「强但非超人」AI 意味着漫长黄金时代而非奇点,引激烈论战
teortaxesTex · x · 2026-10-22
shinboson argues people underthink the implications of generally exceptional but not generally superhuman AI: incredible economic impact, but likely a long golden age rather than a singularity. Charts from TrackingAI show offline IQ scores climbing to 136-151, and the number of people you'd need to find one smarter than AI rising from 61 today.
The thread erupts: autumn proposes a third scenario — "AI makes everything worse forever but not enough to kill us quickly" — citing vibe-coded software rot and unfixable vulnerability discovery. Critics call this guild privilege, mocking decades of failed traditional ML academia; autumn counters that his concerns are material, not purist.
「漫话AGI」频道最新
- 「研究生若知道训练 SOTA LLM 真正有效的招，会气疯」 — code_star · 2026-10-22
- AI 网络攻击代理已具国家级水平：靠供应链式跳板绕过隔离网络 — harris_edouard · 2026-10-22
- 单向数据二极管：关键基础设施为何还没被 AI 网络代理攻破 — harris_edouard · 2026-10-22
- Nathan Calvin 再批 Anthropic:正疯狂冲刺递归自我改进的超级智能 — wfithian · 2026-10-22
- 神经外科住院总与 AI 交流：临床医生自主权正被侵蚀 — zakkohane · 2026-10-22
- AI 时代最黏性的资产不是知识，而是可验证的信誉记录 — curious_vii · 2026-10-22