Archive
A journey through everything I've published, organized over time.
过去很多人赚钱，靠的是需求增长和市场空位。现在大部分标准化需求已经被成熟体系占住，普通人的机会不再在大赛道里，而在新的交换缝隙里。
从现实、概念、语言到 AI 的层层抽象出发，讨论大语言模型为何强大、为何会失真，以及人在 AI 时代为什么仍要负责把抽象重新接回现实。
A reflection on reality, concepts, language, and AI: why large language models are powerful, where abstraction introduces distortion, and why humans still need to reconnect symbols back to the real world.
When AI services blossom everywhere, who connects them? A speculative look at the missing infrastructure—the AI service marketplace—and the protocols, economics, and privacy challenges it must solve.
基于政治经济学与技术史的交叉视角，系统分析AI驱动的生产力革命如何重塑稀缺结构与协调能力，进而探讨制度演化的三条路径与临界条件。
A warm yet rigorous narrative that threads together cold political-economy logic and the history of technology: how AI and automation are re-engineering scarcity and coordination, and why institutions are likely to evolve into a hybrid spectrum of algorithmic capitalism, technocratic welfare states and post-scarcity horizons.
Part (3) brings the layered paradigm into organizations: once AI is expected to act, the winning layer is an auditable control plane that turns workflows and agents into governed operations.
AI makes production cheap. The new scarcity becomes selection, trust, and responsibility—and that shift rewrites what people want, what companies build, and which jobs survive.
当一个臃肿、昂贵、隐私纠缠的系统依然被支持，信号不是`它完美`，而是`行动型 AI 正被加速推入现实`。代价是什么？失业只是第一波。
When a bloated, expensive, privacy-entangled system is still supported, the signal isn't 'it's perfect'—it's 'action-oriented AI is being accelerated into reality.' What's the cost? Unemployment is just the first wave.
Using OpenClaw as a clue, this essay argues personal AI will be won by a control plane that can route intent, enforce boundaries, and make actions auditable across the surfaces where life actually happens.
As LLM apps hit the context wall, the next phase of AI software shifts from chat-first tools to a governable service ecosystem built around discovery, routing, and accountability.
A quiet shift is underway: as AI lowers the cost of building software, the attention-extraction model weakens and a user-aligned personal AI becomes plausible—reviving cognitive sovereignty.
Weather alerts deliver information, not decisions. This essay argues for a proactive, human-centered AI layer that turns extreme weather warnings—cold, heat, blizzards, tsunamis—into quiet, personalized action plans, with governance to avoid alert fatigue.
Open worlds don’t fail because they’re small. They fail because content throughput can’t match player curiosity. This essay explains—technically, but plainly—how generative AI reshapes assets, NPCs, and UGC into constrained systems that keep a world coherent while it keeps growing.
From DeepSeek's global disruption to MiniMax's IPO and Alibaba's Apple partnership—here is the data-driven reality of China's AI market in 2025.
An exploration of future AI product forms centered on Tobby OS: eliminating cognitive friction with semantic-to-structure intelligence, passive assistance, intent routing, and humane interfaces that turn everyday language into reliable outcomes.
After Big Tech’s full entry, do personal/small-team AI SaaS still have room to survive? This piece focuses on market structure, moat types, entry space, and the true difficulty of going from 0 to 1, offering a professional, verifiable judgment framework—not an execution plan.
In the next 3–5 years B2B will scale first, while long term B2C is more disruptive. Why this happens, how to plan, and how B2B2C bridges both.
基于 MIT Technology Review、IDC、McKinsey、NVIDIA GTC、Reuters 等权威来源的系统性综述：2026 年 AI 将在算力与能效、智能体与多模态视频、边缘与行业闭环、治理与绿色 AI 四条主线加速成熟，走向“环境智能层 + 个人/企业 OS”。
A research-backed annual review of AI heading into 2026: compute and efficiency, agentic systems, multimodal/generative video, edge inference, industry closures, governance and green AI.
Revue annuelle vers 2026 : ressources de calcul et efficacité, systèmes agentiques, multimodal/vidéo générative, inférence en périphérie, fermetures industrielles, gouvernance et IA verte.
Jahresrückblick Richtung 2026: Rechenressourcen und Effizienz, agentische Systeme, multimodal/generative Videos, Edge‑Inference, industrielle Abschlüsse, Governance und Green AI.
2026年に向けたAI年次レビュー：計算資源と効率、エージェント化、マルチモーダル/生成動画、エッジ推論、産業閉ループ、ガバナンスとグリーンAI。
2026년 AI 연례 리뷰: 계산 자원과 효율, 에이전트화, 멀티모달/생성 비디오, 엣지 추론, 산업 폐루프, 거버넌스와 그린 AI.
