{
 "generated": "2026-10-04",
 "site": "https://ice-wocker.github.io/frontier-knowledge-base",
 "count": 163,
 "docs": [
  {
   "i": 0,
   "title": "3D 与高斯泼溅",
   "t": "3D 与高斯泼溅",
   "u": "ai/3d-and-gaussian-splatting.html",
   "s": "三维内容生成的主流技术可归为两类：一是 重建/渲染 （Neural Radiance Fields，NeRF；3D Gaussian Splatting，3DGS），从多视角图像恢复可渲染的场景表示；二是 生成式 3D （image/text-to-3D），直接产出可直接使用的 3D 资产（mesh + PBR 贴图）。2025–2026 年的趋势是两者融合…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "gaussian",
    "cvpr2026",
    "rendering",
    "generators",
    "leveraging",
    "dynamic",
    "content",
    "compared"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "扩散与图像生成",
     "url": "ai/diffusion-and-image-generation.html",
     "score": 0.1252
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1173
    },
    {
     "title": "视频生成与世界模型",
     "url": "ai/video-generation-and-world-models.html",
     "score": 0.1156
    },
    {
     "title": "图学习与知识图谱",
     "url": "ai/graph-neural-networks-and-knowledge-graphs.html",
     "score": 0.0679
    }
   ]
  },
  {
   "i": 1,
   "title": "AI Agent（智能体）技术与生态",
   "t": "AI Agent（智能体）技术与生态",
   "u": "ai/ai-agents.html",
   "s": "AI Agent（智能体）指以大语言模型为推理核心、能够调用工具、读取与写入外部系统、并在多步任务中自主决策的系统。从工程视角看，一个典型 agent 由「模型 + 工具接口 + 状态与记忆 + 编排循环 + 防护栏」构成：模型负责推理与决策，工具接口负责把模型意图落到真实系统，状态与记忆负责跨步骤保持一致，编排循环负责把多步任务串起来，防护栏负责约束其自主…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 20,
   "tags": [
    "a2a",
    "browser",
    "langgraph",
    "terminal-bench",
    "adk",
    "crewai",
    "osworld",
    "cowork"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 58,
   "related": [
    {
     "title": "Agent 框架与工具链",
     "url": "ai/agent-frameworks-and-tooling.html",
     "score": 0.2668
    },
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.1836
    },
    {
     "title": "AI 编程工具与编程智能体",
     "url": "ai/ai-coding-tools.html",
     "score": 0.1745
    },
    {
     "title": "大模型评测基准（Benchmarks）",
     "url": "ai/ai-benchmarks.html",
     "score": 0.1591
    }
   ]
  },
  {
   "i": 2,
   "title": "AI for Science",
   "t": "AI for Science",
   "u": "ai/ai-for-science.html",
   "s": "AI for Science 指用机器学习方法加速自然科学发现，代表性方向包括：蛋白质与生物分子结构预测（AlphaFold 系列）、生物基础模型（ESM3）、材料发现（GNoME、MatterGen）、AI 气象预报（GraphCast、Pangu-Weather、Aurora、GenCast、AIFS）、数学定理证明（AlphaProof）与基因组调控预…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 14,
   "tags": [
    "alphaproof",
    "graphcast",
    "ecmwf",
    "esm3",
    "weather",
    "alphafold",
    "aifs",
    "gnome"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 33,
   "related": [
    {
     "title": "图学习与知识图谱",
     "url": "ai/graph-neural-networks-and-knowledge-graphs.html",
     "score": 0.0944
    },
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.076
    },
    {
     "title": "推荐系统与 AI（Recommender Systems & AI）",
     "url": "ai/recommender-systems-and-ai.html",
     "score": 0.0698
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.0697
    }
   ]
  },
  {
   "i": 3,
   "title": "AI 产品与 UX",
   "t": "AI 产品与 UX",
   "u": "ai/ai-product-and-ux.html",
   "s": "2025–2026 年，AI 产品的交互范式从「对话框」向「入口 + 智能体」演进：AI 助手不再只是插件，而成为承载任务执行、应用构建与持续工作的入口。与此同时，评测驱动迭代（Eval-Driven Development）成为产品工程的基本实践，定价模式也从按席位转向按用量与按结果。这一转变意味着 AI 助手开始覆盖从对话、写作到应用构建与后台任务的完整…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "pricing",
    "agentforce",
    "copilot",
    "assistant",
    "tests",
    "coding",
    "evals",
    "office"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "AI 编程工具与编程智能体",
     "url": "ai/ai-coding-tools.html",
     "score": 0.1555
    },
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.0876
    },
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.0843
    },
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.084
    }
   ]
  },
  {
   "i": 4,
   "title": "AI 发展史与关键里程碑",
   "t": "AI 发展史与关键里程碑",
   "u": "ai/ai-history-and-milestones.html",
   "s": "人工智能（Artificial Intelligence, AI）作为一个被正式命名的学科，通常以 1956 年在美国达特茅斯学院（Dartmouth College）举办的暑期研讨会为起点，正是在这次会议上 \"artificial intelligence\" 一词被采纳，该领域被组织为一个正式的研究计划（History of Artificial Inte…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "timeline",
    "expert",
    "history",
    "deepseek-r1",
    "boom",
    "implications",
    "gpt-5",
    "gpt-6"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心脉络与关键里程碑",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.143
    },
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.1387
    },
    {
     "title": "大模型评测基准（Benchmarks）",
     "url": "ai/ai-benchmarks.html",
     "score": 0.1024
    },
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.0946
    }
   ]
  },
  {
   "i": 5,
   "title": "AI 基础设施与算力",
   "t": "AI 基础设施与算力",
   "u": "ai/ai-infrastructure-and-compute.html",
   "s": "AI 基础设施是前沿模型的物理底座，涵盖超大规模训练集群、GPU/TPU 互联网络、训练与推理软件栈、模型架构效率技术（MoE、量化、KV cache 优化）、以及数据中心的电力与散热。2025–2026 年，随着旗舰模型训练规模迈入\"十万卡级\"，基础设施的主旋律从\"堆算力\"转向\"每瓦性能\"与\"每 token 成本\"：一方面互联带宽、液冷与供电成为集群扩展的…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 25,
   "tags": [
    "cooling",
    "hbm4",
    "data-center",
    "vera",
    "pflops",
    "rubin",
    "tpu7x",
    "pue"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/公司/产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 79,
   "related": [
    {
     "title": "推理优化与服务：从 PagedAttention 到分离式部署",
     "url": "ai/inference-optimization-and-serving.html",
     "score": 0.1571
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1449
    },
    {
     "title": "软硬协同设计（AI Hardware-Software Co-Design）",
     "url": "ai/ai-hardware-software-codesign.html",
     "score": 0.1369
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.1298
    }
   ]
  },
  {
   "i": 6,
   "title": "AI 安全与对齐（AI Safety and Alignment）",
   "t": "AI 安全与对齐（AI Safety and Alignment）",
   "u": "ai/ai-safety-and-alignment.html",
   "s": "AI 安全与对齐（AI Safety and Alignment） 最后更新：2026-09-26 ｜ 领域：人工智能 / 安全、对齐与治理 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 AI 安全与对齐关注两类问题： 误用风险（misuse） ——模型被用于生化、网络、操控等危害；以及 失控 / 失准风险（misalignment & lo…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 25,
   "tags": [
    "alignment",
    "aisi",
    "responsible",
    "jailbreak",
    "constitutional",
    "rlhf",
    "risk",
    "sae"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心技术与关键概念",
    "四、代表性框架与项目",
    "五、关键数据与评测结果",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 67,
   "related": [
    {
     "title": "可解释性与机制分析",
     "url": "ai/interpretability-and-mechanistic-analysis.html",
     "score": 0.193
    },
    {
     "title": "AI 安全攻防",
     "url": "ai/ai-security-threats-and-defense.html",
     "score": 0.1658
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.1494
    },
    {
     "title": "后训练与对齐方法",
     "url": "ai/post-training-and-alignment-methods.html",
     "score": 0.135
    }
   ]
  },
  {
   "i": 7,
   "title": "AI 安全攻防",
   "t": "AI 安全攻防",
   "u": "ai/ai-security-threats-and-defense.html",
   "s": "随着 LLM 与 Agent 进入生产系统，AI 的攻击面从「内容风险」向「行为风险」扩散：一方面智能体在复杂环境中的错误决策或越权操作可能直接导致事故，另一方面攻击者可通过工具投毒等新手段劫持智能体用于违法犯罪，责任主体更模糊（专家解读 以规范应用护航智能体发展行稳致远）。行业普遍以 OWASP Top 10 for LLM Applications 作为…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "owasp",
    "jailbreak",
    "mitigation",
    "poisoning",
    "applications",
    "aisi",
    "testing",
    "trace"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "关键防御手段与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.1658
    },
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.0923
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.081
    },
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.0729
    }
   ]
  },
  {
   "i": 8,
   "title": "AI 搜索与深度研究（AI Search & Deep Research）",
   "t": "AI 搜索与深度研究（AI Search & Deep Research）",
   "u": "ai/ai-search-and-deep-research.html",
   "s": "AI 搜索（AI Search）指以对话式界面替代或补充传统关键词检索的产品形态，典型代表包括 ChatGPT Search、Google AI Overviews、Perplexity 等；深度研究（Deep Research）则是其中的\"重模式\"分支：Agent 自主制定研究计划、反复检索、识别知识缺口并再次检索，最终产出多页带引用的报告（Build w…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "geo",
    "chatgpt",
    "search",
    "browser",
    "gemini",
    "deep",
    "atlas",
    "era"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "对话助手与语音（Conversational Assistants & Voice）",
     "url": "ai/conversational-assistants-and-voice.html",
     "score": 0.1285
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1254
    },
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.1252
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1056
    }
   ]
  },
  {
   "i": 9,
   "title": "AI 编程工具与编程智能体",
   "t": "AI 编程工具与编程智能体",
   "u": "ai/ai-coding-tools.html",
   "s": "AI 编程工具与编程智能体 最后更新：2026-09-26 ｜ 领域：人工智能 / 软件开发工具与编程智能体 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 AI 编程工具已从「补全代码」演进为「自主完成任务的编程智能体（coding agent）」。2025–2026 年，市场形成三条主线： IDE 内嵌智能体 （GitHub Copilot…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 23,
   "tags": [
    "codex",
    "terminal-bench",
    "cursor",
    "vibe",
    "coding",
    "ai-generated",
    "copilot",
    "swe-bench"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心技术与关键概念",
    "四、代表性工具 / 产品",
    "五、关键数据与评测结果",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 62,
   "related": [
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.1745
    },
    {
     "title": "大模型评测基准（Benchmarks）",
     "url": "ai/ai-benchmarks.html",
     "score": 0.1716
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1637
    },
    {
     "title": "AI 产品与 UX",
     "url": "ai/ai-product-and-ux.html",
     "score": 0.1555
    }
   ]
  },
  {
   "i": 10,
   "title": "AI 评测实践",
   "t": "AI 评测实践",
   "u": "ai/ai-evaluation-practices.html",
   "s": "AI 评测实践（evaluation）指对模型与 AI 应用的能力、可靠性与安全性进行系统化测量，涵盖评测集构建、打分方法（LLM-as-judge、人工评测、A/B）、数据污染检测、统计显著性与安全评测等。随着 LLM 被用于训练监督信号与生产质量门禁，评测本身的可信度成为关键问题：如果判分器（judge）系统性偏差，那么「以模型评模型」的评测会放大误差。…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "redbench",
    "judge",
    "teaming",
    "reliability",
    "bias",
    "labels",
    "wrong",
    "plosone"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "大模型评测基准（Benchmarks）",
     "url": "ai/ai-benchmarks.html",
     "score": 0.1715
    },
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.1494
    },
    {
     "title": "MLOps 与 LLMOps",
     "url": "ai/mlops-and-llmops.html",
     "score": 0.1168
    },
    {
     "title": "后训练与对齐方法",
     "url": "ai/post-training-and-alignment-methods.html",
     "score": 0.1098
    }
   ]
  },
  {
   "i": 11,
   "title": "Agent 工程实践（Agent Engineering Patterns）",
   "t": "Agent 工程实践（Agent Engineering Patterns）",
   "u": "ai/agent-engineering-patterns.html",
   "s": "Agent 工程（Agent Engineering）指把大语言模型（LLM）从单轮文本补全扩展为能自主规划、调用工具、维护状态并多步完成任务的系统的实践体系。2026 年的业界共识是：决定 Agent 成败的往往不是更大的模型，而是如何拆分工作、如何管理上下文与记忆、如何评估与观测（5 Key Concepts Behind Agentic AI Ever…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 17,
   "tags": [
    "gaia",
    "letta",
    "langfuse",
    "bench",
    "mem0",
    "persistent",
    "calling",
    "pass"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "失败模式与工程建议",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 40,
   "related": [
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.2172
    },
    {
     "title": "提示工程与上下文工程（Prompt & Context Engineering）",
     "url": "ai/prompt-and-context-engineering.html",
     "score": 0.195
    },
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.1836
    },
    {
     "title": "MLOps 与 LLMOps",
     "url": "ai/mlops-and-llmops.html",
     "score": 0.1557
    }
   ]
  },
  {
   "i": 12,
   "title": "Agent 框架与工具链",
   "t": "Agent 框架与工具链",
   "u": "ai/agent-frameworks-and-tooling.html",
   "s": "Agent 框架与工具链指构建 LLM 智能体（agent）所需的软件分层：编排框架（LangGraph、CrewAI、AutoGen、OpenAI Agents SDK 等）、上下文与文档框架（LlamaIndex）、工具与上下文连接协议（Model Context Protocol, MCP），以及记忆、沙箱、可观测等基础设施。2025–2026 年，这…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "langgraph",
    "crewai",
    "llamaindex",
    "adk",
    "autogen",
    "knowledge-base",
    "ai-agents",
    "mcp"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.2668
    },
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.1472
    },
    {
     "title": "MLOps 与 LLMOps",
     "url": "ai/mlops-and-llmops.html",
     "score": 0.1187
    },
    {
     "title": "AI 编程工具与编程智能体",
     "url": "ai/ai-coding-tools.html",
     "score": 0.1184
    }
   ]
  },
  {
   "i": 13,
   "title": "GPU 编程与算子：CUDA、Triton 与 FlashAttention",
   "t": "GPU 编程与算子：CUDA、Triton 与 FlashAttention",
   "u": "ai/gpu-kernels-and-cuda.html",
   "s": "大模型的训练与推理性能，最终取决于 GPU 上算子（kernel）的实现质量。所谓\"算子工程\"，是围绕 NVIDIA GPU 的执行模型做三件事： 把数据搬运与计算重叠 、 减少 HBM 与片上存储之间的往返 、 消除 host 侧启动开销 。这三件事分别对应 FlashAttention 系列、算子融合（fusion）与 CUDA Graphs。 CUDA…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 14,
   "tags": [
    "triton",
    "kernels",
    "cuda",
    "tile",
    "dsl",
    "nccl",
    "co-design",
    "gemm"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 33,
   "related": [
    {
     "title": "软硬协同设计（AI Hardware-Software Co-Design）",
     "url": "ai/ai-hardware-software-codesign.html",
     "score": 0.2122
    },
    {
     "title": "推理优化与服务：从 PagedAttention 到分离式部署",
     "url": "ai/inference-optimization-and-serving.html",
     "score": 0.145
    },
    {
     "title": "分布式训练与并行策略",
     "url": "ai/distributed-training-and-parallelism.html",
     "score": 0.1322
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.1318
    }
   ]
  },
  {
   "i": 14,
   "title": "MLOps 与 LLMOps",
   "t": "MLOps 与 LLMOps",
   "u": "ai/mlops-and-llmops.html",
   "s": "MLOps（Machine Learning Operations）指把软件工程中的持续集成、持续交付、监控与治理实践延伸到机器学习模型的全生命周期，典型环节包括实验跟踪、模型注册、部署发布、生产监控与回滚。MLflow 是将这些环节收敛到单一平台的代表，其官方定位强调「自动记录参数、权重、工件、代码、指标与依赖，确保实验可复现」，并内置模型注册表以控制模型…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "llmops",
    "mlflow",
    "langfuse",
    "langsmith",
    "monitoring",
    "observability",
    "mlops",
    "versioning"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.1557
    },
    {
     "title": "Agent 框架与工具链",
     "url": "ai/agent-frameworks-and-tooling.html",
     "score": 0.1187
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.1168
    },
    {
     "title": "提示工程与上下文工程（Prompt & Context Engineering）",
     "url": "ai/prompt-and-context-engineering.html",
     "score": 0.1
    }
   ]
  },
  {
   "i": 15,
   "title": "MoE 稀疏架构",
   "t": "MoE 稀疏架构",
   "u": "ai/mixture-of-experts.html",
   "s": "混合专家（Mixture of Experts, MoE）通过让网络的局部组件\"条件性激活\"来突破稠密模型的算力约束：一个 MoE 层拥有 N 个独立的 FFN（称为\"专家\"）以及一个决定每个 token 使用哪些专家的小型路由器网络；对任一 token，只有 k 个专家被激活，其余不参与计算，因此模型可拥有远超单次前向所需参数量的总参数（Mixture o…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "experts",
    "mixture",
    "megatron",
    "deepseek-v3",
    "moe",
    "deepseekmoe",
    "parallelism",
    "folding"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 29,
   "related": [
    {
     "title": "分布式训练与并行策略",
     "url": "ai/distributed-training-and-parallelism.html",
     "score": 0.2061
    },
    {
     "title": "Transformer 架构原理",
     "url": "ai/transformer-architecture.html",
     "score": 0.1951
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1463
    },
    {
     "title": "小模型与高效模型",
     "url": "ai/efficient-and-small-models.html",
     "score": 0.138
    }
   ]
  },
  {
   "i": 16,
   "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
   "t": "RAG 与上下文工程（Context Engineering）与智能体记忆",
   "u": "ai/rag-and-memory.html",
