https://guijiagi.com/tags/%E5%8F%AF%E6%89%A9%E5%B1%95%E7%9B%91%E7%9D%A3/Recent content in 可扩展监督 on 硅基 AGI · 智能体学习与测评Hugozh-cn本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权Thu, 16 Jul 2026 10:23:00 +0800https://guijiagi.com/posts/b2-f761472e/Thu, 16 Jul 2026 10:23:00 +0800https://guijiagi.com/posts/b2-f761472e/探讨AGI时代AI对齐面临的核心开放问题，涵盖可扩展监督、机械可解释性与可纠正性挑战https://guijiagi.com/posts/ai-alignment-frontier-scalable-oversight/Tue, 30 Jun 2026 11:25:00 +0800https://guijiagi.com/posts/ai-alignment-frontier-scalable-oversight/深入探讨2026年AI对齐技术的前沿方向，包括可扩展监督、AI反馈（RLAIF）、辩论机制等，分析技术原理与实践挑战https://guijiagi.com/posts/ai-alignment-2026-rlhf-to-constitutional/Tue, 30 Jun 2026 10:15:00 +0800https://guijiagi.com/posts/ai-alignment-2026-rlhf-to-constitutional/系统梳理2026年AI对齐技术的完整演进路径，从RLHF到Constitutional AI再到可扩展监督，分析各方案的优劣与适用场景https://guijiagi.com/posts/superalignment-2026-controlling-ai-beyond-human-intelligence/Sun, 28 Jun 2026 10:11:00 +0800https://guijiagi.com/posts/superalignment-2026-controlling-ai-beyond-human-intelligence/2026 年超级对齐研究进展：从可扩展监督到机械可解释性，探讨当 AI 能力超越人类时的安全控制框架