Годовой обзор к 2026: вычислительные ресурсы и эффективность, агентные системы, мультимодальность/генеративное видео, пограничный вывод, отраслевые замыкания, управление и зелёный ИИ.
Revisión anual hacia 2026: recursos de cómputo y eficiencia, sistemas agentic, multimodal/video generativo, inferencia en el borde, cierres industriales, gobernanza y IA verde.
面向 2026 的年度回顧：算力與能效、智能體、跨模態/生成影片、邊緣推理、產業閉環、治理與綠色 AI。
An SCI-style exposition of AI’s end-state and near-term transformation: a dual-layer architecture of an Environmentalized Intelligence Layer plus a Personal OS, grounded in structured world models, neuro-symbolic fusion, embodied loops, and trustworthy governance.
A field guide to building system-level agents: perception, memory, planning, tools, execution, and evaluation.
From assistance to collaboration: multimodal fusion of imaging–pathology–clinical data and workflow redesign.
Key hardware and system trade‑offs: general‑purpose GPUs, specialized NPUs/ASICs, near‑memory, and distributed orchestration.
A systemic redesign from teaching to assessment: diagnostic evaluation, dynamic content, and orchestrated learning paths.
From product to platform: developer network effects, supply‑chain coordination, and governed openness.
Upgrade prompts into interaction protocols: roles, state, tools, memory, and evaluation loops.
In-depth analysis of Google Research's latest PH-LLM (Physiological Health Large Language Model), exploring its practical applications in Fitbit, technical architecture, and revolutionary impact on personal health management.
Deep analysis of how OpenAI is building safer, more responsible AI systems for sensitive dialogue scenarios - from GPT-5 System Card to ChatGPT practical implementations
ByteDance Seed3D 1.0 delivers high-fidelity asset generation, native physics compatibility, and scalable scene composition—from a single image to robotics-ready 3D assets for simulators like Isaac Sim.
When big players pave highways, small AI SaaS wins with controllable experiences and outcome‑selling. Four case studies and a PEAL playbook.
An in-depth analysis of the survival crisis facing AI startups, revealing the essential path to building sustainable competitive advantages in the era of technological democratization
A practical roadmap for individual developers to start profitable AI businesses. Learn how to leverage existing AI tools, build MVPs quickly, and scale without a team. Build your AI SaaS.
A comprehensive analysis of how artificial intelligence will evolve by 2026, examining technological differentiation, geopolitical divisions, and the shift from parameter wars to practical applications.
Official analysis of DeepSeek-V3.2-Exp's groundbreaking Sparse Attention mechanism, massive API cost reductions, and open-source ecosystem expansion.
Alibaba's DAMO Academy unveils Animate 2.5, a groundbreaking AI video generation model that rivals international standards with 60-second video capabilities, precise motion control, and character consistency—marking China's competitive entry into the AIGC video arena.
An in-depth analysis of Google Learn Your Way tool, exploring how LearnLM technology enables truly personalized education, the science behind its 11% learning improvement, and its profound impact on the future of education.
When information floods become tsunamis, our attention becomes the scarcest resource of our time. AI isn't a lifeboat—it's a more powerful current that can either carry us to new continents or drag us into the depths.
Current AI's remarkable success masks fundamental limitations. Explore the revolutionary architectures—neurosymbolic AI, embodied intelligence, and dual-system thinking—that will define the next era of artificial intelligence. AI Talk.
Beyond content generation: concrete product strategies for agentic workflows, domain-trained models, and natural multimodal UX—plus pricing for uncertainty, data moats, and a 90‑day plan.
A pragmatic guide to AI SaaS: high‑value directions, validation and moats, pricing and unit economics, team execution, risk & compliance, and practical workflows to turn AI into a reliable coworker.
A practical, research-based guide to building AI SaaS: five viable directions, validation tactics, defensible moats, pricing models, and the unit economics discipline to avoid margin death.
2025-09-18. A concise, objective roundup of recent DeepSeek-R1 research: RL to incentivize reasoning, what independent evaluations show, and what it means for product, research, and safety.
Practical, science‑aligned guidance for adult sexual health—contraception, STI screening, fertility, and communication—with AI as a private assistant.
A practical, research-aligned guide to using AI for age-appropriate sex education in early childhood—covering body literacy, boundaries, and safety-by-design.
Practical, stigma‑free guidance for sexual wellbeing after 50—menopause/andropause, intimacy, STI prevention, and digital romance safety—with AI as a private coach.
A practical, research-aligned guide to using AI for age-appropriate sex education for teens: Hormones, Privacy, and Prevention.
