Ryan Greenblatt
立场时间线 4 条 · 4 个议题
同一议题按时间排列，说法变化一目了然。
Greenblatt认为AI研发具有高可验证性与迭代性，一旦AI匹配顶尖人类专家就可能触发递归自我改进的反馈循环，一年内实现四到五年的AI进展。
“我的中位数预期是，AI可能在一年内取得相当于四到五年的人类AI进展。（My median expectation is something like four or five years of AI progress in a single year.）”
来源访谈 →他认为机器学习相对数学是浅层领域，更依赖爬山式累加创新而非深邃抽象，因此AI更容易通过大量可验证子任务训练获得核心研发直觉。
“我认为，相对于数学，机器学习是一个非常浅层的领域。（I think ML is a very shallow domain relative to math.）”
来源访谈 →他认为即使AI在政治手腕等软技能上迁移不足，只要在芯片、工厂、机器人等硬科技研发上达到超人水平，就足以引发全球性变革。
“如果AI真的非常擅长芯片研发、建造晶圆厂、协调工厂、设计和操作机器人，同时也擅长AI研发……我认为那已经是一个非常疯狂的局面了。（If the AIs were really, really good at chip R&D, building fabs, orchestrating factories, designing robots, operating robots, and also at AI R&D... I think that would already be a pretty crazy situation.）”
来源访谈 →他认为Claude宪法等现有规范并未真正把用户作为AI首要的倡导者与守护者，他更倾向于明确AI是用户个人倡导者的宪法。
“我更倾向于的宪法是，它明确指出Claude是你的倡导者。（The thing I would prefer would be a constitution that says Claude is your advocate.）”
来源访谈 →金句墙 4 条
“My median expectation is something like four or five years of AI progress in a single year.”
我的中位数预期是，AI可能在一年内取得相当于四到五年的人类AI进展。
AI自动化研发后：递归自我改进的可能与风险 · 2026/8/11“I think ML is a very shallow domain relative to math.”
我认为，相对于数学，机器学习是一个非常浅层的领域。
AI自动化研发后：递归自我改进的可能与风险 · 2026/8/11“If the AIs were really, really good at chip R&D, building fabs, orchestrating factories, designing robots, operating robots, and also at AI R&D... I think that would already be a pretty crazy situation.”
如果AI真的非常擅长芯片研发、建造晶圆厂、协调工厂、设计和操作机器人，同时也擅长AI研发……我认为那已经是一个非常疯狂的局面了。
AI自动化研发后：递归自我改进的可能与风险 · 2026/8/11“The thing I would prefer would be a constitution that says Claude is your advocate.”
我更倾向于的宪法是，它明确指出Claude是你的倡导者。
AI自动化研发后：递归自我改进的可能与风险 · 2026/8/11