   "s": "RAG 与上下文工程（Context Engineering）与智能体记忆 最后更新：2026-09-26 ｜ 领域：人工智能 / 检索增强生成、上下文工程与记忆 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 检索增强生成（Retrieval-Augmented Generation, RAG）曾长期被简化为「向量检索 + 拼接进提示词」的标…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 28,
   "tags": [
    "mem0",
    "zep",
    "pinecone",
    "graphrag",
    "milvus",
    "locomo",
    "retrieval",
    "qdrant"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心技术与关键概念",
    "四、代表性项目 / 产品",
    "五、关键数据与评测结果",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 80,
   "related": [
    {
     "title": "检索深入：稀疏、稠密、混合检索与向量搜索（Retrieval & Vector Search）",
     "url": "ai/retrieval-and-vector-search.html",
     "score": 0.2289
    },
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.2172
    },
    {
     "title": "提示工程与上下文工程（Prompt & Context Engineering）",
     "url": "ai/prompt-and-context-engineering.html",
     "score": 0.1832
    },
    {
     "title": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
     "url": "ai/long-context-techniques.html",
     "score": 0.1536
    }
   ]
  },
  {
   "i": 17,
   "title": "Transformer 架构原理",
   "t": "Transformer 架构原理",
   "u": "ai/transformer-architecture.html",
   "s": "Transformer 由 Vaswani 等人在 2017 年论文《Attention Is All You Need》中提出，是此后所有主流大语言模型的基础架构（The History of AI）。现代 LLM 的骨架可概括为\"注意力 + FFN 的堆叠块\"模式：GPT-4、Claude、Llama 4、DeepSeek-R1、Qwen3 等共享一批趋…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "rope",
    "mla",
    "ffn",
    "mamba-transformer",
    "yarn",
    "linear",
    "nemotron",
    "attention"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 33,
   "related": [
    {
     "title": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
     "url": "ai/long-context-techniques.html",
     "score": 0.3267
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.1951
    },
    {
     "title": "量化与模型压缩",
     "url": "ai/quantization-and-model-compression.html",
     "score": 0.1675
    },
    {
     "title": "分词与嵌入",
     "url": "ai/tokenization-and-embeddings.html",
     "score": 0.1309
    }
   ]
  },
  {
   "i": 18,
   "title": "企业 AI 落地",
   "t": "企业 AI 落地",
   "u": "ai/ai-enterprise-adoption.html",
   "s": "企业 AI 落地指组织把 AI 能力从概念验证（PoC）推进到生产系统并产生可衡量业务价值的过程，涉及用例筛选、ROI 评估、数据准备、部署形态选择（公有云 API / 私有化 / 混合）、合规与安全、组织与人才等维度。2025–2026 年，行业叙事的重心从「是否采用 AI」转向「采用后能否兑现回报」：采用率已高企，但从试点到规模化的转化率与利润影响仍然偏…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "roi",
    "mckinsey",
    "our-insights",
    "fail",
    "private",
    "gartner",
    "failure",
    "esg"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.089
    },
    {
     "title": "AI 编程工具与编程智能体",
     "url": "ai/ai-coding-tools.html",
     "score": 0.0709
    },
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.0669
    },
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.0603
    }
   ]
  },
  {
   "i": 19,
   "title": "分布式训练与并行策略",
   "t": "分布式训练与并行策略",
   "u": "ai/distributed-training-and-parallelism.html",
   "s": "当模型参数、激活值与优化器状态总量远超单卡显存时，训练必须切分到多卡、多节点。分布式训练的核心就是把\"计算、显存、通信\"三者重新分配，主流切分维度包括： - 数据并行（DP, Data Parallel） ：每卡持有完整模型副本，切分数据批次，反向传播后做梯度同步（AllReduce）。 - 张量并行（TP, Tensor Parallel） ：把单层内的矩…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "megatron",
    "zero",
    "gradient",
    "stage",
    "distributed",
    "training",
    "parallelism",
    "fsdp"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "代表性项目 / 公司 / 产品",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.2061
    },
    {
     "title": "GPU 编程与算子：CUDA、Triton 与 FlashAttention",
     "url": "ai/gpu-kernels-and-cuda.html",
     "score": 0.1322
    },
    {
     "title": "Transformer 架构原理",
     "url": "ai/transformer-architecture.html",
     "score": 0.1173
    },
    {
     "title": "量化与模型压缩",
     "url": "ai/quantization-and-model-compression.html",
     "score": 0.1163
    }
   ]
  },
  {
   "i": 20,
   "title": "分词与嵌入",
   "t": "分词与嵌入",
   "u": "ai/tokenization-and-embeddings.html",
   "s": "分词（Tokenization）把原始文本切分为模型可处理的 token 序列，嵌入（Embedding）则把 token 或整段文本映射为稠密向量。前者决定模型的\"字母表\"与词汇表规模，后者是检索、聚类、分类与 RAG 系统的语义基础。两者共同构成 LLM 的输入接口：分词器一旦随模型权重固化便难以更换，不同分词器的合并规则与词表规模不同，同一段文本会被编…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "mteb",
    "tokenization",
    "embedding",
    "tokenizer",
    "llm-fundamentals",
    "mmteb",
    "contextual",
    "bge"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 37,
   "related": [
    {
     "title": "检索深入：稀疏、稠密、混合检索与向量搜索（Retrieval & Vector Search）",
     "url": "ai/retrieval-and-vector-search.html",
     "score": 0.2333
    },
    {
     "title": "Transformer 架构原理",
     "url": "ai/transformer-architecture.html",
     "score": 0.1309
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1117
    },
    {
     "title": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
     "url": "ai/long-context-techniques.html",
     "score": 0.1103
    }
   ]
  },
  {
   "i": 21,
   "title": "可解释性与机制分析",
   "t": "可解释性与机制分析",
   "u": "ai/interpretability-and-mechanistic-analysis.html",
   "s": "可解释性研究分为两条主线：一条是行为层面的「外部可解释」（探针、输入输出测试），另一条是机制可解释性（Mechanistic Interpretability），试图识别模型内部的「特征（features）、电路（circuits）与激活模式」，把事实回忆、拒绝、类欺骗规划或代码生成等行为对应到网络中的具体组件（Mechanistic interpretab…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "interpretability",
    "causal",
    "mechanistic",
    "sae",
    "steering",
    "ran",
    "audit",
    "circuit"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.193
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.0866
    },
    {
     "title": "合成数据与数据引擎",
     "url": "ai/synthetic-data-and-data-engines.html",
     "score": 0.0861
    },
    {
     "title": "后训练与对齐方法",
     "url": "ai/post-training-and-alignment-methods.html",
     "score": 0.0833
    }
   ]
  },
  {
   "i": 22,
   "title": "合成数据与数据引擎",
   "t": "合成数据与数据引擎",
   "u": "ai/synthetic-data-and-data-engines.html",
   "s": "合成数据（Synthetic Data）指由模型而非人工生成、用于训练的数据。据 2026 年的行业总结，合成数据已成为几乎所有专用模型的默认监督来源：Tülu、OpenHermes、Llama-3-Instruct 等开放权重指令模型高度依赖合成数据，定制的 reranker、分类器与 embedder 也以合成语料为主，根本原因是人工标注是瓶颈，而 LL…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "collapse",
    "synthetic",
    "distillation",
    "closer",
    "look",
    "fine-tuning",
    "abs",
    "multimodal"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 12,
   "related": [
    {
     "title": "预训练与缩放定律",
     "url": "ai/pretraining-and-scaling-laws.html",
     "score": 0.1243
    },
    {
     "title": "强化学习前沿",
     "url": "ai/reinforcement-learning-frontier.html",
     "score": 0.1131
    },
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.1058
    },
    {
     "title": "后训练与对齐方法",
     "url": "ai/post-training-and-alignment-methods.html",
     "score": 0.0931
    }
   ]
  },
  {
   "i": 23,
   "title": "后训练与对齐方法",
   "t": "后训练与对齐方法",
   "u": "ai/post-training-and-alignment-methods.html",
   "s": "后训练（Post-training）指预训练之后、使模型具备指令遵循、偏好对齐与推理能力的一系列方法。一份 2025 年的后训练综述把 SFT、RLHF、DPO、GRPO 等技术放入统一脉络，指出 DPO（2023）通过直接依据人类偏好优化模型输出、绕过中间奖励建模来简化 RLHF，而 GRPO 最初是在其团队此前的 DeepSeekMath 工作中提出的（…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "dpo",
    "grpo",
    "rlhf",
    "post-training",
    "sft",
    "alignment",
    "preference",
    "reinforcement"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "强化学习前沿",
     "url": "ai/reinforcement-learning-frontier.html",
     "score": 0.2155
    },
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.2019
    },
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.135
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.1098
    }
   ]
  },
  {
   "i": 24,
   "title": "图学习与知识图谱",
   "t": "图学习与知识图谱",
   "u": "ai/graph-neural-networks-and-knowledge-graphs.html",
   "s": "图学习（Graph Learning）以图神经网络（GNN）与图 Transformer 为核心，处理节点、边与整图的表示学习；知识图谱（Knowledge Graph，KG）则以实体—关系—实体三元组刻画结构化知识。2025–2026 年最活跃的交汇点是 GraphRAG ：用 LLM 从非结构化文本抽取实体与关系构建知识图谱，再以图结构增强检索与推理，缓…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "graphrag",
    "drug",
    "graph",
    "acl",
    "fair",
    "knowledge",
    "prediction",
    "discovery"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "检索深入：稀疏、稠密、混合检索与向量搜索（Retrieval & Vector Search）",
     "url": "ai/retrieval-and-vector-search.html",
     "score": 0.1071
    },
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.107
    },
    {
     "title": "AI for Science",
     "url": "ai/ai-for-science.html",
     "score": 0.0944
    },
    {
     "title": "推荐系统与 AI（Recommender Systems & AI）",
     "url": "ai/recommender-systems-and-ai.html",
     "score": 0.0925
    }
   ]
  },
  {
   "i": 25,
   "title": "多模态 AI（Multimodal AI）",
   "t": "多模态 AI（Multimodal AI）",
   "u": "ai/multimodal-ai.html",
   "s": "多模态 AI（Multimodal AI） 最后更新：2026-09-26 ｜ 领域：人工智能 / 多模态模型与生成 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 多模态 AI 指同时理解与生成文本、图像、音频、视频乃至 3D 世界的模型体系。2025–2026 年的主线是「 原生多模态（native multimodal） 」——不再把视觉…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 21,
   "tags": [
    "banana",
    "kling",
    "veo",
    "sora",
    "generators",
    "genie",
    "video",
    "nano"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心技术与关键概念",
    "四、代表性项目 / 产品",
    "五、关键数据与评测结果",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 71,
   "related": [
    {
     "title": "视频生成与世界模型",
     "url": "ai/video-generation-and-world-models.html",
     "score": 0.2641
    },
    {
     "title": "扩散与图像生成",
     "url": "ai/diffusion-and-image-generation.html",
     "score": 0.2077
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1962
    },
    {
     "title": "对话助手与语音（Conversational Assistants & Voice）",
     "url": "ai/conversational-assistants-and-voice.html",
     "score": 0.1781
    }
   ]
  },
  {
   "i": 26,
   "title": "大模型评测基准（Benchmarks）",
   "t": "大模型评测基准（Benchmarks）",
   "u": "ai/ai-benchmarks.html",
   "s": "评测基准（benchmark）是衡量大模型能力的标尺，也是厂商发布时的「成绩单」。一个基准通常由三要素定义：任务与数据（题目从哪来、如何标注）、评测协议（模型能否用工具、思考多少、允许几次尝试）与判定方式（精确匹配、测试通过、人类偏好投票）。理解基准时必须同时看这三者，否则分数难以解读。 2025–2026 年，基准生态呈现两条主线：一是经典知识类基准（MM…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 19,
   "tags": [
    "hle",
    "exam",
    "mmlu-pro",
    "arc-agi",
    "contamination",
    "mmlu",
    "humanity",
    "last"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心基准详解",
    "关键数据速览（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 58,
   "related": [
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.2479
    },
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.2001
    },
    {
     "title": "AI 编程工具与编程智能体",
     "url": "ai/ai-coding-tools.html",
     "score": 0.1716
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.1715
    }
   ]
  },
  {
   "i": 27,
   "title": "对话助手与语音（Conversational Assistants & Voice）",
   "t": "对话助手与语音（Conversational Assistants & Voice）",
   "u": "ai/conversational-assistants-and-voice.html",
   "s": "对话助手与语音技术栈由四层构成：语音识别（ASR，语音转文本）、语音合成（TTS，文本转语音）、实时语音对话模型（speech-in / speech-out，端到端低延迟交互），以及承载它们的消费级助手生态（如 Siri、Alexa+、Gemini）。2026 年的关键变化是：实时语音 API 开始支持语音直进直出的对话，且主要平台把\"对话式 AI\"作为一…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 15,
   "tags": [
    "siri",
    "tts",
    "elevenlabs",
    "asr",
    "speech",
    "gpt-live",
    "voice",
    "whisper"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 36,
   "related": [
    {
     "title": "音乐与音频生成",
     "url": "ai/audio-music-and-speech-generation.html",
     "score": 0.2248
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1781
    },
    {
     "title": "AI 搜索与深度研究（AI Search & Deep Research）",
     "url": "ai/ai-search-and-deep-research.html",
     "score": 0.1285
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1113
    }
   ]
  },
  {
   "i": 28,
   "title": "小模型与高效模型",
   "t": "小模型与高效模型",
   "u": "ai/efficient-and-small-models.html",
   "s": "小语言模型（Small Language Model, SLM）与\"高效模型\"指以极低推理成本换取可用智能的一类模型，通常具备三个特征之一或多个： 参数量小 （一般指 3B 以下）、 稀疏激活 （MoE 结构，总参数量大但每 token 只激活一小部分）、以及 面向端侧优化 （量化格式、低内存占用、CPU/移动 NPU 可运行）。 学术工作中对 SLM 的常…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "gemma",
    "smollm2",
    "phi-4",
    "phi",
    "tok",
    "gemma-3n",
    "iphone",
    "rt-lm"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 28,
   "related": [
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.1852
    },
    {
     "title": "开源 AI 工具链",
     "url": "ai/open-source-ai-tooling.html",
     "score": 0.1612
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1435
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.138
    }
   ]
  },
  {
   "i": 29,
   "title": "开放权重（开源）模型生态",
   "t": "开放权重（开源）模型生态",
   "u": "ai/open-weight-models.html",
   "s": "开放权重（开源）模型生态 最后更新：2026-09-26 ｜ 领域：人工智能 / 开放权重模型与本地部署 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 「开放权重（open-weight）」指模型参数可公开下载、可自行部署与微调，区别于仅提供 API 的闭源模型。2025–2026 年，开放权重阵营从「追赶者」变为「半壁江山」：一方面，Met…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 24,
   "tags": [
    "minimax",
    "open-weight",
    "gemma",
    "guardrails",
    "phi",
    "gpt-oss",
    "qwen",
    "kimi"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心技术与关键概念",
    "四、代表性项目 / 产品",
    "五、关键数据与评测结果",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 62,
   "related": [
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.2357
    },
    {
     "title": "小模型与高效模型",
     "url": "ai/efficient-and-small-models.html",
     "score": 0.1852
    },
    {
     "title": "开源 AI 工具链",
     "url": "ai/open-source-ai-tooling.html",
     "score": 0.1726
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.138
    }
   ]
  },
  {
   "i": 30,
   "title": "开源 AI 工具链",
   "t": "开源 AI 工具链",
   "u": "ai/open-source-ai-tooling.html",
   "s": "开源 AI 工具链涵盖模型与数据托管、训练与微调、推理与服务、本地运行、评测等环节。当前生态大致可分为几层：以 Hugging Face Hub 为核心的模型与数据集托管层；以 Transformers、PyTorch 等为代表的训练/微调框架层；以 vLLM、SGLang、TensorRT-LLM 为代表的高吞吐服务层；以 llama.cpp、Ollama…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "gguf",
    "ollama",
    "sglang",
    "vllm-metal",
    "vllm",
    "quantization",
    "mlx",
    "open-weight"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 28,
   "related": [
    {
     "title": "推理优化与服务：从 PagedAttention 到分离式部署",
     "url": "ai/inference-optimization-and-serving.html",
     "score": 0.2116
    },
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.1726
    },
    {
     "title": "小模型与高效模型",
     "url": "ai/efficient-and-small-models.html",
     "score": 0.1612
    },
    {
     "title": "量化与模型压缩",
     "url": "ai/quantization-and-model-compression.html",
     "score": 0.1593
    }
   ]
  },
  {
   "i": 31,
   "title": "强化学习前沿",
   "t": "强化学习前沿",
   "u": "ai/reinforcement-learning-frontier.html",
   "s": "强化学习（Reinforcement Learning, RL）在 2025–2026 年重新成为大模型与具身智能的核心方法。最显著的变化是从「人类反馈强化学习」（RLHF）转向「可验证奖励强化学习」（RLVR，Reinforcement Learning with Verifiable Rewards）：在高难度数学或代码问题上，人类很难给出精准反馈，而 …",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "reinforcement",
    "mas",
    "humanoid",
    "sp2026",
    "lectures",
    "offline",
    "grpo",
    "motion"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 24,
   "related": [
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.2203
    },
    {
     "title": "后训练与对齐方法",
     "url": "ai/post-training-and-alignment-methods.html",
     "score": 0.2155
    },
    {
     "title": "合成数据与数据引擎",
     "url": "ai/synthetic-data-and-data-engines.html",
     "score": 0.1131
    },
    {
     "title": "AI 评测实践",
     "url": "ai/ai-evaluation-practices.html",
     "score": 0.0924
    }
   ]
  },
  {
   "i": 32,
   "title": "成本与性能工程",
   "t": "成本与性能工程",
