Andrej Karpathy

本期周刊以‘AI次方变革’为核心隐喻，系统串联杨斌的组织心智重构、Karpathy的Software 3.0范式、Demis的AGI三缺口，揭示AI已从‘+AI’工具叠加迈入底数质变驱动的指数级重构阶段。

The bottleneck in engineering management has shifted from coding speed to verification and collaboration. Fiona Fung from Anthropic proposes restructuring processes, cutting outdated roadmaps and design docs, and treating code as the single source of truth. Managers must return to writing code, adopt JIT planning, shift QA left to automation, and focus on onboarding time and PR lifecycle rather than vanity metrics.

Recursive self-improvement is driving AI models into a 'new Moore’s Law' era; SpaceX is building a trillion-dollar 'Elon Web Services' ecosystem via Starlink and Colossus compute; American fear of AI stems from job displacement anxiety, CEO communication failures, and regulatory misalignment—not the technology itself.

ASD-STE00航空维修标准被用于优化AI回复规则，通过明确条件和操作顺序提升工程师与AI协作的准确性。

ASD-STE100规范通过严格写作规则提升技术文档清晰度，被用于优化大模型Prompt效果。

Andrej Karpathy展示了一种通过16,200次经纬度判断实验验证LLM地理识别能力的方法，结果图像显示模型表现优异。

Andrej Karpathy展示Opus 5模型生成复杂Three.js代码的能力，揭示LLM在游戏世界构建中的潜力与局限。

混合架构（云端大模型+本地小模型）将成为AI发展主流，兼顾隐私与效率。

2026年AGI时间线因Claude Opus 4.5突破发生剧变，专家观点从悲观转向乐观，但核心分歧仍存。

Agent Wikis通过预处理生成知识库降低查询成本，四家团队实现相同架构但各有差异，系统分三层结构并包含摄入/查询/检查三类操作。

与LLM交互时，通过语音输入的长时间无序思考可提升输出质量。

autoresearch是一个开源工具，允许AI代理自动优化LLM训练过程，通过实验找到模型改进点。

设计师应主动定义与AI协作的实践，而非被动接受“vibe coding”标签，强调判断力与意图性。

Claude Fable 5 是 Claude Mythos 的升级版，性能显著提升，尤其在复杂任务和长篇问题解决上表现突出。

文章强调了通过设计自主循环系统来提升AI代理效率，而非依赖人工干预。

Cursor achieved a $20 billion valuation in 3 years, scaling ARR from 0 to $2 billion with almost no marketing spend. Its growth engine combines product design and user habit migration: forking VS Code to reduce friction, using Tab and Composer to eliminate coding friction, and leveraging KOLs like Karpathy to spread 'vibe coding' virally, creating a 'can't go back after one week' retention loop.

Anthropic has appointed Andrej Karpathy and other prominent figures as MTS (Technical Employee), sparking controversy. MTS is a flat position title designed to prevent headhunters from poaching talent, enhance cultural identity, and break down departmental walls. However, some comments suggest that this approach may be superficial and potentially disadvantageous to young engineers in their career development.

Karpathy points out that 90% of the AI coding bill is spent on unnecessary context, and optimizing context usage and routing strategies can significantly reduce costs.

This article explores the necessity and practical methods of transitioning from vibe coding to spec-driven development, emphasizing the advantages of the latter in team collaboration and project management.

Karpathy在Sequoia Ascent 2026的炉边谈话中强调了LLMs超越加速现有技术的新领域，如无需代码的应用menugen、安装.md技能代替.sh脚本，以及LLM知识库处理非结构化数据的能力。

Vibe coding builds software applications by completely ignoring the code generated by LLMs, only developing through prompts and feedback, but poses risks in maintainability, correctness, and security. This technique suits non-programmers creating disposable applications but is unsuitable for production environments.

Andrej Karpathy讨论了AI编程的变革，从“vibe coding”到软件3.0时代，如何通过上下文而非传统编码实现功能，强调可验证性在AI自动化中的作用，并提出智能体工程的新视角，同时提醒保持人类判断与立场的重要性。

Andrej Karpathy joined Anthropic to lead AI-powered pre-training research using Claude, marking a strategic shift toward autonomous AI development.

AI hype drives capital market frenzy, with Anthropic reaching a $90 billion valuation, while traditional SaaS companies face transformation pressures and AI infrastructure/IPOs become key topics.

The article introduces the concept of 'agentic engineering', emphasizing that developers must master AI collaboration skills to meet future development challenges.

Andrej Karpathy proposes a paradigm shift: neural networks as the primary computational substrate, with CPUs relegated to co-processors — redefining hardware/software hierarchy in AI-native systems.

Andrej Karpathy提出Software 3.0范式：软件开发正从手写代码（S1.0）和机器学习建模（S2.0）转向以提示工程驱动智能体协作的新阶段，核心能力是设计提示、调试行为与编排AI工作流。

探索AI代理与人类协作的通用接口设计，通过Notion-like页面整合LLM生成内容，但缺乏系统性技术方案。

提示工程逐渐被图结构取代，成为AI系统的核心层。Andrej Karpathy在讲座中指出，LLMs→Prompts→Agents→Graphs是技术演进路径，图是最终的生存形态。

Andrej Karpathy joins Anthropic to lead Claude pre-training research, focusing on AI self-improvement capabilities (RSI), potentially accelerating AI's independent R&D progress.