration ai safety superintelligence research paradigm reinforcement learning 或 搜索:

协同改进:迈向更安全、更可实现的超级智能之路 / AI & Human Co-Improvement for Safer Co-Superintelligence


1️⃣ 一句话总结

本文提出并论证了“协同改进”作为替代“自我改进”的AI发展新范式,主张通过人类研究者与AI系统在整个研究周期内深度协作,以更快、更安全地实现对人类有益的超级智能。


2️⃣ 论文创新点

1. 从“自我改进”到“协同改进”的范式转变

2. 人机协同研究(Co-improvement)框架

3. 双向协同改进与协同超级智能愿景

4. 管理开放性原则


3️⃣ 主要结果与价值

结果亮点

实际价值


4️⃣ 术语表

源自 arXiv: 2512.05356