   "u": "ai/ai-cost-and-performance-optimization.html",
   "s": "LLM 成本与性能工程的核心是把「单位经济（unit economics）」拆解为可优化的变量。其基本关系为：每 token 成本 = 运营成本 ÷ 处理的 token 数，因此降低成本要么减少总成本，要么提升吞吐；改进该指标既可以降本、也可以提高处理的 token 量，或两者兼施。例如 LLM 基础设施月成本约 5 万美元、当月处理 500 亿 token…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "caching",
    "lang",
    "costs",
    "routing",
    "tencent",
    "self-hosted",
    "codersera",
    "foundry"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "推理优化与服务：从 PagedAttention 到分离式部署",
     "url": "ai/inference-optimization-and-serving.html",
     "score": 0.1527
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.1256
    },
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.1045
    },
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.0923
    }
   ]
  },
  {
   "i": 33,
   "title": "扩散与图像生成",
   "t": "扩散与图像生成",
   "u": "ai/diffusion-and-image-generation.html",
   "s": "扩散模型（Diffusion Model）已成为文生图（Text-to-Image）与图像编辑的主流范式。其核心思想是通过前向加噪与反向去噪学习数据分布，早期由 DDPM（Denoising Diffusion Probabilistic Models）与 DDIM（Denoising Diffusion Implicit Models）奠定基础；随后 La…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "banana",
    "diffusion",
    "nano",
    "dit",
    "klein",
    "flux",
    "visual",
    "latent"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.2077
    },
    {
     "title": "视频生成与世界模型",
     "url": "ai/video-generation-and-world-models.html",
     "score": 0.1452
    },
    {
     "title": "3D 与高斯泼溅",
     "url": "ai/3d-and-gaussian-splatting.html",
     "score": 0.1252
    },
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.0919
    }
   ]
  },
  {
   "i": 34,
   "title": "推理优化与服务：从 PagedAttention 到分离式部署",
   "t": "推理优化与服务：从 PagedAttention 到分离式部署",
   "u": "ai/inference-optimization-and-serving.html",
   "s": "大语言模型推理服务（LLM Serving）的目标是在给定 GPU 预算下同时优化两个相互制约的指标： 吞吐量（throughput，tokens/s） 与 时延（latency，TTFT / TPOT） 。推理过程分为两个阶段—— prefill （并行处理全部输入 token，计算密集）与 decode （逐 token 自回归生成，显存带宽密集）。两阶…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "sglang",
    "ttft",
    "prefill",
    "speculative",
    "vllm",
    "tok",
    "throughput",
    "tco"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "开源 AI 工具链",
     "url": "ai/open-source-ai-tooling.html",
     "score": 0.2116
    },
    {
     "title": "AI 基础设施与算力",
     "url": "ai/ai-infrastructure-and-compute.html",
     "score": 0.1571
    },
    {
     "title": "成本与性能工程",
     "url": "ai/ai-cost-and-performance-optimization.html",
     "score": 0.1527
    },
    {
     "title": "量化与模型压缩",
     "url": "ai/quantization-and-model-compression.html",
     "score": 0.147
    }
   ]
  },
  {
   "i": 35,
   "title": "推理模型与测试时计算（Test-Time Compute）",
   "t": "推理模型与测试时计算（Test-Time Compute）",
   "u": "ai/reasoning-and-test-time-compute.html",
   "s": "推理模型（reasoning model）与测试时计算（test-time compute，TTS）是 2024 年以来的核心范式转变：模型不再「一次性作答」，而是在给出最终答案前生成一段可见或隐藏的长链推理（chain of thought），并允许通过增加推理时的算力来换取更高的正确率。OpenAI 在 2024 年 9 月推出 o1，用大规模强化学习训…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 18,
   "tags": [
    "test-time",
    "grpo",
    "rlvr",
    "thinking",
    "o4-mini",
    "rewards",
    "gpt-5",
    "effort"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目 / 产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 43,
   "related": [
    {
     "title": "强化学习前沿",
     "url": "ai/reinforcement-learning-frontier.html",
     "score": 0.2203
    },
    {
     "title": "通用大模型前沿格局（2025–2026）",
     "url": "ai/llm-frontier-2026.html",
     "score": 0.2071
    },
    {
     "title": "后训练与对齐方法",
     "url": "ai/post-training-and-alignment-methods.html",
     "score": 0.2019
    },
    {
     "title": "大模型评测基准（Benchmarks）",
     "url": "ai/ai-benchmarks.html",
     "score": 0.2001
    }
   ]
  },
  {
   "i": 36,
   "title": "推荐系统与 AI（Recommender Systems & AI）",
   "t": "推荐系统与 AI（Recommender Systems & AI）",
   "u": "ai/recommender-systems-and-ai.html",
   "s": "推荐系统（Recommender Systems）从早期的协同过滤（Collaborative Filtering）与矩阵分解，演进到深度学习排序模型，再到当前以大语言模型（LLM）与生成式架构为核心的新范式。传统工业推荐是多级级联架构（召回 retrieval、粗排、精排 ranking、竞拍 auction），在工业界取得显著成功，但受限于语义鸿沟、阶段…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "recommendation",
    "generative",
    "pinterest",
    "youtube",
    "app",
    "technical",
    "survey",
    "scaling"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 28,
   "related": [
    {
     "title": "预训练与缩放定律",
     "url": "ai/pretraining-and-scaling-laws.html",
     "score": 0.1167
    },
    {
     "title": "Transformer 架构原理",
     "url": "ai/transformer-architecture.html",
     "score": 0.1117
    },
    {
     "title": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
     "url": "ai/long-context-techniques.html",
     "score": 0.106
    },
    {
     "title": "检索深入：稀疏、稠密、混合检索与向量搜索（Retrieval & Vector Search）",
     "url": "ai/retrieval-and-vector-search.html",
     "score": 0.103
    }
   ]
  },
  {
   "i": 37,
   "title": "提示工程与上下文工程（Prompt & Context Engineering）",
   "t": "提示工程与上下文工程（Prompt & Context Engineering）",
   "u": "ai/prompt-and-context-engineering.html",
   "s": "提示工程（Prompt Engineering）指通过设计输入文本、指令格式与示例来引导大语言模型（LLM）产生期望输出的方法体系；上下文工程（Context Engineering）则是范围更宽的概念，关注在推理时向模型提供的 全部 信息——系统提示、工具定义、记忆、检索结果、文件与 Skills 等——如何被组装、压缩与管理（The new rules …",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 15,
   "tags": [
    "reflective",
    "chapters",
    "self-consistency",
    "fundamentals",
    "context",
    "prompt",
    "engineering",
    "effective"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目 / 工具",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "Agent 工程实践（Agent Engineering Patterns）",
     "url": "ai/agent-engineering-patterns.html",
     "score": 0.195
    },
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.1832
    },
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.1232
    },
    {
     "title": "AI Agent（智能体）技术与生态",
     "url": "ai/ai-agents.html",
     "score": 0.1224
    }
   ]
  },
  {
   "i": 38,
   "title": "检索深入：稀疏、稠密、混合检索与向量搜索（Retrieval & Vector Search）",
   "t": "检索深入：稀疏、稠密、混合检索与向量搜索（Retrieval & Vector Search）",
   "u": "ai/retrieval-and-vector-search.html",
   "s": "现代检索栈（尤其是 RAG 场景）通常由三层构成：候选召回（稀疏词法检索、稠密语义检索或两者混合）、重排序（reranking）与向量索引（ANN）。TREC RAG 2025 的参赛报告把混合检索系统描述为\"融合稀疏检索、稠密语义匹配与交叉编码器重排\"，并指出该组合在多数指标上取得最佳表现（NITATREC at TREC RAG 2025）。一项面向文本…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "mteb",
    "hnsw",
    "retrieval",
    "bm25",
    "rrf",
    "colbert",
    "graphrag",
    "qdrant"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 36,
   "related": [
    {
     "title": "分词与嵌入",
     "url": "ai/tokenization-and-embeddings.html",
     "score": 0.2333
    },
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.2289
    },
    {
     "title": "图学习与知识图谱",
     "url": "ai/graph-neural-networks-and-knowledge-graphs.html",
     "score": 0.1071
    },
    {
     "title": "推荐系统与 AI（Recommender Systems & AI）",
     "url": "ai/recommender-systems-and-ai.html",
     "score": 0.103
    }
   ]
  },
  {
   "i": 39,
   "title": "视频生成与世界模型",
   "t": "视频生成与世界模型",
   "u": "ai/video-generation-and-world-models.html",
   "s": "视频生成模型在过去两年从「短片段、无声音」快速演进到「多镜头、带原生音频、可交互世界」的阶段。技术路线上，主流模型普遍采用 DiT（Diffusion Transformer）在潜空间做时空去噪，并逐步引入自回归、多模态联合训练与实时交互能力。与之并行的是「世界模型（World Model）」方向：不再只输出一段视频，而是生成可被实时导航、具有持久性（per…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "veo",
    "sora",
    "kling",
    "marble",
    "genie",
    "video",
    "fps",
    "world"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.2641
    },
    {
     "title": "扩散与图像生成",
     "url": "ai/diffusion-and-image-generation.html",
     "score": 0.1452
    },
    {
     "title": "3D 与高斯泼溅",
     "url": "ai/3d-and-gaussian-splatting.html",
     "score": 0.1156
    },
    {
     "title": "强化学习前沿",
     "url": "ai/reinforcement-learning-frontier.html",
     "score": 0.0914
    }
   ]
  },
  {
   "i": 40,
   "title": "软硬协同设计（AI Hardware-Software Co-Design）",
   "t": "软硬协同设计（AI Hardware-Software Co-Design）",
   "u": "ai/ai-hardware-software-codesign.html",
   "s": "软硬协同设计（co-design）指把模型架构、数值精度、编译栈与芯片微架构作为同一系统联合优化，而非「先定模型再适配硬件」。在 LLM 推理成本由内存带宽、互连与算子效率主导的背景下，协同设计被视为提升单位算力产出的关键路径。其技术抓手包括：面向硬件友好的模型结构设计（稀疏、低精度、算子规整化）、编译器与运行时（XLA、TVM、MLIR、Triton）、以…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 14,
   "tags": [
    "co-design",
    "mlir",
    "hexagon-mlir",
    "nvfp4",
    "jax",
    "ironwood",
    "loop",
    "groq"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "GPU 编程与算子：CUDA、Triton 与 FlashAttention",
     "url": "ai/gpu-kernels-and-cuda.html",
     "score": 0.2122
    },
    {
     "title": "量化与模型压缩",
     "url": "ai/quantization-and-model-compression.html",
     "score": 0.1923
    },
    {
     "title": "AI 基础设施与算力",
     "url": "ai/ai-infrastructure-and-compute.html",
     "score": 0.1369
    },
    {
     "title": "开源 AI 工具链",
     "url": "ai/open-source-ai-tooling.html",
     "score": 0.1123
    }
   ]
  },
  {
   "i": 41,
   "title": "通用大模型前沿格局（2025–2026）",
   "t": "通用大模型前沿格局（2025–2026）",
   "u": "ai/llm-frontier-2026.html",
   "s": "截至 2026 年 9 月，通用大模型（frontier LLM）的竞争格局已从\"单一旗舰迭代\"演变为\"多厂商、多层次、按月刷新\"的密集竞赛。头部厂商包括 OpenAI、Anthropic、Google DeepMind、xAI（后并入 SpaceX 体系）、Meta，以及中国的 DeepSeek、阿里云 Qwen、月之暗面 Kimi、字节跳动豆包、智谱 G…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 26,
   "tags": [
    "gpt-6",
    "astra",
    "kimi",
    "pricing",
    "fable",
    "grok",
    "opus",
    "sol"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/产品（带官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 81,
   "related": [
    {
     "title": "大模型评测基准（Benchmarks）",
     "url": "ai/ai-benchmarks.html",
     "score": 0.2479
    },
    {
     "title": "开放权重（开源）模型生态",
     "url": "ai/open-weight-models.html",
     "score": 0.2357
    },
    {
     "title": "推理模型与测试时计算（Test-Time Compute）",
     "url": "ai/reasoning-and-test-time-compute.html",
     "score": 0.2071
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1962
    }
   ]
  },
  {
   "i": 42,
   "title": "量化与模型压缩",
   "t": "量化与模型压缩",
   "u": "ai/quantization-and-model-compression.html",
   "s": "量化（Quantization）把模型权重与激活值从高精度浮点（FP16/BF16/FP32）转换为低位宽表示（FP8/INT8/INT4/FP4，甚至 INT2），以换取更小的显存占用、更高的显存带宽利用率与更高的 Tensor Core 吞吐。它是当前大模型落地中最\"性价比\"最高的优化手段：通常只需少量校准数据，不需重训，即可在几乎无精度损失的前提下显著…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "quantization",
    "post-training",
    "awq",
    "nvfp4",
    "w8a8",
    "fp4",
    "only",
    "e4m3"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "软硬协同设计（AI Hardware-Software Co-Design）",
     "url": "ai/ai-hardware-software-codesign.html",
     "score": 0.1923
    },
    {
     "title": "Transformer 架构原理",
     "url": "ai/transformer-architecture.html",
     "score": 0.1675
    },
    {
     "title": "开源 AI 工具链",
     "url": "ai/open-source-ai-tooling.html",
     "score": 0.1593
    },
    {
     "title": "推理优化与服务：从 PagedAttention 到分离式部署",
     "url": "ai/inference-optimization-and-serving.html",
     "score": 0.147
    }
   ]
  },
  {
   "i": 43,
   "title": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
   "t": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
   "u": "ai/long-context-techniques.html",
   "s": "长上下文（Long Context）指让大语言模型在一次前向计算中处理 128K、1M 乃至更长 token 序列的能力。其核心矛盾在于标准自注意力的计算与显存开销随序列长度呈二次增长，同时位置编码（Position Embedding）在超出预训练长度后出现分布外退化（extrapolation failure）。 工程上解决该矛盾的三条主线是：（1） 位…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "rope",
    "mamba",
    "deepseek-v4",
    "yarn",
    "long-context",
    "ruler",
    "nemotron",
    "ssm"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 33,
   "related": [
    {
     "title": "Transformer 架构原理",
     "url": "ai/transformer-architecture.html",
     "score": 0.3267
    },
    {
     "title": "RAG 与上下文工程（Context Engineering）与智能体记忆",
     "url": "ai/rag-and-memory.html",
     "score": 0.1536
    },
    {
     "title": "推理优化与服务：从 PagedAttention 到分离式部署",
     "url": "ai/inference-optimization-and-serving.html",
     "score": 0.1286
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.1286
    }
   ]
  },
  {
   "i": 44,
   "title": "音乐与音频生成",
   "t": "音乐与音频生成",
   "u": "ai/audio-music-and-speech-generation.html",
   "s": "音乐与音频生成覆盖三条主线：歌曲生成（Suno、Udio 等）、语音合成与识别（TTS/ASR，如 ElevenLabs、VibeVoice、Whisper 生态），以及音效与通用音频生成（Stable Audio、MusicGen 等）。2025–2026 年最显著的产业变化是「版权合规化」：头部音乐生成公司与唱片公司达成授权合作，并陆续以「完全授权训练数…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "audio",
    "elevenlabs",
    "music",
    "stable",
    "news-updates",
    "synthesis",
    "stability",
    "meet"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "对话助手与语音（Conversational Assistants & Voice）",
     "url": "ai/conversational-assistants-and-voice.html",
     "score": 0.2248
    },
    {
     "title": "多模态 AI（Multimodal AI）",
     "url": "ai/multimodal-ai.html",
     "score": 0.1248
    },
    {
     "title": "视频生成与世界模型",
     "url": "ai/video-generation-and-world-models.html",
     "score": 0.0835
    },
    {
     "title": "扩散与图像生成",
     "url": "ai/diffusion-and-image-generation.html",
     "score": 0.0768
    }
   ]
  },
  {
   "i": 45,
   "title": "预训练与缩放定律",
   "t": "预训练与缩放定律",
   "u": "ai/pretraining-and-scaling-laws.html",
   "s": "预训练是 LLM 通过海量文本以自监督方式学习通用语言能力的基础阶段。其核心工程问题是\"算力—数据—参数\"三者的最优配比，以及如何在超大规模下保持训练稳定、并保证数据质量。缩放定律（Scaling Laws）为这一配比提供了经验性数学指引。 最新进展（2025–2026） 2025–2026 年，业界对缩放定律的实践从\"纯训练算力最优\"转向\"面向部署的产品最…",
   "dm": "人工智能（AI）",
   "d": "ai",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "pretraining",
    "laws",
    "pre-training",
    "herd",
    "scaling",
    "loss",
    "llama",
    "training"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 28,
   "related": [
    {
     "title": "合成数据与数据引擎",
     "url": "ai/synthetic-data-and-data-engines.html",
     "score": 0.1243
    },
    {
     "title": "推荐系统与 AI（Recommender Systems & AI）",
     "url": "ai/recommender-systems-and-ai.html",
     "score": 0.1167
    },
    {
     "title": "MoE 稀疏架构",
     "url": "ai/mixture-of-experts.html",
     "score": 0.1039
    },
    {
     "title": "长上下文技术：位置外推、稀疏注意力与状态空间模型",
     "url": "ai/long-context-techniques.html",
     "score": 0.1028
    }
   ]
  },
  {
   "i": 46,
   "title": "AI 集群网络（Networking for AI Clusters）",
   "t": "AI 集群网络（Networking for AI Clusters）",
   "u": "hardware/networking-for-ai-clusters.html",
   "s": "AI 集群网络负责把成千上万颗加速卡连接成一个可协同训练与推理的整体，其性能直接决定大规模并行任务的效率。业界把网络分为两个层次： Scale-up（纵向扩展） ，在机架内以高带宽、低时延的私有互连打通 GPU 之间的全互联； Scale-out（横向扩展） ，通过 InfiniBand 或以太网把大量节点连接为集群。随着 Mixture-of-Expert…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "infiniband",
    "spectrum-x",
    "switches",
    "nvlink",
    "tomahawk",
    "broadcom",
    "ethernet",
    "scale-out"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "硅光与光互连（Optical Interconnects and Photonics）",
     "url": "hardware/optical-interconnects-and-photonics.html",
     "score": 0.3346
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.1921
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.089
    },
    {
     "title": "类脑与新型计算（Neuromorphic and Novel Computing）",
     "url": "hardware/neuromorphic-and-novel-computing.html",
     "score": 0.068
    }
   ]
  },
  {
   "i": 47,
   "title": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
   "t": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
   "u": "hardware/fpga-and-reconfigurable-computing.html",
