我们已越过 AI 起飞的事件视界，GPT-4 和 o3 等系统在许多领域超越人类智能。 We are past the event horizon of AI takeoff, with systems like GPT-4 and o3 surpassing human intelligence in many areas.
AI 将推动更快的科学进步和生产力提升，极大改善生活质量。 AI will drive faster scientific progress and productivity, vastly improving quality of life.
2025 年出现能进行真正认知工作的智能体，2026 年预计出现能发现新见解的系统，2027 年出现能完成现实世界任务的机器人。 Agents capable of real cognitive work emerged in 2025, with novel insights expected by 2026 and real-world robots by 2027.
2030 年代智能和能源将变得极其丰富，消除进步的基本限制因素。 Intelligence and energy will become wildly abundant in the 2030s, removing fundamental limiters on progress.
通过 AI 辅助研究和自动化基础设施建设，递归式自我改进正在进行中。 Recursive self-improvement is underway via AI-assisted research and automated infrastructure buildout.
解决对齐问题并广泛分配超级智能对于最大化收益和最小化风险至关重要。 Solving alignment and distributing superintelligence widely are critical for maximizing upside and minimizing downside.
局限 · Limitations
奇点被描述为'温和'且渐进的，但实际速度和社会影响仍不确定。 The singularity is described as 'gentle' and gradual, but the actual pace and societal impact remain uncertain.
对齐问题被承认但未完全解决；社交媒体信息流被引用为对齐失败的例子。 Alignment problem is acknowledged but not fully solved; social media feeds are cited as a misaligned example.
提及了就业替代和社会契约变化等困难，但未提供具体解决方案。 Job displacement and social contract changes are mentioned as hard parts, but concrete solutions are not provided.
鉴于当前限制，2030 年代智能和能源丰富的预测可能过于乐观。 The prediction of abundant intelligence and energy by 2030s may be overly optimistic given current constraints.
文章侧重于收益，淡化了滥用、不平等或失控等潜在风险。 The essay focuses on benefits and downplays potential risks like misuse, inequality, or loss of control.