一位著名计算机科学家认为，围绕 AI 和 AGI 的危言耸听和过度乐观叙事使年轻人士气低落，分散了他们对机器学习与系统构建真正机会的注意力。
A prominent computer scientist argues that alarmist and exuberant narratives around AI and AGI demoralize young people, distracting from real opportunities in machine learning and systems building.
要点 · TL;DR
AI 炒作分散了对真正问题的关注，伤害了年轻创新者。 AI hype distracts from real problems and harms young innovators.
不确定性量化对于负责任的 AI 系统至关重要。 Uncertainty quantification is essential for responsible AI systems.
经济学和统计学必须与机器学习结合，以构建更好的 AI。 Economics and statistics must integrate with machine learning for better AI.
核心观点 · Key points
AI 应该用于辅助人类，而不是用超级智能取代人类。 AI should be about aiding humans, not replacing them with superintelligence.
当前的 AI 炒作分散了对医疗和交通等实际问题的关注。 Current AI hype distracts from real problems like healthcare and transportation.
不确定性量化至关重要；LLM 缺乏这一点，需要统计方法。 Uncertainty quantification is critical; LLMs lack it and need statistical methods.
经济学和统计学必须与机器学习结合，以实现负责任的 AI。 Economics and statistics must integrate with machine learning for responsible AI.
数据市场需要激励结构来尊重隐私并奖励贡献者。 Data markets need incentive structures to respect privacy and reward contributors.
反共识 · Contrarian takes
AGI 是一个公关术语，扭曲了研究并让年轻人感到沮丧。 AGI is a PR term that distorts research and demoralizes young people.
构建不理解系统的做法没问题，但需要经济上的保障措施。 Building systems without understanding them is fine, but requires economic safeguards.
AI 灭绝风险被夸大；真正的危险是劳动力和资本问题。 The risk of AI extinction is overblown; real dangers are labor and capital issues.
LLM 并不智能；它们模仿人类文本，没有推理或理解。 LLMs are not intelligent; they mimic human text without reasoning or understanding.
市场比任何自上而下设计的 AI 系统更能减少不确定性。 Markets reduce uncertainty better than any AI system designed top-down.
本期章节 · Chapters（共 26）
引言与背景Introduction and Background
批判 AI 炒作与 AGICritique of AI Hype and AGI
批判硅谷思维Critique of Silicon Valley mindset
智能的社会科学视角Social science perspective on intelligence
非批评者，旨在改进 AINot a critic, but aiming to improve AI
第一步谬误与系统局限First step fallacy and limitations of current systems
回应多智能体方法Response to multi-agent approach
颠覆与隐喻Disruption and metaphors
机械可解释性与系统理解Mechanistic interpretability and understanding systems
行为主义与 AlphaFold 示例Behaviorism and AlphaFold example
AlphaFold 预测中的科学假设偏差Bias in AlphaFold predictions for scientific hypotheses
理解 vs.优化Understanding vs. Optimization
多学科视角Multi-disciplinary Perspectives
数据市场中的激励机制Incentives in Data Markets
将市场建模为动态系统Modeling markets as dynamical systems
数据 vs.社会知识与谦逊Data vs. social knowledge and humility
抽象、市场与自下而上涌现Abstraction, markets, and bottom-up emergence
文化与抽象Culture and Abstraction
批判 AI 末日论及其对年轻人的影响Critique of AI doomerism and its impact on young people
AI 作为增强而非替代AI as augmentation, not replacement
自主系统与人类监督Autonomous systems and human oversight
批判 AI 炒作与思想领袖Critique of AI hype and thought leaders
博弈论 vs.机制设计Game Theory vs Mechanism Design
不确定性量化与 E 值Uncertainty Quantification and E-values
任意时间推断与统计契约理论Anytime Inference and Statistical Contract Theory