-
如何優化 AI Agent:從常駐指令、Skills 到驗證方法
-
AI Agent Weekly Overview W34: The Harness Becomes the Shared Capability and Security Boundary (Industry × Research × Open Source)
-
AI Agent 週報總覽 W34:harness 成為能力與安全的共同邊界(產業×研究×開源)
-
AI Agent Open-Source Weekly W34: Harnesses Split into Plugins, Context, and Isolated Computers (DeepSeek Harness, Headroom, OpenBot)
-
AI Agent 開源週報 W34:harness 生態拆出插件、context 與獨立電腦(DeepSeek Harness、Headroom、OpenBot)
-
AI Agent Frontier Research Weekly W34: The Harness Becomes a Trainable Capability Layer (ClawGym II, HarnessEval-W, Anything2Skill)
-
AI Agent 前沿研究週報 W34:harness 從介面升級為可訓練能力層(ClawGym II、HarnessEval-W、Anything2Skill)
-
AI Agent Industry Weekly W34: Capability Pushes Control Down into the Runtime (OpenAI, NVIDIA AVO, Alipay)
-
AI Agent 產業週報 W34:能力越強,控制邊界越往 runtime 下沉(OpenAI、NVIDIA AVO、支付寶)
-
AI Agent Weekly Overview W33: The Control Plane Replaces Single-Model Worship (Industry × Research × Open Source)
-
AI Agent 週報總覽 W33:控制面取代單模型崇拜(產業×研究×開源)
-
AI Agent Open-Source Weekly W33: Multi-Agent Engineering Adds Orchestration and Testing (agent-orchestrator, ADK Go, aimock)
-
AI Agent 開源週報 W33:多 Agent 工程補齊編排與測試(agent-orchestrator、ADK Go、aimock)
-
AI Agent Frontier Research Weekly W33: Interfaces and Trajectories Are the Real Variables (SWE-Bench ProMax, Recursive Synthesis, TraceSafe)
-
AI Agent 前沿研究週報 W33:真正的變數是介面與軌跡(SWE-Bench ProMax、Recursive Synthesis、TraceSafe)
-
AI Agent Industry Weekly W33: Competition Moves to Routing and Governance (Gemini 3.7 Flash, Grok 4.6, Switchyard)
-
AI Agent 產業週報 W33:模型競賽進入路由與治理層(Gemini 3.7 Flash、Grok 4.6、Switchyard)
-
別讓 AI 重學三百年物理:PIML 真正重要的不是多一個 Loss
-
PCA 壓縮自然影像:為什麼找出主要方向還不夠?
-
Laplacian operator 介紹:一個算子如何看見曲率、擴散與邊界
-
Grid cells 深度介紹:從六角放電到路徑整合假說
-
Edward Tolman:當老鼠在迷宮裡學會一張看不見的地圖
-
PCA 到底在做什麼?從幾何直覺、特徵值到 SVD 的完整推導
-
海馬的認知地圖不是地圖:從 Successor Representation 讀懂 place cells 與 grid cells