An in-depth exploration of AI sex robots' technological status, social impact, and ethical challenges, from the loneliness epidemic to the future reshaping of human relationships
How a Chinese startup with $5.6M budget challenged OpenAI's dominance through open-source innovation, algorithmic efficiency, and the promise of autonomous AI agents
Explore how personal AI systems are transforming from generic tools to personalized digital companions, reshaping human cognition, social relationships, and the very nature of individual potential through advanced personalization, federated learning, and ethical AI frameworks.
How Montreal became an unlikely AI powerhouse through strategic government investment, collaborative culture, and the unwavering perseverance of the 'Canadian Mafia' during the AI winter.
How an ancient university steeped in theoretical tradition became the birthplace of both foundational computer science and the world's most ambitious AI company, proving that the deepest science can yield the most practical results.
Master the art of combining advanced prompt techniques through practical scenarios. Learn how experts structure AI conversations for programming, content creation, and strategic decision-making.
Master advanced prompt engineering techniques that transform AI from a simple Q&A tool into a sophisticated reasoning partner capable of complex problem-solving and self-verification.
How UC Berkeley became the moral and methodological compass of artificial intelligence through open science, theoretical rigor, and ethical leadership.
As AI healthcare technologies advance rapidly, we examine the critical regulatory, ethical, and social challenges that must be addressed to ensure responsible implementation and equitable access to AI-powered medical care.
Master the art of prompt engineering through role prompting, structured instructions, and chain-of-thought reasoning. Transform weak prompts into powerful tools with proven frameworks.
As the concluding piece of the AI Origins series, this article shifts from technological history to intellectual history, exploring the deep patterns of AI development, philosophical reflections, and the relationship with human civilization, revealing the eternal themes of intelligence exploration.
Explore how artificial intelligence is revolutionizing surgical precision through advanced navigation systems, robotic assistance, and real-time guidance, transforming surgeons into precision artists in the operating room.
Explore how AI is transforming healthcare from reactive treatment to proactive prevention through personalized health management, digital phenotyping, and intelligent wearables that serve as your personal health butler.
Explore AI's revolutionary leap from single-text processing to multimodal perception and embodied intelligence, witnessing how breakthrough technologies like CLIP, DALL-E, and Sora redefine human-computer interaction, and how robotics enables AI to truly enter the physical world.
An in-depth analysis of how AI is shattering the traditional pharmaceutical 'ten years, ten billion dollars' rule, exploring the complete chain revolution from target discovery to clinical trials, and examining how technologies like Insilico Medicine, Exscientia, and AlphaFold are redefining the speed and cost of drug discovery.
Demystifying the three fundamental pillars of large language models: tokenization, probability prediction, and attention mechanisms. Understand the mathematical foundations that make prompt engineering possible.
Demystifying AI large language models and prompt engineering - understanding the science behind effective AI communication through token prediction mechanisms and probability distributions.
From 'understanding' language to 'generating' worlds, how one architecture opened a new era of AI
Explore how artificial intelligence revolutionized medical imaging analysis, from CNN breakthroughs in radiology to FDA-approved diagnostic tools, transforming how doctors detect diseases with unprecedented accuracy.
Discover how Carnegie Mellon University's engineering-first culture and systems integration approach quietly revolutionized AI, from pioneering speech recognition to autonomous vehicles, establishing the blueprint for practical artificial intelligence.
Exploring the historical inevitability and initial opportunities that led AI to penetrate the sacred halls of medicine, from early expert systems to modern deep learning breakthroughs.
Exploring the deep causes of the second AI winter, from the collapse of expert systems to the underground revival of connectionism, revealing the most critical turning point in AI development history.
A deep dive into how East Coast academic rigor and West Coast entrepreneurial spirit created two distinct but equally powerful AI research traditions
Exploring the academic DNA and cultural foundations that made certain universities the birthplaces of artificial intelligence breakthroughs
Exploring the fascinating parallel evolution and ultimate convergence of artificial intelligence and medicine from isolated disciplines to revolutionary partnership.
How expert systems transformed AI from academic curiosity to commercial reality in the 1980s, creating the first AI boom and establishing the foundation for modern knowledge-based systems.
Exploring the brilliant achievements of 1950s-1970s artificial intelligence and the subsequent arrival of the first AI winter, revealing the profound contradictions between technological ambition and reality's limitations
Explore the historic 1956 Dartmouth Conference that officially launched artificial intelligence as a field, featuring the Logic Theorist program and the visionary scientists who opened AI's golden age.
Exploring the thousand-year intellectual journey from ancient myths of intelligent machines to the mathematical foundations that made modern AI possible.