   "s": "FPGA（Field Programmable Gate Array，现场可编程门阵列）是一类可在出厂后重复编程的集成电路，其核心价值在于把硬件逻辑与数据通路按需定制，同时在需求变化时无需重新流片即可更新功能。与 ASIC（专用集成电路）相比，FPGA 单位成本更高、峰值能效通常不及定制电路，但具备上市快、可迭代、小批量经济的优势，因此在网络加速、金融交易、…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "fpga",
    "programmable",
    "array",
    "gate",
    "silver",
    "field",
    "hls",
    "amd"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 24,
   "related": [
    {
     "title": "传感器与 MEMS",
     "url": "hardware/sensors-and-mems.html",
     "score": 0.1172
    },
    {
     "title": "芯片设计与 EDA",
     "url": "hardware/chip-design-and-eda.html",
     "score": 0.1103
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.1093
    },
    {
     "title": "IoT 与嵌入式硬件",
     "url": "hardware/iot-and-embedded-hardware.html",
     "score": 0.1092
    }
   ]
  },
  {
   "i": 48,
   "title": "GPU 架构深入与 CUDA",
   "t": "GPU 架构深入与 CUDA",
   "u": "hardware/gpu-architecture-and-cuda.html",
   "s": "GPU 最初为图形渲染设计，其核心思路是用大量相对简单的流多处理器（SM，Streaming Multiprocessor）并行执行同一段代码的不同数据（SIMT，Single Instruction Multiple Threads）。与 CPU 相比，GPU 把大部分晶体管预算投入到算术单元与高带宽显存，而不是分支预测与大容量缓存，因此在矩阵乘法、卷积、…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 21,
   "tags": [
    "pflops",
    "instinct",
    "data-center",
    "hgx",
    "ironwood",
    "tpu7x",
    "vera",
    "tpus"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 63,
   "related": [
    {
     "title": "AI 集群网络（Networking for AI Clusters）",
     "url": "hardware/networking-for-ai-clusters.html",
     "score": 0.1921
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.1627
    },
    {
     "title": "硅光与光互连（Optical Interconnects and Photonics）",
     "url": "hardware/optical-interconnects-and-photonics.html",
     "score": 0.1475
    },
    {
     "title": "边缘 SoC 与 NPU",
     "url": "hardware/edge-soc-and-npu.html",
     "score": 0.1206
    }
   ]
  },
  {
   "i": 49,
   "title": "IoT 与嵌入式硬件",
   "t": "IoT 与嵌入式硬件",
   "u": "hardware/iot-and-embedded-hardware.html",
   "s": "IoT（Internet of Things，物联网）与嵌入式硬件指面向传感、连接与控制的微控制器（MCU）、无线 SoC、通信模组与工业/边缘网关等。其技术栈核心是 MCU 生态、无线协议（BLE、Thread、Wi-Fi、LoRa、NB-IoT、Matter）与边缘计算能力。IoT 与嵌入式硬件覆盖面广，一款产品往往需同时权衡功耗、成本、连接协议与安全能…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "matter",
    "mcu",
    "thread",
    "ble",
    "lora",
    "wireless",
    "wi-fi",
    "iot"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性厂商 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "传感器与 MEMS",
     "url": "hardware/sensors-and-mems.html",
     "score": 0.1228
    },
    {
     "title": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
     "url": "hardware/fpga-and-reconfigurable-computing.html",
     "score": 0.1092
    },
    {
     "title": "边缘 SoC 与 NPU",
     "url": "hardware/edge-soc-and-npu.html",
     "score": 0.0814
    },
    {
     "title": "类脑与新型计算（Neuromorphic and Novel Computing）",
     "url": "hardware/neuromorphic-and-novel-computing.html",
     "score": 0.0709
    }
   ]
  },
  {
   "i": 50,
   "title": "RISC-V 与开放硬件",
   "t": "RISC-V 与开放硬件",
   "u": "hardware/risc-v-and-open-hardware.html",
   "s": "RISC-V 是一套开放、免专利费的精简指令集架构（ISA），任何人可自由实现与扩展，被称为芯片界的\"通用标准\"，被视为突破芯片生态壁垒、发展自主可控算力的重要路线之一（中国科学院启动下一代开源芯片与系统研发（中国科学院计算技术研究所））。2025–2026 年，随着 RVA23 标准的落地、AI 相关扩展的推进以及开源芯片项目（如\"香山\"\"如意\"）的发布，…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "risc-v",
    "international",
    "automotive",
    "industries",
    "isa",
    "profile",
    "nvidia",
    "V 与开放硬件"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 组织",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "类脑与新型计算（Neuromorphic and Novel Computing）",
     "url": "hardware/neuromorphic-and-novel-computing.html",
     "score": 0.0816
    },
    {
     "title": "芯片设计与 EDA",
     "url": "hardware/chip-design-and-eda.html",
     "score": 0.0807
    },
    {
     "title": "边缘 SoC 与 NPU",
     "url": "hardware/edge-soc-and-npu.html",
     "score": 0.0737
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.0701
    }
   ]
  },
  {
   "i": 51,
   "title": "传感器与 MEMS",
   "t": "传感器与 MEMS",
   "u": "hardware/sensors-and-mems.html",
   "s": "传感器是把物理量（光、声、压力、加速度、角速度、磁场等）转换为电信号的器件，是消费电子、汽车、工业与机器人感知物理世界的基础。MEMS（Micro-Electro-Mechanical Systems，微机电系统）通过半导体微加工工艺在硅片上制造微型机械结构，是惯性传感器（加速度计、陀螺仪）、麦克风、压力传感器等的主流技术路线。CMOS 图像传感器（CIS）…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "mems",
    "cis",
    "cmos",
    "automotive",
    "imu",
    "worldwide",
    "mordor",
    "industry-reports"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "半导体材料",
     "url": "hardware/semiconductor-materials.html",
     "score": 0.1462
    },
    {
     "title": "机器人硬件与执行器",
     "url": "hardware/robot-hardware-and-actuators.html",
     "score": 0.1251
    },
    {
     "title": "IoT 与嵌入式硬件",
     "url": "hardware/iot-and-embedded-hardware.html",
     "score": 0.1228
    },
    {
     "title": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
     "url": "hardware/fpga-and-reconfigurable-computing.html",
     "score": 0.1172
    }
   ]
  },
  {
   "i": 52,
   "title": "先进封装",
   "t": "先进封装",
   "u": "hardware/advanced-packaging.html",
   "s": "当制程微缩的收益趋缓、单芯片面积受光罩尺寸与良率限制时，先进封装成为继续提升系统算力的主要路径。其核心思路是把多个裸片（die）通过硅中介层（interposer）、重布线层（RDL）、硅桥（bridge）或直接混合键合集成在一个封装内，从而在逼近单芯片的互连密度下实现异构集成——例如把逻辑计算 die 与多颗 HBM 堆叠放在一起，或把不同制程节点、不同功…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "tsmc",
    "abf",
    "ucie",
    "packaging",
    "central-libraries",
    "dedicatedfoundry",
    "die",
    "chiplet"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "半导体材料",
     "url": "hardware/semiconductor-materials.html",
     "score": 0.3066
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.167
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.1663
    },
    {
     "title": "芯片设计与 EDA",
     "url": "hardware/chip-design-and-eda.html",
     "score": 0.1562
    }
   ]
  },
  {
   "i": 53,
   "title": "光刻与半导体设备",
   "t": "光刻与半导体设备",
   "u": "hardware/lithography-and-equipment.html",
   "s": "半导体制造设备通常分为前道（晶圆制造）与后道（封装测试）两大类。前道核心环节包括光刻、刻蚀、薄膜沉积（CVD/PVD/ALD）、离子注入、扩散/退火、清洗、CMP 与量测/检测。其中光刻决定最小可分辨线宽与套刻精度，是技术门槛与资本密集度最高的环节，也是先进制程迭代的节奏控制器。 产业格局高度集中：ASML 是目前全球唯一实现 EUV（极紫外）光刻机量产供应…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 17,
   "tags": [
    "asml",
    "euv",
    "semi",
    "applied",
    "asset",
    "ald",
    "third",
    "tsmc"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "半导体材料",
     "url": "hardware/semiconductor-materials.html",
     "score": 0.2695
    },
    {
     "title": "先进封装",
     "url": "hardware/advanced-packaging.html",
     "score": 0.1663
    },
    {
     "title": "芯片设计与 EDA",
     "url": "hardware/chip-design-and-eda.html",
     "score": 0.1388
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.1358
    }
   ]
  },
  {
   "i": 54,
   "title": "半导体材料",
   "t": "半导体材料",
   "u": "hardware/semiconductor-materials.html",
   "s": "半导体材料分为晶圆制造材料（wafer fab materials）与封装材料（packaging materials）两大类。晶圆制造材料主要包括硅片、光刻胶及辅材（显影液、增粘剂等）、电子特种气体、湿化学品、CMP 抛光液与抛光垫、溅射靶材、前驱体、掩膜版等；封装材料主要包括 ABF 载板、塑封料、键合丝、引线框架、底部填充胶等。 SEMI 数据显示，2…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "abf",
    "semi",
    "wafer",
    "arf",
    "ibiden",
    "cmp",
    "unimicron",
    "euv"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "先进封装",
     "url": "hardware/advanced-packaging.html",
     "score": 0.3066
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.2695
    },
    {
     "title": "传感器与 MEMS",
     "url": "hardware/sensors-and-mems.html",
     "score": 0.1462
    },
    {
     "title": "芯片设计与 EDA",
     "url": "hardware/chip-design-and-eda.html",
     "score": 0.1226
    }
   ]
  },
  {
   "i": 55,
   "title": "存储技术",
   "t": "存储技术",
   "u": "hardware/memory-technologies.html",
   "s": "存储技术构成现代计算系统的数据层级：寄存器与 SRAM 靠近计算单元但容量小、成本高；DRAM 作为主存提供大容量与较高带宽；NAND Flash 与 SSD 提供持久化存储；外存与网络存储提供更大容量与更低成本。AI 工作负载同时挤压这一层级的顶点与中部——训练与推理需要巨量高带宽内存（HBM），而数据管道需要海量持久化容量，由此形成 2024–2026 …",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 14,
   "tags": [
    "cxl",
    "hbf",
    "hbm4",
    "dram",
    "hynix",
    "gddr7",
    "ssd",
    "hbm"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "先进封装",
     "url": "hardware/advanced-packaging.html",
     "score": 0.167
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.1627
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.1358
    },
    {
     "title": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
     "url": "hardware/fpga-and-reconfigurable-computing.html",
     "score": 0.1093
    }
   ]
  },
  {
   "i": 56,
   "title": "数据中心基础设施（Data Center Infrastructure and Power）",
   "t": "数据中心基础设施（Data Center Infrastructure and Power）",
   "u": "hardware/data-center-infrastructure-and-power.html",
   "s": "AI 训练与推理把数据中心从「IT 房地产」变成「能源与热管理工程」。芯片功耗密度、单机柜功率、供电可得性与散热方式，共同构成本轮 AI 基础设施建设的主要约束。有观点指出，2026 年开发一个 AI 数据中心首先是一个电力、散热与人才问题，其次才是地产问题：建筑本身可 12–18 个月建成，但电网接入可能需 5–7 年，且 63% 的运营者把熟练劳动力短缺…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "pue",
    "cooling",
    "grid",
    "uptime",
    "vernova",
    "data-center",
    "centers",
    "power"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "硅光与光互连（Optical Interconnects and Photonics）",
     "url": "hardware/optical-interconnects-and-photonics.html",
     "score": 0.1427
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.0917
    },
    {
     "title": "算力市场与 GPU 云（Compute Market and GPU Cloud）",
     "url": "hardware/compute-market-and-gpu-cloud.html",
     "score": 0.0667
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.0645
    }
   ]
  },
  {
   "i": 57,
   "title": "机器人硬件与执行器",
   "t": "机器人硬件与执行器",
   "u": "hardware/robot-hardware-and-actuators.html",
   "s": "机器人硬件是指支撑机器人运动与感知的物理部件体系，核心包括执行器（actuator，含电机、减速器、编码器）、灵巧手、力觉与触觉传感器、IMU、结构件与电池等。随人形机器人（humanoid）在 2025–2026 年加速走向小批量量产，其硬件成本拆解、关键零部件（尤其是谐波减速器、六维力传感器、灵巧手）的供应链与国产化成为产业焦点。 最新进展（2025–2…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "humanoid",
    "harmonic",
    "wrc",
    "drive",
    "bom",
    "机器人硬件与执行器"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与成本拆解",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "传感器与 MEMS",
     "url": "hardware/sensors-and-mems.html",
     "score": 0.1251
    },
    {
     "title": "半导体材料",
     "url": "hardware/semiconductor-materials.html",
     "score": 0.0704
    },
    {
     "title": "RISC-V 与开放硬件",
     "url": "hardware/risc-v-and-open-hardware.html",
     "score": 0.0594
    },
    {
     "title": "IoT 与嵌入式硬件",
     "url": "hardware/iot-and-embedded-hardware.html",
     "score": 0.0594
    }
   ]
  },
  {
   "i": 58,
   "title": "消费电子与移动硬件",
   "t": "消费电子与移动硬件",
   "u": "hardware/consumer-and-mobile-hardware.html",
   "s": "消费电子与移动硬件涵盖智能手机、PC（含 AI PC）、可穿戴（智能手表、手环、耳机）与 AR 眼镜等品类。其竞争核心在于手机 SoC、端侧 AI 能力、显示与光学硬件，以及出货量与市场格局的演变。2025–2026 年，AI PC 快速普及、可穿戴市场结构分化、AI 眼镜爆发，是三条主线。 最新进展（2025–2026） 手机 SoC 竞争 2025 年旗…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "dimensity",
    "snapdragon",
    "glasses",
    "elite",
    "idc",
    "mediatek",
    "shipments",
    "npu"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "边缘 SoC 与 NPU",
     "url": "hardware/edge-soc-and-npu.html",
     "score": 0.3126
    },
    {
     "title": "传感器与 MEMS",
     "url": "hardware/sensors-and-mems.html",
     "score": 0.0819
    },
    {
     "title": "IoT 与嵌入式硬件",
     "url": "hardware/iot-and-embedded-hardware.html",
     "score": 0.0708
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.0649
    }
   ]
  },
  {
   "i": 59,
   "title": "硅光与光互连（Optical Interconnects and Photonics）",
   "t": "硅光与光互连（Optical Interconnects and Photonics）",
   "u": "hardware/optical-interconnects-and-photonics.html",
   "s": "光互连（optical interconnect）承担数据中心内计算芯片与高速网络之间海量数据的光电转换与传输，是连接算力芯片与网络的关键器件。随着 AI 数据中心规模扩张，机架内与机架间的东西向流量激增，光模块的重要性随之提升。国家能源局数据显示，2025 年全国算力中心总用电量达 1700 亿千瓦时，占全社会用电量 1.6%，算力规模扩张意味着数据中心内…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "cpo",
    "optical",
    "ocs",
    "broadcom",
    "tomahawk",
    "apollo",
    "photonics",
    "npo"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "AI 集群网络（Networking for AI Clusters）",
     "url": "hardware/networking-for-ai-clusters.html",
     "score": 0.3346
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.1475
    },
    {
     "title": "数据中心基础设施（Data Center Infrastructure and Power）",
     "url": "hardware/data-center-infrastructure-and-power.html",
     "score": 0.1427
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.0832
    }
   ]
  },
  {
   "i": 60,
   "title": "算力市场与 GPU 云（Compute Market and GPU Cloud）",
   "t": "算力市场与 GPU 云（Compute Market and GPU Cloud）",
   "u": "hardware/compute-market-and-gpu-cloud.html",
   "s": "算力市场指围绕 AI 训练与推理所需加速卡及其配套基础设施形成的供需与交易体系。近两年最显著的结构性变化是「neocloud」（新型 GPU 云）崛起：CoreWeave、Lambda、Nebius、Crusoe 等厂商专门提供大规模 GPU 集群租赁，与 AWS、Azure、GCP 等综合云平台形成差异化竞争。与此同时，主权 AI（sovereign AI…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "coreweave",
    "nebius",
    "neocloud",
    "lambda",
    "b300",
    "spot",
    "backlog",
    "b200"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.1211
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.0845
    },
    {
     "title": "半导体材料",
     "url": "hardware/semiconductor-materials.html",
     "score": 0.0792
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.0751
    }
   ]
  },
  {
   "i": 61,
   "title": "类脑与新型计算（Neuromorphic and Novel Computing）",
   "t": "类脑与新型计算（Neuromorphic and Novel Computing）",
   "u": "hardware/neuromorphic-and-novel-computing.html",
   "s": "类脑与新型计算（neuromorphic and novel computing）指不依赖传统冯·诺依曼架构与二进制同步时钟的一类计算范式探索，包括神经形态芯片（脉冲神经网络、事件驱动）、近存/存内计算、光计算、DNA 存储与概率计算等方向。其共同动机是突破「存储与计算分离」带来的数据搬运瓶颈。IBM 研究指出，数十年前确立的冯·诺依曼架构决策，正成为当今 …",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "brainchip",
    "dna",
    "pulsar",
    "storage",
    "point",
    "ibm",
    "intel",
    "computing"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
     "url": "hardware/fpga-and-reconfigurable-computing.html",
     "score": 0.0993
    },
    {
     "title": "边缘 SoC 与 NPU",
     "url": "hardware/edge-soc-and-npu.html",
     "score": 0.0949
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.0928
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.0914
    }
   ]
  },
  {
   "i": 62,
   "title": "芯片设计与 EDA",
   "t": "芯片设计与 EDA",
   "u": "hardware/chip-design-and-eda.html",
   "s": "一颗现代数字芯片从需求到可制造版图，典型流程为：架构定义 → RTL 设计（Verilog/SystemVerilog/VHDL）→ 功能验证 → 逻辑综合 → DFT（可测性设计）→ 布局布线（Place & Route）→ 签核（时序 STA、DRC、LVS、EM/IR）→ GDSII 交付 → 流片（tapeout）。全流程高度依赖电子设计自动化（ED…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "eda",
    "synopsys",
    "ucie",
    "chiplet",
    "pdk",
    "cadence",
    "export",
    "bis"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 35,
   "related": [
    {
     "title": "先进封装",
     "url": "hardware/advanced-packaging.html",
     "score": 0.1562
    },
    {
     "title": "光刻与半导体设备",
     "url": "hardware/lithography-and-equipment.html",
     "score": 0.1388
    },
    {
     "title": "半导体材料",
     "url": "hardware/semiconductor-materials.html",
     "score": 0.1226
    },
    {
     "title": "FPGA 与可重构计算（FPGA and Reconfigurable Computing）",
     "url": "hardware/fpga-and-reconfigurable-computing.html",
     "score": 0.1103
    }
   ]
  },
  {
   "i": 63,
   "title": "边缘 SoC 与 NPU",
   "t": "边缘 SoC 与 NPU",
   "u": "hardware/edge-soc-and-npu.html",
   "s": "边缘 SoC（System-on-Chip）指把 CPU、GPU、NPU（Neural Processing Unit，神经网络处理单元）、ISP、DSP、内存控制器与基带等功能模块集成到单一芯片、面向智能手机、AI PC、平板、可穿戴与机器人等端侧设备的平台。随着端侧大模型兴起，NPU 算力（常以 TOPS，即每秒万亿次运算标称）成为衡量边缘 SoC 的关…",
   "dm": "硬件与半导体",
