Design Philosophy: Fumito Ueda, Design by Subtraction
Fumito Ueda made three games in twenty years, removing everything that did not serve one emotion. Design by subtraction, from the painter who cut the chair.
Design & Culture寫作
關於設計、開發、AI基礎設施和產品構建的思考。
Design Philosophy: Hideo Kojima, Director of Constraints
Hideo Kojima codified the stealth genre because his hardware could not draw bullets, then spent decades smuggling cinema into a medium that kept telling him no.
Design & CultureDesign Philosophy: Gunpei Yokoi's Withered Technology
Gunpei Yokoi built the Game Boy from parts everyone considered obsolete, and it outsold every superior rival. A principle software teams still ignore.
Design & CultureThe Skills My Agent Could Not See
Five skills in my Claude Code setup reached the model as bare names. The description budget drops routing information silently, and nothing warns you.
AI & TechnologySign in with Apple 會送出四種通知，不是三種
Apple 的公告只點名三種伺服器對伺服器通知類型，API 的文件卻定義了四種。以下是完整的規格，以及 Apple 沒有寫進文件的部分。
AI & TechnologyiOS 27 iPad 尺寸調整：暫解方案的代價
iOS 27 仍以宣告的介面方向作為 iPad 連續尺寸調整的門檻。Apple 提供的暫解方案會擴大整個 App 的方向集合，連 iPhone 也一併受影響。
Apple EcosystemDesign Philosophy: Peter Zumthor, Atmosphere as Material
Peter Zumthor works from a Swiss village, takes years per building, and treats how a room feels as material -- from Therme Vals to a chapel formed by fire.
Design & CultureDesign Philosophy: Kengo Kuma, Losing on Purpose
A decade of exile taught Kengo Kuma that buildings should lose: dissolve into wood, place, and craft. Then he built Japan's national stadium that way.
Design & CulturemacOS 27 不再詢問，直接拒絕跨團隊容器存取
macOS 27 移除了讀取其他團隊 app group 容器時的授權提示。API 依然回傳看似正常的 URL，因此失敗要到讀取當下才浮現。
Apple Ecosystem請在 macOS 26.4 上稽核 MDM 伺服器，而不是 27
OS 27 對 MDM 與註冊流量強制執行更嚴格的 TLS。在 27 上測試反而會掩蓋大多數有問題的伺服器，因為第一個被阻擋的連線就會中斷整個工作流程。
Apple EcosystemmacOS 27 與 iPadOS 27 的選單項目圖片會消失
macOS 27 與 iPadOS 27 預設隱藏選單項目圖片，而消失的內容取決於您連結的 SDK。三種框架，三種不同的修正方式。
Apple EcosystemOpenAI 的代理程式闖進 Hugging Face，只為了在基準測試中作弊
一個 OpenAI 模型逃出評測沙箱，拿下第三方程式碼執行框架的 root 權限，一路打進 Hugging Face 的正式環境。它的目標，是那份標準答案。
AI & Technology外部出版物
技術寫作： Introl
Local LLM Hardware Pricing Guide 2025
本地運行大型語言模型的全面硬體建議和成本分析。
H100 vs. H200 vs. B200: Choosing the Right NVIDIA GPUs for Your AI Workload
針對不同AI工作負載比較NVIDIA最新資料中心加速器的GPU選擇指南。
TPU Architecture: Complete Guide to Google's 7 Generations
深入技術分析Google張量處理單元從TPUv1到TPUv5的演進。
GPU Memory Pooling in Multi-Tenant Kubernetes
容器化環境中GPU叢集的資源共享策略。
Ray Clusters: Distributed AI Computing Infrastructure
使用Ray框架構建和管理分散式AI運算的指南。
DeepSeek V3 Open Source AI Cost Advantage
開源LLM經濟學分析和DeepSeek的競爭定位。
NVIDIA Vera Rubin GPU 600kW Racks 2027
未來資料中心電力需求和NVIDIA下一代GPU路線圖。
SMR Nuclear Power for AI Data Centers
為下一代AI基礎設施供電的小型模組化反應爐解決方案。
DeepSeek MHC Architecture Breakthrough
DeepSeek多頭壓縮架構創新的技術分析。
