三个观察

Three Observations

摘要 · Abstract

管理编辑、发布、注销和登录的观察。

ManageEdit PostLogoutLogin[](https://posthaven.com/)

AGI 的智能与资源（训练算力、数据、推理算力）的对数成正比。
AGI intelligence scales as log of resources (training compute, data, inference compute).
使用特定水平 AI 的成本每 12 个月下降约 10 倍，快于摩尔定律。
Cost to use a given AI level drops ~10x every 12 months, faster than Moore's Law.
线性增长的智能带来的社会经济价值是超指数级的。
Socioeconomic value of linearly increasing intelligence is super-exponential.
AI 智能体将成为虚拟同事，可扩展到数千或数百万个。
AI agents will become virtual co-workers, scaling to thousands or millions.
AGI 将成为人类意志的最大杠杆，使个人影响力更大。
AGI will be the biggest lever for human willfulness, enabling greater individual impact.
开源和个体赋权是平衡安全与自由的关键。
Open-sourcing and individual empowerment are key to balancing safety and freedom.

缩放定律可能在极端资源水平或新任务上失效。
Scaling laws may break down at extreme resource levels or for novel tasks.
成本下降趋势可能因物理或算法限制而放缓。
Cost reduction trends may slow due to physical or algorithmic limits.
超指数价值增长假设没有颠覆性的社会或监管变化。
Super-exponential value growth assumes no disruptive societal or regulatory changes.
AI 智能体可能在需要创造力或长期规划的任务上失败。
AI agents may fail on tasks requiring creativity or long-term planning.
若无政策干预，AGI 收益分配不均可能加剧不平等。
Uneven distribution of AGI benefits could worsen inequality without policy intervention.