达里奥和丹妮拉·阿莫代伊离开 OpenAI 创立 Anthropic，如今这家万亿级 AI 公司正竞相构建安全的超级智能，同时应对伦理困境与政府合作。
Dario and Daniela Amodei left OpenAI to found Anthropic, now a trillion-dollar AI company racing to build safe superintelligence while navigating ethical dilemmas and government partnerships.
要点 · TL;DR
规模定律是 AI 进步的关键驱动力，而非算法突破。 Scaling laws are the key driver of AI progress, not algorithmic breakthroughs.
企业级聚焦符合安全价值观，避免消费者参与成瘾。 Enterprise focus aligns with safety, avoiding consumer engagement addiction.
AI 将淘汰许多白领工作，需要全民基本收入和新角色。 AI will eliminate many white-collar jobs, requiring UBI and new roles.
核心观点 · Key points
缩放定律推动了 AI 进步；即使算法不变，更多数据和算力也能提升模型。 Scaling laws drove AI progress; more data and compute improve models even without algorithmic changes.
企业级业务与安全价值观一致，避免了消费者领域的成瘾和低质内容。 Enterprise focus aligns with safety values, avoiding consumer engagement addiction and slop.
AI 将淘汰许多白领工作，社会需要寻找新角色并制定政策，如全民基本收入。 AI will eliminate many white-collar jobs, requiring society to find new roles and policies like UBI.
负责任的 AI 开发需要发布前测试、审计和监管，而非放任自流。 Responsible AI development requires pre-release testing, auditing, and regulation, not laissez-faire.
军事 AI 应保持人类在决策环中，防止灾难性错误。 Military AI should keep humans in the loop for final decisions to prevent catastrophic mistakes.
Anthropic 致力于成为安全第一的公司，通过行动而非空话赢得信任。 Anthropic aims to be a safety-first company, earning trust through actions, not just words.
反共识 · Contrarian takes
AI 可能导致高 GDP 增长伴随高失业和不平等，这是一种罕见的组合。 AI could cause high GDP growth alongside high unemployment and inequality, a rare combination.
软件行业将变得更大，尽管一些现有企业可能贬值或倒闭。 Software industry will grow larger despite some incumbents losing value or going out of business.
Anthropic 的 CEO 认为 AI 总体上降低战争可能性，与升级的担忧相反。 Anthropic's CEO believes AI on balance makes war less likely, contrary to fears of escalation.
私营部门主导 AI 发展是危险且不稳定的；政府应发挥作用。 Private sector leading AI development is dangerous and unstable; government should have a role.
Mythos 模型尽管造成商业损失仍被限制发布，优先考虑安全而非利润。 Mythos model was withheld from public release despite commercial harm, prioritizing safety over profit.
Dario Amodei 认同 Leo Szilard 而非奥本海默，视奥本海默为失败案例。 Dario Amodei identifies with Leo Szilard, not Oppenheimer, seeing Oppenheimer as a failure case.
本期章节 · Chapters（共 31）
引言与背景Introduction and Setting
达里奥·阿莫迪与 Anthropic 崛起Dario Amodei's Profile and Anthropic's Rise
阿莫迪兄妹组合The Amodei Sibling Duo
Anthropic 使命与挑战Anthropic's Mission and Challenges
达里奥谈 AI 发展Dario's Perspective on AI Growth
童年与成长背景Childhood and Background
职业路径与离开 OpenAICareer Path and OpenAI Departure
创立 AnthropicFounding Anthropic
安全使命与 Claude 宪法Anthropic's safety mission and Claude's constitution
商业模式：企业 vs 消费者Business model: enterprise focus vs consumer
对软件行业与就业影响Impact on software industry and job displacement
鲍里斯·切尔尼与 Claude CodeBoris Cherney's journey and Claude Code creation
编程代理愿景Coding Agent Vision
开发者大会与增长Developer Conference and Growth
对工程师与就业影响Impact on Engineers and Jobs
AI 风险与信息传达On AI risks and messaging
AI 对就业的影响AI impact on jobs
Anthropic 领导结构Anthropic's leadership structure
地缘政治与出口管制Geopolitics and export controls
反战立场与国防合同Anti-war stance and defense contracts
Anthropic 与五角大楼冲突Anthropic's Conflict with the Pentagon
AI 战争与全球稳定AI Warfare and Global Stability
应对压力与内部沟通Handling Pressure and Internal Communication
Mythos 简介Introduction to Mythos
Glasswing 项目与访问控制Project Glasswing and Access Control