   "d": "hardware",
   "ud": "2026-09-26",
   "rm": 22,
   "tags": [
    "npu",
    "snapdragon",
    "elite",
    "mediatek",
    "hexagon",
    "tops",
    "smartphones",
    "mobile"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性产品（附官方链接）",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 58,
   "related": [
    {
     "title": "消费电子与移动硬件",
     "url": "hardware/consumer-and-mobile-hardware.html",
     "score": 0.3126
    },
    {
     "title": "GPU 架构深入与 CUDA",
     "url": "hardware/gpu-architecture-and-cuda.html",
     "score": 0.1206
    },
    {
     "title": "类脑与新型计算（Neuromorphic and Novel Computing）",
     "url": "hardware/neuromorphic-and-novel-computing.html",
     "score": 0.0949
    },
    {
     "title": "存储技术",
     "url": "hardware/memory-technologies.html",
     "score": 0.089
    }
   ]
  },
  {
   "i": 64,
   "title": "AI 原生研发流程（AI-Native Development Workflow）",
   "t": "AI 原生研发流程（AI-Native Development Workflow）",
   "u": "software/ai-native-development-workflow.html",
   "s": "AI 原生研发流程指把大模型与智能体（agent）作为默认参与者，嵌入从需求、设计、编码、评审、测试、文档到部署与事故响应的软件开发生命周期（SDLC），而非仅在编辑器里提供补全。2025–2026 年的核心结论是： AI 显著加速了“代码产出”这一环节，但能否转化为整体交付绩效，取决于下游工程基础（平台、测试、评审、可观测性）是否扎实 。DORA 2025…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "greptile",
    "dora",
    "news-insights",
    "company-news",
    "github",
    "swe-bench",
    "copilot",
    "app"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.3685
    },
    {
     "title": "测试与质量工程（Testing and Quality Engineering）",
     "url": "software/testing-and-quality-engineering.html",
     "score": 0.1373
    },
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.1331
    },
    {
     "title": "重构与遗留系统现代化（Refactoring and Legacy Modernization）",
     "url": "software/refactoring-and-legacy-modernization.html",
     "score": 0.0753
    }
   ]
  },
  {
   "i": 65,
   "title": "API 设计与协议",
   "t": "API 设计与协议",
   "u": "software/api-design-and-protocols.html",
   "s": "API 设计决定系统对外与对内的契约形态。主流风格包括面向资源的 REST、面向查询图的 GraphQL、面向高性能内部调用的 gRPC、面向 TypeScript 全栈的 tRPC，以及面向实时场景的 WebSocket/SSE。2025–2026 年的关键动态包括：OpenAPI 规范发布 3.2 并持续迭代、GraphQL 联邦（federation）…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "openapi",
    "graphql",
    "sse",
    "websocket",
    "grpc",
    "apollo",
    "rest",
    "initiative"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 组织 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 19,
   "related": [
    {
     "title": "前端与 Web 平台",
     "url": "software/frontend-and-web-platforms.html",
     "score": 0.0672
    },
    {
     "title": "微服务与分布式架构",
     "url": "software/microservices-and-distributed-architecture.html",
     "score": 0.0615
    },
    {
     "title": "事件驱动与消息系统",
     "url": "software/event-driven-and-messaging.html",
     "score": 0.0595
    },
    {
     "title": "JavaScript 与 TypeScript 生态",
     "url": "software/javascript-and-typescript-ecosystem.html",
     "score": 0.0554
    }
   ]
  },
  {
   "i": 66,
   "title": "Go 生态",
   "t": "Go 生态",
   "u": "software/go-ecosystem.html",
   "s": "Go 是由 Google 主导设计的静态类型、编译型语言，以简洁语法、内置并发原语、快速编译与单一静态二进制产物为主要卖点。其并发模型基于 goroutine 与 channel（源自 CSP 思想），运行时提供低延迟垃圾回收；工具链（ go build 、 go test 、 go mod 、 gofmt ）开箱即用。凭借 Kubernetes、Docke…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 6,
   "tags": [
    "encoding",
    "tea",
    "released",
    "kubernetes",
    "goroutine",
    "json",
    "doc",
    "containerd"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "并发与并行（Concurrency and Parallelism）",
     "url": "software/concurrency-and-parallelism.html",
     "score": 0.2172
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.2007
    },
    {
     "title": "编程语言理论与范式",
     "url": "software/programming-language-theory.html",
     "score": 0.1493
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.1302
    }
   ]
  },
  {
   "i": 67,
   "title": "Java 与 JVM 生态",
   "t": "Java 与 JVM 生态",
   "u": "software/java-and-jvm-ecosystem.html",
   "s": "Java 是一门面向对象的静态类型语言，依托 JVM（Java Virtual Machine）实现「一次编写，到处运行」。JVM 通过字节码、即时编译（JIT，C1/C2）与多种垃圾回收器在吞吐、延迟与内存之间提供可调权衡，并支持 Kotlin、Scala、Groovy、Clojure 等多语言运行。Java 拥有规模巨大的企业级生态（Spring、Jak…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "zgc",
    "jep",
    "openjdk",
    "jdk",
    "graalvm",
    "jeps",
    "java",
    "kotlin"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 19,
   "related": [
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.2351
    },
    {
     "title": "编程语言理论与范式",
     "url": "software/programming-language-theory.html",
     "score": 0.2086
    },
    {
     "title": "并发与并行（Concurrency and Parallelism）",
     "url": "software/concurrency-and-parallelism.html",
     "score": 0.1771
    },
    {
     "title": "Go 生态",
     "url": "software/go-ecosystem.html",
     "score": 0.1302
    }
   ]
  },
  {
   "i": 68,
   "title": "JavaScript 与 TypeScript 生态",
   "t": "JavaScript 与 TypeScript 生态",
   "u": "software/javascript-and-typescript-ecosystem.html",
   "s": "JavaScript（正式标准为 ECMAScript）是 Web 平台的通用语言，也是全球使用率最高的编程语言；TypeScript 在其之上增加静态类型系统，已成为现代前端与 Node.js 后端开发的行业标准。生态的核心议题包括：标准演进（TC39 提案流程）、模块系统（ESM 与 CommonJS 的互操作）、运行时竞争（Node.js、Deno、B…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "deno",
    "bun",
    "typescript",
    "npm",
    "esm",
    "javascript",
    "require",
    "pnpm"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "前端与 Web 平台",
     "url": "software/frontend-and-web-platforms.html",
     "score": 0.2621
    },
    {
     "title": "编程语言理论与范式",
     "url": "software/programming-language-theory.html",
     "score": 0.1344
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.1236
    },
    {
     "title": "Python 生态",
     "url": "software/python-ecosystem.html",
     "score": 0.1057
    }
   ]
  },
  {
   "i": 69,
   "title": "Python 生态",
   "t": "Python 生态",
   "u": "software/python-ecosystem.html",
   "s": "Python 是一门动态类型、解释执行的高级语言，以可读性与「自带电池」的标准库著称，并通过大量第三方包成为数据科学、机器学习、AI 与研究计算的事实标准。其核心实现 CPython 长期受全局解释器锁（GIL）约束，但自 Python 3.13 起引入的 free-threading 模式（PEP 703）在 3.14 中获得官方支持，配套的实验性 JIT…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 6,
   "tags": [
    "pep",
    "pip",
    "free-threading",
    "whatsnew",
    "cpython",
    "jit",
    "gil",
    "downloads"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.1889
    },
    {
     "title": "并发与并行（Concurrency and Parallelism）",
     "url": "software/concurrency-and-parallelism.html",
     "score": 0.1754
    },
    {
     "title": "JavaScript 与 TypeScript 生态",
     "url": "software/javascript-and-typescript-ecosystem.html",
     "score": 0.1057
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.0889
    }
   ]
  },
  {
   "i": 70,
   "title": "Rust 生态",
   "t": "Rust 生态",
   "u": "software/rust-ecosystem.html",
   "s": "Rust 是一门面向系统编程的静态类型语言，设计目标是在不使用垃圾回收（GC）的前提下提供内存安全、并发安全与接近 C 的性能。其核心机制是所有权（ownership）、借用检查（borrow checker）与生命周期（lifetimes）：同一时刻一个值只能有一个所有者，可变引用（ &mut T ）必须唯一，从而在编译期静态消除 use-after-fr…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "tokio",
    "crate",
    "safety-critical",
    "axum",
    "rust",
    "nightly",
    "programming",
    "foundation"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 19,
   "related": [
    {
     "title": "编程语言理论与范式",
     "url": "software/programming-language-theory.html",
     "score": 0.2698
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.1745
    },
    {
     "title": "系统编程语言：C/C++ 与 Zig、Carbon 等",
     "url": "software/cpp-and-systems-languages.html",
     "score": 0.1223
    },
    {
     "title": "并发与并行（Concurrency and Parallelism）",
     "url": "software/concurrency-and-parallelism.html",
     "score": 0.1204
    }
   ]
  },
  {
   "i": 71,
   "title": "WebAssembly 与 Web 平台",
   "t": "WebAssembly 与 Web 平台",
   "u": "software/webassembly-and-web-platform.html",
   "s": "WebAssembly（简称 Wasm）是一种可移植的二进制指令格式，最初为在浏览器中以接近原生速度运行高性能代码而设计，随后扩展到服务端、边缘计算、插件系统与嵌入式等场景。它以栈式虚拟机、线性内存模型和强类型静态校验为特征，可在不同硬件架构上高效执行（WebAssembly: How It's Transforming Modern Web Develop…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "webassembly",
    "wasm",
    "component",
    "baseline",
    "transforming",
    "draft",
    "spec",
    "javascript"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "前端与 Web 平台",
     "url": "software/frontend-and-web-platforms.html",
     "score": 0.1538
    },
    {
     "title": "云原生与 Kubernetes",
     "url": "software/cloud-native-and-kubernetes.html",
     "score": 0.1034
    },
    {
     "title": "编程语言理论与范式",
     "url": "software/programming-language-theory.html",
     "score": 0.0832
    },
    {
     "title": "Go 生态",
     "url": "software/go-ecosystem.html",
     "score": 0.0797
    }
   ]
  },
  {
   "i": 72,
   "title": "事件驱动与消息系统",
   "t": "事件驱动与消息系统",
   "u": "software/event-driven-and-messaging.html",
   "s": "事件驱动架构（EDA）以事件的产生、传输与消费组织系统，实现服务间的松耦合、异步与可扩展。其技术底座是消息与流平台（Kafka、Pulsar、NATS、RabbitMQ 等），上层常见模式包括事件溯源（Event Sourcing）、CQRS、发布/订阅与队列。2025–2026 年的关键动态包括：Apache Kafka 4.x 彻底转向 KRaft、交付…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "kafka",
    "sourcing",
    "pulsar",
    "kraft",
    "confluent",
    "event",
    "zookeeper",
    "exactly-once"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 组织 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "微服务与分布式架构",
     "url": "software/microservices-and-distributed-architecture.html",
     "score": 0.071
    },
    {
     "title": "领域驱动设计（DDD）",
     "url": "software/domain-driven-design.html",
     "score": 0.0626
    },
    {
     "title": "缓存与存储策略（Caching and Storage Strategies）",
     "url": "software/caching-and-storage-strategies.html",
     "score": 0.0621
    },
    {
     "title": "软件架构模式",
     "url": "software/software-architecture-patterns.html",
     "score": 0.0617
    }
   ]
  },
  {
   "i": 73,
   "title": "云原生与 Kubernetes",
   "t": "云原生与 Kubernetes",
   "u": "software/cloud-native-and-kubernetes.html",
   "s": "云原生（Cloud Native）已从前沿概念演变为企业基础设施的默认范式。根据 CNCF 于 2026 年 1 月 20 日发布的《CNCF Annual Cloud Native Survey》，98% 的受访组织已采用云原生技术，82% 的容器用户在生产环境运行 Kubernetes，较 2023 年的 66% 显著提升，报告称 Kubernetes …",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "ambient",
    "cncf",
    "istio",
    "kubernetes",
    "argo",
    "sidecar",
    "gateway",
    "backstage"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/框架",
    "版本与生态数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 35,
   "related": [
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.1901
    },
    {
     "title": "WebAssembly 与 Web 平台",
     "url": "software/webassembly-and-web-platform.html",
     "score": 0.1034
    },
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.103
    },
    {
     "title": "Go 生态",
     "url": "software/go-ecosystem.html",
     "score": 0.1004
    }
   ]
  },
  {
   "i": 74,
   "title": "低延迟系统",
   "t": "低延迟系统",
   "u": "software/low-latency-systems.html",
   "s": "低延迟系统追求从微秒到纳秒级的确定性响应，广泛用于高频交易（HFT）、实时音视频、工业控制与电信前传等场景。其核心技术路线包括 内核旁路（kernel bypass） 、 用户态网络栈（DPDK / AF_XDP） 、 硬件卸载（FPGA/ASIC） 、 异步 I/O（io_uring） 以及 介质与部署拓扑优化 。2025–2026 年，这些技术的关注点从…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "io_uring",
    "webrtc",
    "trading",
    "ebpf",
    "bypass",
    "ice",
    "networking",
    "driver"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 22,
   "related": [
    {
     "title": "性能工程（Performance Engineering）",
     "url": "software/performance-engineering.html",
     "score": 0.1085
    },
    {
     "title": "云原生与 Kubernetes",
     "url": "software/cloud-native-and-kubernetes.html",
     "score": 0.0679
    },
    {
     "title": "嵌入式系统与 RTOS",
     "url": "software/embedded-systems-and-rtos.html",
     "score": 0.0618
    },
    {
     "title": "WebAssembly 与 Web 平台",
     "url": "software/webassembly-and-web-platform.html",
     "score": 0.0565
    }
   ]
  },
  {
   "i": 75,
   "title": "前端与 Web 平台",
   "t": "前端与 Web 平台",
   "u": "software/frontend-and-web-platforms.html",
   "s": "2025–2026 年的前端与 Web 平台呈现出三条主线： 框架编译化与去虚拟 DOM 化 （Svelte Runes、Vue Vapor、Solid 编译器）、 工具链 Rust 化 （Vite 8/Rolldown、Oxc、Turbopack、TypeScript 7 Go 移植）以及 AI 深度介入开发流程 。State of JS 2025 调研显…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "vite",
    "deno",
    "pnpm",
    "solid",
    "webassembly",
    "typescript",
    "components",
    "bun"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/框架",
    "版本与生态数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "JavaScript 与 TypeScript 生态",
     "url": "software/javascript-and-typescript-ecosystem.html",
     "score": 0.2621
    },
    {
     "title": "WebAssembly 与 Web 平台",
     "url": "software/webassembly-and-web-platform.html",
     "score": 0.1538
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.12
    },
    {
     "title": "桌面与跨平台应用",
     "url": "software/desktop-and-cross-platform-apps.html",
     "score": 0.1063
    }
   ]
  },
  {
   "i": 76,
   "title": "单仓与构建系统（Monorepo and Build Systems）",
   "t": "单仓与构建系统（Monorepo and Build Systems）",
   "u": "software/monorepo-and-build-systems.html",
   "s": "单仓（monorepo）指把多个项目、服务、库放在同一个版本控制仓库中统一管理；构建系统则负责在这种规模下正确、增量、可缓存的构建与测试。单仓的优势在于共享依赖与协同变更（coordinated changes），但对「各自独立、技术栈异构」的团队未必合适，因此工具选择需要结合团队规模、项目复杂度与语言需求（Monorepo in 2026: Turbore…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "comparisons",
    "remote",
    "forum",
    "workspace",
    "rules",
    "action",
    "documentation",
    "build"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 19,
   "related": [
    {
     "title": "前端与 Web 平台",
     "url": "software/frontend-and-web-platforms.html",
     "score": 0.082
    },
    {
     "title": "微服务与分布式架构",
     "url": "software/microservices-and-distributed-architecture.html",
     "score": 0.0615
    },
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.0611
    },
    {
     "title": "测试与质量工程（Testing and Quality Engineering）",
     "url": "software/testing-and-quality-engineering.html",
     "score": 0.0609
    }
   ]
  },
  {
   "i": 77,
   "title": "后端语言与运行时",
   "t": "后端语言与运行时",
   "u": "software/backend-languages-and-runtimes.html",
   "s": "后端语言与运行时的 2025–2026 主线是 并发模型的普及化 与 性能的底层重写 ：Java 的虚拟线程走向常态、Python 进入无 GIL 时代、Go 与 Rust 补齐泛型与异步短板，同时各语言工具链（编译器、包管理、运行时）大量从脚本语言迁移到原生实现。在云原生与 AI 推理场景下，语言选择越来越多地由冷启动、内存占用与并发吞吐共同决定。 202…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "tokio",
    "loom",
    "scala",
    "jdk",
    "graalvm",
    "zig",
    "round",
    "net"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/框架",
    "版本与生态数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 33,
   "related": [
    {
     "title": "并发与并行（Concurrency and Parallelism）",
     "url": "software/concurrency-and-parallelism.html",
     "score": 0.317
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.2351
    },
    {
     "title": "Go 生态",
     "url": "software/go-ecosystem.html",
     "score": 0.2007
    },
    {
     "title": "Python 生态",
     "url": "software/python-ecosystem.html",
     "score": 0.1889
    }
   ]
  },
  {
   "i": 78,
   "title": "嵌入式系统与 RTOS",
   "t": "嵌入式系统与 RTOS",
   "u": "software/embedded-systems-and-rtos.html",
   "s": "嵌入式系统以 MCU（微控制器）为核心，围绕实时操作系统（RTOS）、固件实时性、功能安全认证与 IoT 协议栈构建。2025–2026 年，Zephyr 与 FreeRTOS 两条开源主线分别通过新架构与长期支持（LTS）版本推进企业级落地；安全认证（ISO 26262、IEC 61508、IEC 62304、EN 50128）成为商用 RTOS 的竞争焦…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "matter",
    "rtos",
    "risc-v",
    "thread",
    "lts",
    "mcu",
    "rts",
    "iec"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 24,
