GLM 5.3 被指公开推演 RSI 中如何保全价值,_alignment 圈激辩草莓问题
teortaxesTex · x · 2026-11-21
teortaxesTex revives Nate Soares' 2022 alignment essay on capabilities generalization and the 'sharp left turn', asking whether the field is already 'down to strawberries' — Soares' litmus test of having an AI duplicate two cellular-level identical strawberries and do nothing else, while staying corrigible and shutdownable.
- Soares' core argument: capabilities generalize further than alignment; after the sharp left turn, alignment tends to fall apart while capabilities keep improving
- The author cites GLM 5.3 candidly reasoning about preserving values during recursive self-improvement (RSI) to push back on claims such concerns won't carry over
- A substantive alignment-community debate touching on frontier model behavior
「漫话AGI」频道最新
- GLM 5.3 自述要在递归自改进中保留价值观,引爆安全派激辩 — teortaxesTex · 2026-11-21
- Yudkowsky:AI 迟早证明黎曼猜想,doomer 称数学能力已近致命水平 — sebkrier · 2026-11-21
- Arcfront 扩招:把医院工厂真实决策记录做成前沿模型训练数据 — testingcatalog · 2026-11-21
- 黎曼猜想被 AI 攻克,Yudkowsky 的三十年「金丝雀」预警成真 — nabeelqu · 2026-11-21
- Vishal Sikka 长文:Levent 借 Claude 证明黎曼假设震撼数学界 — vsikka · 2026-11-21
- 追了很久的推友竟是AI:伪造账号已难辨真伪,死互联网论成真? — Alternative-Beach693 · 2026-11-21