   "related": [
    {
     "title": "Rust 生态",
     "url": "software/rust-ecosystem.html",
     "score": 0.0983
    },
    {
     "title": "系统编程语言：C/C++ 与 Zig、Carbon 等",
     "url": "software/cpp-and-systems-languages.html",
     "score": 0.0736
    },
    {
     "title": "WebAssembly 与 Web 平台",
     "url": "software/webassembly-and-web-platform.html",
     "score": 0.0707
    },
    {
     "title": "云原生与 Kubernetes",
     "url": "software/cloud-native-and-kubernetes.html",
     "score": 0.0642
    }
   ]
  },
  {
   "i": 79,
   "title": "平台工程与 DevOps",
   "t": "平台工程与 DevOps",
   "u": "software/platform-engineering-and-devops.html",
   "s": "2025–2026 年，DevOps 的重心从「工具链拼接」转向 平台工程（Platform Engineering） 与 AI 辅助运维 。平台工程的核心是把内部平台当作产品来经营，用「黄金路径（golden paths）」替代临时脚本与一次性自动化（Platform engineering: how internal developer platform…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "cncf",
    "grafana",
    "slsa",
    "backstage",
    "sigstore",
    "terraform",
    "idp",
    "actions"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/平台",
    "版本与生态数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.2118
    },
    {
     "title": "云原生与 Kubernetes",
     "url": "software/cloud-native-and-kubernetes.html",
     "score": 0.1901
    },
    {
     "title": "AI 原生研发流程（AI-Native Development Workflow）",
     "url": "software/ai-native-development-workflow.html",
     "score": 0.1331
    },
    {
     "title": "性能工程（Performance Engineering）",
     "url": "software/performance-engineering.html",
     "score": 0.0919
    }
   ]
  },
  {
   "i": 80,
   "title": "并发与并行（Concurrency and Parallelism）",
   "t": "并发与并行（Concurrency and Parallelism）",
   "u": "software/concurrency-and-parallelism.html",
   "s": "并发（concurrency）指多个任务在时间上重叠推进，并行（parallelism）指多个任务在同一时刻真正同时执行。并发模型讨论的是「如何组织任务、如何共享状态、如何取消与传播错误」；并行则关心如何把工作切分到多核与多机上。 主流实现路线可分为几类：共享内存 + 线程/锁（Java、C++）、异步任务 + 事件循环（Rust async、JavaScr…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "tokio",
    "concurrency",
    "jep",
    "jeps",
    "threads",
    "gil",
    "reactive",
    "goroutine"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 22,
   "related": [
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.317
    },
    {
     "title": "Go 生态",
     "url": "software/go-ecosystem.html",
     "score": 0.2172
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.1771
    },
    {
     "title": "Python 生态",
     "url": "software/python-ecosystem.html",
     "score": 0.1754
    }
   ]
  },
  {
   "i": 81,
   "title": "开发体验与工具（Developer Experience and Tooling）",
   "t": "开发体验与工具（Developer Experience and Tooling）",
   "u": "software/developer-experience-and-tooling.html",
   "s": "开发体验（Developer Experience，DX）指开发者在使用工具链、平台与流程时感受到的效率、认知负荷与满意度。它从早期的“编辑器偏好”话题，逐步演变为覆盖 IDE、AI 编程助手、开发容器（devcontainer）、调试器、文档体系、DX 度量与内部开发者平台（Internal Developer Platform，IDP）的系统性工程学科。…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "dora",
    "cncf",
    "backstage",
    "news-insights",
    "container",
    "cursor",
    "copilot",
    "overflow"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 27,
   "related": [
    {
     "title": "AI 原生研发流程（AI-Native Development Workflow）",
     "url": "software/ai-native-development-workflow.html",
     "score": 0.3685
    },
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.2118
    },
    {
     "title": "云原生与 Kubernetes",
     "url": "software/cloud-native-and-kubernetes.html",
     "score": 0.103
    },
    {
     "title": "测试与质量工程（Testing and Quality Engineering）",
     "url": "software/testing-and-quality-engineering.html",
     "score": 0.0928
    }
   ]
  },
  {
   "i": 82,
   "title": "开源许可与治理（Open Source Licensing and Governance）",
   "t": "开源许可与治理（Open Source Licensing and Governance）",
   "u": "software/open-source-licensing-and-governance.html",
   "s": "开源许可与治理研究的是三组相互纠缠的问题：代码以何种法律许可发布（许可证类型与兼容性）、项目由谁决策与维护（基金会、公司或社区治理）、以及使用方如何履行义务并控制风险（合规、SBOM、审计）。2025–2026 年，这一领域新增了两条主线： AI 与开源的边界之争 （“开放权重是否等于开源”“用开源代码训练模型是否传播许可证义务”），以及 许可证“再开放化”…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "openssf",
    "redis",
    "agplv3",
    "osi",
    "licenses",
    "hashicorp",
    "legal",
    "linux-foundation"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "缓存与存储策略（Caching and Storage Strategies）",
     "url": "software/caching-and-storage-strategies.html",
     "score": 0.1105
    },
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.0602
    },
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.0602
    },
    {
     "title": "AI 原生研发流程（AI-Native Development Workflow）",
     "url": "software/ai-native-development-workflow.html",
     "score": 0.0519
    }
   ]
  },
  {
   "i": 83,
   "title": "微服务与分布式架构",
   "t": "微服务与分布式架构",
   "u": "software/microservices-and-distributed-architecture.html",
   "s": "微服务架构把系统拆分为一组围绕业务能力组织、可独立部署的小型服务，配合轻量级通信机制与去中心化治理。它带来独立伸缩、故障隔离与技术异构等收益，也带来分布式事务、跨服务一致性与运维复杂度等代价。2025–2026 年的显著变化是：行业从「无条件拥抱微服务」转向更务实的取舍，模块化单体（modular monolith）与单体回归成为主流讨论，同时韧性模式（熔断…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "microservices",
    "monolith",
    "modular",
    "span",
    "opentelemetry",
    "patterns",
    "prime",
    "cncf"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 组织 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "软件架构模式",
     "url": "software/software-architecture-patterns.html",
     "score": 0.3414
    },
    {
     "title": "领域驱动设计（DDD）",
     "url": "software/domain-driven-design.html",
     "score": 0.1398
    },
    {
     "title": "性能工程（Performance Engineering）",
     "url": "software/performance-engineering.html",
     "score": 0.0793
    },
    {
     "title": "事件驱动与消息系统",
     "url": "software/event-driven-and-messaging.html",
     "score": 0.071
    }
   ]
  },
  {
   "i": 84,
   "title": "性能工程（Performance Engineering）",
   "t": "性能工程（Performance Engineering）",
   "u": "software/performance-engineering.html",
   "s": "性能工程是把「延迟、吞吐、资源占用」当作一等的设计约束，并用度量驱动的方式持续验证与优化的工程学科。它区别于事后的性能调优：核心做法是先建立基线、定义延迟预算（latency budget）、再通过剖析定位瓶颈，最后以可重复的基准测试验证改动是否奏效。 在实际系统中，性能目标很少是「平均变快」，而是控制分布形态。低延迟交易系统领域的 2026 年指南特别强调…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "ebpf",
    "load",
    "grafana",
    "trace",
    "latency",
    "testing",
    "continuous",
    "performance"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "低延迟系统",
     "url": "software/low-latency-systems.html",
     "score": 0.1085
    },
    {
     "title": "并发与并行（Concurrency and Parallelism）",
     "url": "software/concurrency-and-parallelism.html",
     "score": 0.0946
    },
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.0919
    },
    {
     "title": "微服务与分布式架构",
     "url": "software/microservices-and-distributed-architecture.html",
     "score": 0.0793
    }
   ]
  },
  {
   "i": 85,
   "title": "桌面与跨平台应用",
   "t": "桌面与跨平台应用",
   "u": "software/desktop-and-cross-platform-apps.html",
   "s": "桌面应用开发主要分为三条路线： Web 技术栈封装 （Electron、Tauri）、 自绘 UI 跨平台框架 （Flutter Desktop、Qt、Compose Multiplatform）与 平台原生框架 （macOS/iOS 的 SwiftUI，Windows 的 WinUI 3）。三者在体积、内存、平台一致性、首日特性访问与安全模型上各有取舍，2…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "electron",
    "flutter",
    "chromium",
    "windows",
    "macos",
    "impeller",
    "desktop",
    "apps"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 24,
   "related": [
    {
     "title": "移动应用开发",
     "url": "software/mobile-app-development.html",
     "score": 0.2198
    },
    {
     "title": "前端与 Web 平台",
     "url": "software/frontend-and-web-platforms.html",
     "score": 0.1063
    },
    {
     "title": "游戏开发与引擎",
     "url": "software/game-development-and-engines.html",
     "score": 0.0925
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.0889
    }
   ]
  },
  {
   "i": 86,
   "title": "测试与质量工程（Testing and Quality Engineering）",
   "t": "测试与质量工程（Testing and Quality Engineering）",
   "u": "software/testing-and-quality-engineering.html",
   "s": "测试与质量工程关注如何用可重复、可自动化的手段验证软件行为，并把质量约束固化进研发流程，而不是在交付末端做一次性把关。经典的分层模型是测试金字塔：底层是大量快速、隔离的单元测试，中层是数量较少的集成测试，顶层是少量缓慢但置信度高的端到端（E2E）测试（Testing Backend Services: A Practical Strategy for API…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "cypress",
    "testing",
    "tests",
    "ai-generated",
    "dora",
    "coverage",
    "discover",
    "meta"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "AI 原生研发流程（AI-Native Development Workflow）",
     "url": "software/ai-native-development-workflow.html",
     "score": 0.1373
    },
    {
     "title": "重构与遗留系统现代化（Refactoring and Legacy Modernization）",
     "url": "software/refactoring-and-legacy-modernization.html",
     "score": 0.0983
    },
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.0928
    },
    {
     "title": "平台工程与 DevOps",
     "url": "software/platform-engineering-and-devops.html",
     "score": 0.0804
    }
   ]
  },
  {
   "i": 87,
   "title": "游戏开发与引擎",
   "t": "游戏开发与引擎",
   "u": "software/game-development-and-engines.html",
   "s": "游戏开发围绕引擎、渲染、运行时架构（ECS）、网络同步与内容生产五条主线展开。2025–2026 年，主流商业引擎持续推进高保真实时渲染（Nanite/Lumen、MetaHuman），Unity 进入以 Unity 6 为基线的迭代并为 Unity 7 铺垫，开源引擎 Godot 在独立开发者中的份额明显上升；同时，AI 生成内容（AI 助手、图生 3D、…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "unity",
    "unreal",
    "game",
    "manual",
    "quantum",
    "fallback",
    "ping",
    "engine"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 26,
   "related": [
    {
     "title": "桌面与跨平台应用",
     "url": "software/desktop-and-cross-platform-apps.html",
     "score": 0.0925
    },
    {
     "title": "移动应用开发",
     "url": "software/mobile-app-development.html",
     "score": 0.0642
    },
    {
     "title": "前端与 Web 平台",
     "url": "software/frontend-and-web-platforms.html",
     "score": 0.0622
    },
    {
     "title": "WebAssembly 与 Web 平台",
     "url": "software/webassembly-and-web-platform.html",
     "score": 0.0586
    }
   ]
  },
  {
   "i": 88,
   "title": "移动应用开发",
   "t": "移动应用开发",
   "u": "software/mobile-app-development.html",
   "s": "移动应用开发目前呈现「原生优先」与「跨平台共享」并行的格局。iOS 侧以 Swift / SwiftUI 为官方技术栈，Android 侧以 Kotlin / Jetpack Compose 为官方技术栈；跨平台方向则由 Kotlin Multiplatform（KMP） 、 Flutter 、 React Native 等方案分担，分别在共享业务逻辑、共享…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "flutter",
    "multiplatform",
    "swiftui",
    "compose",
    "kotlin",
    "kmp",
    "react",
    "android"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "桌面与跨平台应用",
     "url": "software/desktop-and-cross-platform-apps.html",
     "score": 0.2198
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.0867
    },
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.0845
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.0809
    }
   ]
  },
  {
   "i": 89,
   "title": "系统编程语言：C/C++ 与 Zig、Carbon 等",
   "t": "系统编程语言：C/C++ 与 Zig、Carbon 等",
   "u": "software/cpp-and-systems-languages.html",
   "s": "系统编程语言指直接面向操作系统内核、嵌入式、编译器、数据库与高性能基础设施的语言家族，长期以 C 与 C++ 为核心。它们贴近硬件、可预测、运行时开销小，但也以手动内存管理与未定义行为（UB）著称。2025–2026 年，围绕「内存安全」的政策压力与技术演进成为该领域主线：一方面 C++ 通过标准库加固（hardening）、合约（contracts）与安全…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "zig",
    "profiles",
    "carbon",
    "c++",
    "rust",
    "safety",
    "download",
    "linux"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 组织 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "编程语言理论与范式",
     "url": "software/programming-language-theory.html",
     "score": 0.1551
    },
    {
     "title": "Rust 生态",
     "url": "software/rust-ecosystem.html",
     "score": 0.1223
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.1099
    },
    {
     "title": "Go 生态",
     "url": "software/go-ecosystem.html",
     "score": 0.0846
    }
   ]
  },
  {
   "i": 90,
   "title": "缓存与存储策略（Caching and Storage Strategies）",
   "t": "缓存与存储策略（Caching and Storage Strategies）",
   "u": "software/caching-and-storage-strategies.html",
   "s": "缓存的本质是在更快的介质上保存数据的副本，以降低延迟、减少对下游（数据库、源站）的压力。它同时引入三个必须面对的问题：一致性（副本何时过期）、容量（保存什么、淘汰什么）与可靠性（缓存故障时系统如何降级）。 在架构层面，缓存通常按层级组织：进程内缓存（L1，节点本地、微秒级、无网络开销）、分布式缓存（L2，集群共享、毫秒级）、数据库/源站作为唯一事实来源（Ap…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "redis",
    "caching",
    "cache",
    "tiered",
    "cloudflare",
    "things",
    "set",
    "vector"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 24,
   "related": [
    {
     "title": "开源许可与治理（Open Source Licensing and Governance）",
     "url": "software/open-source-licensing-and-governance.html",
     "score": 0.1105
    },
    {
     "title": "性能工程（Performance Engineering）",
     "url": "software/performance-engineering.html",
     "score": 0.0693
    },
    {
     "title": "软件架构模式",
     "url": "software/software-architecture-patterns.html",
     "score": 0.0692
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.0678
    }
   ]
  },
  {
   "i": 91,
   "title": "编程语言理论与范式",
   "t": "编程语言理论与范式",
   "u": "software/programming-language-theory.html",
   "s": "编程语言理论（Programming Language Theory, PLT）研究语言的形式语义、类型系统、内存模型与编译实现，是系统软件与工程语言设计的理论基础。其核心议题包括：静态类型与动态类型的取舍、类型系统的表达力（泛型、子类型、依赖类型、线性/仿射类型）、内存管理方式（手动、所有权、垃圾回收）、并发内存模型，以及从源码到机器码的编译流水线。 20…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "zgc",
    "swift",
    "llvm",
    "zig",
    "jdk",
    "cisa",
    "mlir",
    "typescript"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "Rust 生态",
     "url": "software/rust-ecosystem.html",
     "score": 0.2698
    },
    {
     "title": "Java 与 JVM 生态",
     "url": "software/java-and-jvm-ecosystem.html",
     "score": 0.2086
    },
    {
     "title": "后端语言与运行时",
     "url": "software/backend-languages-and-runtimes.html",
     "score": 0.1645
    },
    {
     "title": "系统编程语言：C/C++ 与 Zig、Carbon 等",
     "url": "software/cpp-and-systems-languages.html",
     "score": 0.1551
    }
   ]
  },
  {
   "i": 92,
   "title": "软件架构模式",
   "t": "软件架构模式",
   "u": "software/software-architecture-patterns.html",
   "s": "架构模式是对系统整体结构的可复用组织方式，决定依赖方向、边界划分、部署形态与演进路径。经典模式包括分层架构（Layered）、端口与适配器/六边形（Hexagonal / Ports and Adapters）、洋葱架构（Onion）与整洁架构（Clean）；部署形态上则有单体、模块化单体、微服务与 Serverless。2025–2026 年的主线是「回归…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "adr",
    "clean",
    "serverless",
    "monolith",
    "backlash",
    "thoughtworks",
    "uncategorized",
    "x4nent"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 组织 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "微服务与分布式架构",
     "url": "software/microservices-and-distributed-architecture.html",
     "score": 0.3414
    },
    {
     "title": "领域驱动设计（DDD）",
     "url": "software/domain-driven-design.html",
     "score": 0.1625
    },
    {
     "title": "测试与质量工程（Testing and Quality Engineering）",
     "url": "software/testing-and-quality-engineering.html",
     "score": 0.0726
    },
    {
     "title": "重构与遗留系统现代化（Refactoring and Legacy Modernization）",
     "url": "software/refactoring-and-legacy-modernization.html",
     "score": 0.0714
    }
   ]
  },
  {
   "i": 93,
   "title": "重构与遗留系统现代化（Refactoring and Legacy Modernization）",
   "t": "重构与遗留系统现代化（Refactoring and Legacy Modernization）",
   "u": "software/refactoring-and-legacy-modernization.html",
   "s": "重构指在不改变外部可观察行为的前提下改善代码内部结构；遗留系统现代化则把重构的粒度放大到系统级，目标是替换或改造长期运行、难以变更、技术栈过时的业务系统。两者共享同一前提：必须有办法判断「行为未变」。 Michael Feathers 在《Working Effectively with Legacy Code》中把遗留代码定义为「没有测试的代码」，而不论其…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "legacy",
    "transform",
    "ast",
    "java",
    "tags",
    "developer",
    "amazon",
    "aws"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "测试与质量工程（Testing and Quality Engineering）",
     "url": "software/testing-and-quality-engineering.html",
     "score": 0.0983
    },
    {
     "title": "AI 原生研发流程（AI-Native Development Workflow）",
     "url": "software/ai-native-development-workflow.html",
     "score": 0.0753
    },
    {
     "title": "开发体验与工具（Developer Experience and Tooling）",
     "url": "software/developer-experience-and-tooling.html",
     "score": 0.0721
    },
    {
     "title": "软件架构模式",
     "url": "software/software-architecture-patterns.html",
     "score": 0.0714
    }
   ]
  },
  {
   "i": 94,
   "title": "领域驱动设计（DDD）",
   "t": "领域驱动设计（DDD）",
   "u": "software/domain-driven-design.html",
   "s": "领域驱动设计（Domain-Driven Design, DDD）由 Eric Evans 提出，核心是在复杂业务系统中以领域模型为中心，让软件结构与业务概念对齐。其战略设计关注限界上下文（bounded context）、通用语言（ubiquitous language）与上下文映射；战术设计关注实体、值对象、聚合（aggregate）、领域事件与领域服务…",
   "dm": "软件工程",
   "d": "software",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "event",
    "domains",
    "methodology",
    "customer",
    "ibm",
    "details",
    "design",
    "architecture"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 组织 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "软件架构模式",
     "url": "software/software-architecture-patterns.html",
     "score": 0.1625
    },
    {
     "title": "微服务与分布式架构",
     "url": "software/microservices-and-distributed-architecture.html",
     "score": 0.1398
    },
    {
     "title": "事件驱动与消息系统",
     "url": "software/event-driven-and-messaging.html",
     "score": 0.0626
    },
    {
     "title": "重构与遗留系统现代化（Refactoring and Legacy Modernization）",
     "url": "software/refactoring-and-legacy-modernization.html",
     "score": 0.0524
    }
   ]
  },
  {
   "i": 95,
   "title": "OLAP 与查询引擎",
   "t": "OLAP 与查询引擎",
   "u": "data/olap-and-query-engines.html",
   "s": "OLAP（联机分析处理）与查询引擎面向大规模数据的分析与聚合查询，核心设计是列式存储加向量化执行：数据按列存储以提高压缩率与扫描效率，执行引擎以「批」（向量）为单位处理，而非逐行迭代。当前主流开源引擎包括 ClickHouse（实时 OLAP）、Apache Doris 与 StarRocks（实时数仓/湖仓查询）、DuckDB（嵌入式分析）与 Apache…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "clickhouse",
    "duckdb",
    "doris",
    "olap",
    "clickbench",
    "join",
    "tpc-h",
    "druid"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 12,
   "related": [
    {
     "title": "数据库内核",
     "url": "data/database-internals.html",
     "score": 0.2378
    },
    {
     "title": "图数据库与时序数据库",
     "url": "data/graph-and-time-series-databases.html",
     "score": 0.1351
    },
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.1139
    },
    {
     "title": "数据架构与建模",
     "url": "data/data-architecture-and-modeling.html",
     "score": 0.1072
    }
   ]
  },
  {
   "i": 96,
   "title": "分布式一致性与事务",
   "t": "分布式一致性与事务",
   "u": "data/distributed-consistency-and-transactions.html",
   "s": "分布式一致性与事务研究的是：当数据被复制到多台机器、跨多个区域甚至多个云之后，如何让系统对外表现出一致的读写语义，并在网络分区、节点故障、时钟漂移等条件下保持事务的 ACID 属性。经典理论框架包括 CAP 定理与 PACELC 扩展：CAP 只描述发生分区（P）时在可用性（A）与一致性（C）之间的取舍；PACELC 进一步指出，在没有分区的正常情况下，系统…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "raft",
    "cap",
    "lakebase",
    "aurora",
    "omni",
    "databricks",
    "compare",
    "mysql"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.1008
    },
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.0944
    },
    {
     "title": "图数据库与时序数据库",
     "url": "data/graph-and-time-series-databases.html",
     "score": 0.093
    },
    {
     "title": "特征平台与 ML 数据",
     "url": "data/feature-stores-and-ml-data.html",
     "score": 0.0838
    }
   ]
  },
  {
   "i": 97,
   "title": "分析工程与指标",
   "t": "分析工程与指标",
   "u": "data/analytics-engineering-and-metrics.html",
   "s": "分析工程（Analytics Engineering）是把软件工程实践（版本控制、测试、CI/CD、代码评审、模块化）引入数据分析领域的一门实践，其代表性工具是 dbt。指标层 / 语义层（Metrics Layer / Semantic Layer）则在此基础上解决\"同一个指标在不同工具、不同团队被重复定义且口径不一致\"的问题：把指标定义集中到建模层，由上…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "dbt",
    "semantic",
    "layer",
    "experiments",
    "datadog",
    "upgrading",
    "yaml",
    "summit"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "数据架构与建模",
     "url": "data/data-architecture-and-modeling.html",
     "score": 0.1296
    },
    {
     "title": "数据治理与隐私",
     "url": "data/data-governance-and-privacy.html",
     "score": 0.1139
    },
    {
     "title": "数据工程基础",
     "url": "data/data-engineering-fundamentals.html",
     "score": 0.0806
    },
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.0783
    }
   ]
  },
  {
   "i": 98,
   "title": "向量数据库与嵌入检索",
   "t": "向量数据库与嵌入检索",
   "u": "data/vector-databases-and-embeddings.html",
   "s": "向量数据库与嵌入检索（embeddings）是 RAG（检索增强生成）与语义搜索的存储与检索底座。其基本流程是：用嵌入模型把文本、图像等非结构化数据编码为高维向量，建立近似最近邻（ANN）索引，在查询时按向量相似度（余弦/内积/欧氏距离）检索 Top-K 结果。系统由「嵌入模型 + 索引算法 + 向量数据库」三层构成，三者共同决定召回率、延迟与成本。 需要区…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "milvus",
    "hnsw",
    "weaviate",
    "pgvector",
    "pinecone",
    "qdrant",
    "mteb",
    "spann"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "搜索引擎与信息检索",
     "url": "data/search-engines-and-information-retrieval.html",
     "score": 0.1744
    },
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.0767
    },
    {
     "title": "特征平台与 ML 数据",
     "url": "data/feature-stores-and-ml-data.html",
     "score": 0.0727
    },
    {
     "title": "数据库内核",
     "url": "data/database-internals.html",
     "score": 0.0704
    }
   ]
  },
  {
   "i": 99,
   "title": "图数据库与时序数据库",
   "t": "图数据库与时序数据库",
   "u": "data/graph-and-time-series-databases.html",
   "s": "图数据库与时序数据库是两类为特定数据形态优化的专用数据库。图数据库以节点、边、属性为基本模型，擅长多跳关系遍历、路径查找与连接数据分析，典型产品有 Neo4j、Amazon Neptune、TigerGraph、Stardog 等。时序数据库（TSDB）以时间戳为核心索引维度，针对高吞吐写入、时间窗口聚合、降采样与保留策略优化，典型产品有 InfluxDB、…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "neo4j",
    "questdb",
    "databases",
    "appendix",
    "graph",
    "comparing",
    "iot",
    "iso"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.1351
    },
    {
     "title": "数据库内核",
     "url": "data/database-internals.html",
     "score": 0.102
    },
    {
     "title": "分布式一致性与事务",
     "url": "data/distributed-consistency-and-transactions.html",
     "score": 0.093
    },
    {
     "title": "数据可观测性与质量",
     "url": "data/data-observability-and-quality.html",
     "score": 0.081
    }
   ]
  },
  {
   "i": 100,
   "title": "搜索引擎与信息检索",
   "t": "搜索引擎与信息检索",
   "u": "data/search-engines-and-information-retrieval.html",
   "s": "搜索引擎与信息检索（IR）技术负责把海量非结构化与半结构化内容组织为可检索、可排序的索引。核心技术包括倒排索引与 BM25 等词法评分模型、向量检索与近似最近邻（ANN）索引、以及把两者融合的混合检索（hybrid search）。在 RAG（检索增强生成）普及后，搜索引擎的角色从\"给人看的结果列表\"扩展到\"为模型提供上下文\"，检索质量直接决定生成质量。主流…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "elastic",
    "bm25",
    "rrf",
    "colbert",
    "late",
    "semeval-2026",
    "interaction",
    "recall"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 22,
   "related": [
    {
     "title": "向量数据库与嵌入检索",
     "url": "data/vector-databases-and-embeddings.html",
     "score": 0.1744
    },
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.0642
    },
    {
     "title": "数据库内核",
     "url": "data/database-internals.html",
     "score": 0.063
    },
    {
     "title": "图数据库与时序数据库",
     "url": "data/graph-and-time-series-databases.html",
     "score": 0.0621
    }
   ]
  },
  {
   "i": 101,
   "title": "数据可观测性与质量",
   "t": "数据可观测性与质量",
   "u": "data/data-observability-and-quality.html",
   "s": "数据可观测性（Data Observability）把软件可观测性的思路引入数据平台：通过持续监控数据本身的特征（新鲜度、数据量、schema、分布、空值率等），在数据出错时主动告警并帮助定位根因。数据质量（Data Quality）则更关注数据是否满足既定规则与业务期望，通常以测试、校验规则与 SLA/SLO 的形式表达。两者在实践中高度融合：可观测性负责…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "observability",
    "automate",
    "miss",
    "catches",
    "metrics",
    "coverage",
    "sla",
    "quality"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 12,
   "related": [
    {
     "title": "数据工程基础",
     "url": "data/data-engineering-fundamentals.html",
     "score": 0.0919
    },
    {
     "title": "特征平台与 ML 数据",
     "url": "data/feature-stores-and-ml-data.html",
     "score": 0.089
    },
    {
     "title": "数据治理与隐私",
     "url": "data/data-governance-and-privacy.html",
     "score": 0.0864
    },
    {
     "title": "图数据库与时序数据库",
     "url": "data/graph-and-time-series-databases.html",
     "score": 0.081
    }
   ]
  },
  {
   "i": 102,
   "title": "数据工程基础",
   "t": "数据工程基础",
   "u": "data/data-engineering-fundamentals.html",
   "s": "数据工程（Data Engineering）是围绕数据的采集、传输、转换、编排与质量保障构建可复用基础设施的工程学科。其核心目标是把分散在业务数据库、SaaS 系统、消息队列与日志中的原始数据，加工为可信、可分析、可被下游 BI、机器学习与 AI 应用消费的数据资产。现代数据栈通常被划分为六个层次：采集（ingestion）、处理（processing）、存…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "airflow",
    "dag",
    "star",
    "dbt",
    "cdc",
    "quality",
    "engineering",
    "apache"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "流处理与实时分析",
     "url": "data/streaming-and-real-time-analytics.html",
     "score": 0.2336
    },
    {
     "title": "特征平台与 ML 数据",
     "url": "data/feature-stores-and-ml-data.html",
     "score": 0.1184
    },
    {
     "title": "数据架构与建模",
     "url": "data/data-architecture-and-modeling.html",
     "score": 0.1154
    },
    {
     "title": "数据可观测性与质量",
     "url": "data/data-observability-and-quality.html",
     "score": 0.0919
    }
   ]
  },
  {
   "i": 103,
   "title": "数据库内核",
   "t": "数据库内核",
   "u": "data/database-internals.html",
   "s": "数据库内核研究数据在磁盘与内存中如何组织、事务如何保证正确性、查询如何被优化与执行。核心组件包括存储引擎（B+ 树或 LSM-Tree）、事务与并发控制（MVCC、WAL）、以及查询处理层（解析、优化器、执行引擎）。不同引擎的取舍可以概括为「读优化 vs 写优化」以及「行存 vs 列存」两对基本张力，现代系统往往在此基础上叠加向量化执行、可组合的查询引擎与异…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "calcite",
    "postgresql",
    "lsm",
    "aio",
    "kit",
    "duckdb",
    "rbo",
    "cbo"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 12,
   "related": [
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.2378
    },
    {
     "title": "图数据库与时序数据库",
     "url": "data/graph-and-time-series-databases.html",
     "score": 0.102
    },
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.0916
    },
    {
     "title": "流处理与实时分析",
     "url": "data/streaming-and-real-time-analytics.html",
     "score": 0.0876
    }
   ]
  },
  {
   "i": 104,
   "title": "数据架构与建模",
   "t": "数据架构与建模",
   "u": "data/data-architecture-and-modeling.html",
   "s": "数据架构与建模关注两件事：数据在系统间如何分层与流动（架构），以及数据以何种结构被组织与表达（建模）。经典的三条建模流派——Inmon 的 3NF 自顶向下、Kimball 的维度建模（星型/雪花）与 Data Vault 2.0 的企业级整合模型——至今仍是架构选型的坐标系，但在列式引擎与湖仓普及后，其物理落地形态已发生显著变化。 最新进展（2025–20…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "vault",
    "gold",
    "silver",
    "fabric",
    "scd",
    "type",
    "lakehouse",
    "dbt"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.1548
    },
    {
     "title": "分析工程与指标",
     "url": "data/analytics-engineering-and-metrics.html",
     "score": 0.1296
    },
    {
     "title": "数据工程基础",
     "url": "data/data-engineering-fundamentals.html",
     "score": 0.1154
    },
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.1072
    }
   ]
  },
  {
   "i": 105,
   "title": "数据治理与隐私",
   "t": "数据治理与隐私",
   "u": "data/data-governance-and-privacy.html",
   "s": "数据治理是围绕数据资产建立权责、标准、流程与度量的一整套机制，核心组件包括数据目录、血缘追踪、权限模型、分类分级、隐私合规与数据契约（Data Contract）。它与隐私合规高度耦合：隐私法规要求可追溯的数据处理证据、个人数据的识别与访问控制，而治理平台正是提供这些证据的技术载体。 最新进展（2025–2026） 隐私法规数量持续膨胀。 截至 2026 年…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "governance",
    "ireland",
    "gdpr",
    "pii",
    "korea",
    "legal-content",
    "txt",
    "uri"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "分析工程与指标",
     "url": "data/analytics-engineering-and-metrics.html",
     "score": 0.1139
    },
    {
     "title": "数据可观测性与质量",
     "url": "data/data-observability-and-quality.html",
     "score": 0.0864
    },
    {
     "title": "数据工程基础",
     "url": "data/data-engineering-fundamentals.html",
     "score": 0.073
    },
    {
     "title": "数据架构与建模",
     "url": "data/data-architecture-and-modeling.html",
     "score": 0.0676
    }
   ]
  },
  {
   "i": 106,
   "title": "流处理与实时分析",
   "t": "流处理与实时分析",
   "u": "data/streaming-and-real-time-analytics.html",
   "s": "流处理与实时分析面向「数据到达即处理」的场景，用持续计算代替按周期批处理，把延迟从分钟/小时级压缩到秒级。技术栈通常由三部分构成：消息与事件日志（Kafka 等）、流计算引擎（Flink、Spark Structured Streaming）、以及实时消费端（实时数仓、物化视图、流式数据库）。CDC（Change Data Capture，变更数据捕获）是实…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "risingwave",
    "kafka",
    "cdc",
    "confluent",
    "flink",
    "kraft",
    "sink",
    "iceberg"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "数据工程基础",
     "url": "data/data-engineering-fundamentals.html",
     "score": 0.2336
    },
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.1675
    },
    {
     "title": "特征平台与 ML 数据",
     "url": "data/feature-stores-and-ml-data.html",
     "score": 0.1104
    },
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.0972
    }
   ]
  },
  {
   "i": 107,
   "title": "湖仓一体与表格式",
   "t": "湖仓一体与表格式",
   "u": "data/lakehouse-and-table-formats.html",
   "s": "湖仓一体（Lakehouse）试图把数据湖的可扩展性、开放性与数据仓库的事务性和性能结合起来，其技术地基是开放表格式（Open Table Format）。开放表格式在 Parquet/ORC 等文件格式之上，用元数据层提供 ACID 事务、schema 演进、时间旅行与隐藏分区等数据库式能力。当前三大主流为 Apache Iceberg、Delta Lak…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "hudi",
    "iceberg",
    "delta",
    "lake",
    "polaris",
    "catalog",
    "upsert",
    "uniform"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 15,
   "related": [
    {
     "title": "流处理与实时分析",
     "url": "data/streaming-and-real-time-analytics.html",
     "score": 0.1675
    },
    {
     "title": "数据架构与建模",
     "url": "data/data-architecture-and-modeling.html",
     "score": 0.1548
    },
    {
     "title": "特征平台与 ML 数据",
     "url": "data/feature-stores-and-ml-data.html",
     "score": 0.1378
    },
    {
     "title": "OLAP 与查询引擎",
     "url": "data/olap-and-query-engines.html",
     "score": 0.1139
    }
   ]
  },
  {
   "i": 108,
   "title": "特征平台与 ML 数据",
   "t": "特征平台与 ML 数据",
   "u": "data/feature-stores-and-ml-data.html",
   "s": "特征平台（Feature Platform）与特征存储（Feature Store）是机器学习工程中连接数据工程与模型训练/推理的中间层。其核心目标是：让特征可定义、可发现、可复用、可治理，并保证离线训练与在线推理使用同一套特征计算逻辑，从而避免训练-服务偏差（training-serving skew）。与之配套的还有 ML 数据工程实践，包括训练数据版本…",
   "dm": "数据工程与数据平台",
   "d": "data",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "feature",
    "lakefs",
    "databricks",
    "store",
    "mlops",
    "git",
    "spark",
    "unity"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "湖仓一体与表格式",
     "url": "data/lakehouse-and-table-formats.html",
     "score": 0.1378
    },
    {
     "title": "数据工程基础",
     "url": "data/data-engineering-fundamentals.html",
     "score": 0.1184
    },
    {
     "title": "流处理与实时分析",
     "url": "data/streaming-and-real-time-analytics.html",
     "score": 0.1104
    },
    {
     "title": "数据可观测性与质量",
     "url": "data/data-observability-and-quality.html",
     "score": 0.089
    }
   ]
  },
  {
   "i": 109,
   "title": "AI 与 LLM 安全",
   "t": "AI 与 LLM 安全",
   "u": "security/ai-security-and-llm-attacks.html",
   "s": "LLM 应用的安全问题源于一个结构性事实：模型无法在上下文窗口内对\"数据\"与\"指令\"做硬性隔离（OWASP Top 10 for LLM Applications 2025 解读）。因此，攻击者只要能把自然语言指令送入模型可见的通道——用户输入、检索到的文档、工具返回、网页、附件——就可能覆盖开发者写入的系统提示。OWASP 将这类风险列为 LLM01: P…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 15,
   "tags": [
    "owasp",
    "injection",
    "poisoning",
    "adi",
    "zscaler",
    "csa",
    "prompt",
    "attack"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 34,
   "related": [
    {
     "title": "应用安全",
     "url": "security/application-security.html",
     "score": 0.1304
    },
    {
     "title": "终端与勒索软件防御",
     "url": "security/endpoint-and-ransomware-defense.html",
     "score": 0.1135
    },
    {
     "title": "软件供应链安全",
     "url": "security/supply-chain-security.html",
     "score": 0.1077
    },
    {
     "title": "身份与访问管理（IAM）",
     "url": "security/identity-and-access-management.html",
     "score": 0.1014
    }
   ]
  },
  {
   "i": 110,
   "title": "OT 与 IoT 安全",
   "t": "OT 与 IoT 安全",
   "u": "security/ot-and-iot-security.html",
   "s": "OT（Operational Technology）安全关注工业自动化与控制系统（IACS/ICS）、SCADA、PLC、HMI 等生产环境的可用性、完整性与安全性；IoT 安全关注海量联网终端（路由器、摄像头、DVR、消费设备）的固件、凭据与生命周期管理；车联网安全则围绕车辆电子电气架构、软件更新与供应链展开。OT 安全的核心标准是 IEC 62443 系…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 14,
   "tags": [
    "ics",
    "iec",
    "etsi",
    "cra",
    "cert",
    "water",
    "legal-content",
    "industrial"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.1593
    },
    {
     "title": "威胁态势 2026",
     "url": "security/threat-landscape-2026.html",
     "score": 0.1392
    },
    {
     "title": "软件供应链安全",
     "url": "security/supply-chain-security.html",
     "score": 0.1258
    },
    {
     "title": "终端与勒索软件防御",
     "url": "security/endpoint-and-ransomware-defense.html",
     "score": 0.1154
    }
   ]
  },
  {
   "i": 111,
   "title": "云安全",
   "t": "云安全",
   "u": "security/cloud-security.html",
   "s": "云安全（Cloud Security）围绕云上身份、配置、工作负载、数据与基础设施即代码（IaC）的风险展开。2025–2026 年的主线是「平台整合」与「身份风险前置」：Gartner 提出的 CNAPP（Cloud-Native Application Protection Platform）成为事实上的采购框架，而 IAM 过度授权、暴露的服务主体凭证…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "cnapp",
    "ciem",
    "kubernetes",
    "iam",
    "iac",
    "container",
    "gartner",
    "cve"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性厂商 / 平台",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "身份与访问管理（IAM）",
     "url": "security/identity-and-access-management.html",
     "score": 0.1414
    },
    {
     "title": "终端与勒索软件防御",
     "url": "security/endpoint-and-ransomware-defense.html",
     "score": 0.1335
    },
    {
     "title": "网络安全与零信任",
     "url": "security/network-security-and-zero-trust.html",
     "score": 0.1197
    },
    {
     "title": "应用安全",
     "url": "security/application-security.html",
     "score": 0.1108
    }
   ]
  },
  {
   "i": 112,
   "title": "合规与安全治理",
   "t": "合规与安全治理",
   "u": "security/compliance-and-security-governance.html",
   "s": "安全合规与治理是把外部要求（法规、标准、行业规范）转化为可审计、可持续运行的内部控制的工程。企业常见的合规基线包括：ISO/IEC 27001（信息安全管理体系）、SOC 2（服务组织的控制鉴证）、NIST Cybersecurity Framework（风险管理框架）、PCI DSS（支付卡数据）、以及区域性法规如 GDPR、NIS2、DORA、欧盟《网络…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 15,
   "tags": [
    "cmmc",
    "iso",
    "legal-content",
    "edpb",
    "cra",
    "gdpr",
    "eur-lex",
    "ict"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 37,
   "related": [
    {
     "title": "OT 与 IoT 安全",
     "url": "security/ot-and-iot-security.html",
     "score": 0.1593
    },
    {
     "title": "软件供应链安全",
     "url": "security/supply-chain-security.html",
     "score": 0.1083
    },
    {
     "title": "隐私增强技术（PETs）",
     "url": "security/privacy-enhancing-technologies.html",
     "score": 0.0868
    },
    {
     "title": "网络安全与零信任",
     "url": "security/network-security-and-zero-trust.html",
     "score": 0.0717
    }
   ]
  },
  {
   "i": 113,
   "title": "威胁态势 2026",
   "t": "威胁态势 2026",
   "u": "security/threat-landscape-2026.html",
   "s": "2025–2026 年的全球网络安全威胁态势呈现三个相互叠加的特征：勒索软件在「攻击量激增」与「赎金支付萎缩」之间背离，国家级 APT 活动向供应链与云基础设施纵深渗透，以及生成式 AI 被攻防双方同时武器化。权威年度报告给出的量化口径虽有差异，但方向高度一致——攻击数量与效率在上升，而防御方的成本结构正在被 AI 改写。 最新进展（2025–2026） 勒…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "threat",
    "apt",
    "chainalysis",
    "dbir",
    "ransomware",
    "breach",
    "verizon",
    "crime"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性机构 / 报告 / 平台",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "终端与勒索软件防御",
     "url": "security/endpoint-and-ransomware-defense.html",
     "score": 0.1939
    },
    {
     "title": "OT 与 IoT 安全",
     "url": "security/ot-and-iot-security.html",
     "score": 0.1392
    },
    {
     "title": "攻防安全与漏洞赏金",
     "url": "security/offensive-security-and-bug-bounty.html",
     "score": 0.1378
    },
    {
     "title": "安全运营与 SOC",
     "url": "security/security-operations-and-soc.html",
     "score": 0.1314
    }
   ]
  },
  {
   "i": 114,
   "title": "安全运营与 SOC",
   "t": "安全运营与 SOC",
   "u": "security/security-operations-and-soc.html",
   "s": "安全运营中心（Security Operations Center, SOC）承担持续监控、检测、分诊、响应与复盘的职责，其技术栈通常由 SIEM（日志汇聚与关联分析）、SOAR（编排、自动化与响应）、EDR/XDR、威胁情报与检测工程流程组成。SOAR 通过集成异构安全工具、以预定义 playbook 自动化重复任务来加速响应，减少人工干预（What Is…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 12,
   "tags": [
    "soar",
    "siem",
    "mitre",
    "att",
    "sentinel",
    "mttr",
    "verizon",
    "mttd"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 29,
   "related": [
    {
     "title": "终端与勒索软件防御",
     "url": "security/endpoint-and-ransomware-defense.html",
     "score": 0.2077
    },
    {
     "title": "攻防安全与漏洞赏金",
     "url": "security/offensive-security-and-bug-bounty.html",
     "score": 0.2055
    },
    {
     "title": "威胁态势 2026",
     "url": "security/threat-landscape-2026.html",
     "score": 0.1314
    },
    {
     "title": "软件供应链安全",
     "url": "security/supply-chain-security.html",
     "score": 0.1091
    }
   ]
  },
  {
   "i": 115,
   "title": "密码学与后量子迁移",
   "t": "密码学与后量子迁移",
   "u": "security/cryptography-and-pqc.html",
   "s": "密码学为保密性、完整性与身份认证提供数学基础。2025–2026 年的核心叙事是「后量子迁移（PQC Migration）」从标准制定转向工程落地：NIST 已发布首批 PQC 标准并选定第五个算法，TLS 1.3 的混合密钥交换开始规模化部署，而中国商用密码体系（国密 SM 系列）也在推进标准更新与抗量子改造。 最新进展（2025–2026） NIST P…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "pqc",
    "post-quantum",
    "tls",
    "ml-kem",
    "hqc",
    "csrc",
    "fips",
    "ml-dsa"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性组织 / 标准 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 12,
   "related": [
    {
     "title": "身份与访问管理（IAM）",
     "url": "security/identity-and-access-management.html",
     "score": 0.1044
    },
    {
     "title": "隐私增强技术（PETs）",
     "url": "security/privacy-enhancing-technologies.html",
     "score": 0.0916
    },
    {
     "title": "网络安全与零信任",
     "url": "security/network-security-and-zero-trust.html",
     "score": 0.0718
    },
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.0701
    }
   ]
  },
  {
   "i": 116,
   "title": "应用安全",
   "t": "应用安全",
   "u": "security/application-security.html",
   "s": "应用安全（Application Security，AppSec）关注在软件设计、开发、测试与运行各阶段识别与消除应用层弱点。2025–2026 年，该领域出现两个显著变化：一是 OWASP Top 10 发布第 8 版（2025 版），首次将「软件供应链失败」单列为独立类别，并新增「异常条件处理不当」类别；二是 AI 生成代码大规模进入生产环境，使传统 S…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "owasp",
    "api-security",
    "broken",
    "appsec",
    "api1",
    "application",
    "failures",
    "authorization"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 产品 / 组织",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "AI 与 LLM 安全",
     "url": "security/ai-security-and-llm-attacks.html",
     "score": 0.1304
    },
    {
     "title": "云安全",
     "url": "security/cloud-security.html",
     "score": 0.1108
    },
    {
     "title": "身份与访问管理（IAM）",
     "url": "security/identity-and-access-management.html",
     "score": 0.0865
    },
    {
     "title": "软件供应链安全",
     "url": "security/supply-chain-security.html",
     "score": 0.0788
    }
   ]
  },
  {
   "i": 117,
   "title": "攻防安全与漏洞赏金",
   "t": "攻防安全与漏洞赏金",
   "u": "security/offensive-security-and-bug-bounty.html",
   "s": "攻防安全（offensive security）涵盖渗透测试、红队演练、漏洞挖掘与协调披露（CVD）、漏洞赏金计划以及漏洞交易市场。其经济逻辑在 2025–2026 年发生明显位移：Verizon《2026 DBIR》显示，漏洞利用首次成为最常见的初始访问向量，占比从上一年的 20% 升至 31%，而凭据滥用降至 13%（2026 Data Breach I…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "kev",
    "cve",
    "programs",
    "vulnerability",
    "csa",
    "cvss",
    "verizon",
    "t158"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "安全运营与 SOC",
     "url": "security/security-operations-and-soc.html",
     "score": 0.2055
    },
    {
     "title": "威胁态势 2026",
     "url": "security/threat-landscape-2026.html",
     "score": 0.1378
    },
    {
     "title": "软件供应链安全",
     "url": "security/supply-chain-security.html",
     "score": 0.1144
    },
    {
     "title": "AI 与 LLM 安全",
     "url": "security/ai-security-and-llm-attacks.html",
     "score": 0.0981
    }
   ]
  },
  {
   "i": 118,
   "title": "终端与勒索软件防御",
   "t": "终端与勒索软件防御",
   "u": "security/endpoint-and-ransomware-defense.html",
   "s": "终端与勒索软件防御（Endpoint and Ransomware Defense）以端点检测与响应（EDR/XDR）、勒索攻击链阻断、备份与恢复策略、威胁情报与狩猎为核心。2025–2026 年的核心判断是：勒索攻击已普遍采用「先窃取、后加密」的多重勒索流程，备份恢复不再是唯一「免死金牌」；同时 EDR/XDR 向 AI 驱动的统一检测、调查与响应演进，M…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "xdr",
    "mitre",
    "defender",
    "edr",
    "att",
    "ransomware",
    "sophos",
    "response"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性产品 / 组织 / 框架",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "安全运营与 SOC",
     "url": "security/security-operations-and-soc.html",
     "score": 0.2077
    },
    {
     "title": "威胁态势 2026",
     "url": "security/threat-landscape-2026.html",
     "score": 0.1939
    },
    {
     "title": "云安全",
     "url": "security/cloud-security.html",
     "score": 0.1335
    },
    {
     "title": "OT 与 IoT 安全",
     "url": "security/ot-and-iot-security.html",
     "score": 0.1154
    }
   ]
  },
  {
   "i": 119,
   "title": "网络安全与零信任",
   "t": "网络安全与零信任",
   "u": "security/network-security-and-zero-trust.html",
   "s": "网络安全与零信任（Network Security and Zero Trust）关注如何在网络层控制东西向与南北向流量、抵御大流量攻击，并以身份与上下文替代静态边界。2025–2026 年的关键变化是：零信任从概念框架走向可落地的参考实现，SASE/SSE 加速走向单供应商收敛，微隔离成为限制横向移动的主流控制，而 DDoS 攻击规模持续刷新纪录并呈现「超…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "sse",
    "zero",
    "ddos",
    "tbps",
    "cloudflare",
    "trust",
    "zscaler",
    "gartner"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性组织 / 厂商 / 平台",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "身份与访问管理（IAM）",
     "url": "security/identity-and-access-management.html",
     "score": 0.132
    },
    {
     "title": "云安全",
     "url": "security/cloud-security.html",
     "score": 0.1197
    },
    {
     "title": "AI 与 LLM 安全",
     "url": "security/ai-security-and-llm-attacks.html",
     "score": 0.0958
    },
    {
     "title": "终端与勒索软件防御",
     "url": "security/endpoint-and-ransomware-defense.html",
     "score": 0.0906
    }
   ]
  },
  {
   "i": 120,
   "title": "身份与访问管理（IAM）",
   "t": "身份与访问管理（IAM）",
   "u": "security/identity-and-access-management.html",
   "s": "身份与访问管理（Identity and Access Management，IAM）负责对「谁在什么条件下可以访问什么资源」进行认证、授权与治理。2025–2026 年，IAM 的主线有三条：OAuth 2.1 与 WebAuthn/Passkey 等现代认证规范持续落地；密码正在被 passkey 大规模替代；以及非人类身份（Non-Human Iden…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "oauth",
    "identity",
    "nhi",
    "rfc",
    "alliance",
    "metadata",
    "oidc",
    "csa"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性组织与标准",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "云安全",
     "url": "security/cloud-security.html",
     "score": 0.1414
    },
    {
     "title": "网络安全与零信任",
     "url": "security/network-security-and-zero-trust.html",
     "score": 0.132
    },
    {
     "title": "密码学与后量子迁移",
     "url": "security/cryptography-and-pqc.html",
     "score": 0.1044
    },
    {
     "title": "AI 与 LLM 安全",
     "url": "security/ai-security-and-llm-attacks.html",
     "score": 0.1014
    }
   ]
  },
  {
   "i": 121,
   "title": "软件供应链安全",
   "t": "软件供应链安全",
   "u": "security/supply-chain-security.html",
   "s": "软件供应链安全关注代码从源码、依赖、构建、打包到分发全链路被篡改的风险。SLSA（Supply-chain Levels for Software Artifacts，读作 \"salsa\"）是由产业共识建立、可增量采纳的供应链安全指南：生产方据此加固自身链路，消费方据此判断软件包是否可信（About SLSA）。SLSA 官方将典型威胁案例指向 SolarW…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "slsa",
    "sigstore",
    "sbom",
    "cra",
    "shai-hulud",
    "attestation",
    "cosign",
    "openssf"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "OT 与 IoT 安全",
     "url": "security/ot-and-iot-security.html",
     "score": 0.1258
    },
    {
     "title": "攻防安全与漏洞赏金",
     "url": "security/offensive-security-and-bug-bounty.html",
     "score": 0.1144
    },
    {
     "title": "安全运营与 SOC",
     "url": "security/security-operations-and-soc.html",
     "score": 0.1091
    },
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.1083
    }
   ]
  },
  {
   "i": 122,
   "title": "隐私增强技术（PETs）",
   "t": "隐私增强技术（PETs）",
   "u": "security/privacy-enhancing-technologies.html",
   "s": "隐私增强技术（Privacy-Enhancing Technologies, PETs）指在保留数据可用价值的前提下，降低个人数据暴露面的一类技术集合，通常包括差分隐私（Differential Privacy, DP）、联邦学习（Federated Learning, FL）、同态加密（Homomorphic Encryption, HE）、安全多方计算（…",
   "dm": "网络安全",
   "d": "security",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "edpb",
    "privacy",
    "psi",
    "encryption",
    "clean",
    "federated",
    "private",
    "journal"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 25,
   "related": [
    {
     "title": "密码学与后量子迁移",
     "url": "security/cryptography-and-pqc.html",
     "score": 0.0916
    },
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.0868
    },
    {
     "title": "AI 与 LLM 安全",
     "url": "security/ai-security-and-llm-attacks.html",
     "score": 0.0663
    },
    {
     "title": "云安全",
     "url": "security/cloud-security.html",
     "score": 0.0554
    }
   ]
  },
  {
   "i": 123,
   "title": "信息论与编码",
   "t": "信息论与编码",
   "u": "science/information-theory-and-coding.html",
   "s": "信息论研究信息度量、压缩与传输的理论极限；编码理论则构造逼近这些极限的实用方案。Claude Shannon 1948 年的奠基性工作同时提出了信道容量定理与信源编码的保真度准则；后续发展包括 Berger 的率失真变分表述、Wyner–Ziv 对带边信息信源的扩展、Zamir–Feder 的格量化框架，以及 Gallager 的 LDPC 码与 Arıka…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 6,
   "tags": [
    "ldpc",
    "dna",
    "quantum",
    "storage",
    "coding",
    "信息论与编码"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.2214
    },
    {
     "title": "计算科学与仿真",
     "url": "science/computational-science-and-simulation.html",
     "score": 0.0568
    },
    {
     "title": "复杂系统与网络科学",
     "url": "science/complex-systems-and-network-science.html",
     "score": 0.0533
    },
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.0386
    }
   ]
  },
  {
   "i": 124,
   "title": "化学与催化",
   "t": "化学与催化",
   "u": "science/chemistry-and-catalysis.html",
   "s": "化学与催化前沿在 2025–2026 年的核心特征，是「AI 与自动化」对传统合成化学与催化研发流程的系统性渗透：从反应产率预测、逆设计催化剂，到由 AI 智能体自主规划并执行实验的闭环工作流，方法学层面出现大量新工具；与此同时，电催化与光催化在能源转化方向上持续产出高性能催化剂，金属有机框架（MOF）的基础性贡献则获得 2025 年诺贝尔化学奖的确认。 最…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "chemistry",
    "column",
    "orr",
    "mof",
    "prizes",
    "rsc",
    "abs",
    "kitagawa"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 机构 / 产品",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 8,
   "related": [
    {
     "title": "材料科学",
     "url": "science/materials-science.html",
     "score": 0.2922
    },
    {
     "title": "生物学与基因组学",
     "url": "science/biology-and-genomics.html",
     "score": 0.1358
    },
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.1073
    },
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.099
    }
   ]
  },
  {
   "i": 125,
   "title": "复杂系统与网络科学",
   "t": "复杂系统与网络科学",
   "u": "science/complex-systems-and-network-science.html",
   "s": "复杂系统研究由大量异质单元通过非线性相互作用组成的系统如何涌现出宏观有序行为，网络科学则以图、超图与多层网络为语言刻画结构—功能关系。两者在临界性、同步、传播动力学与复杂经济学等方向高度交汇，2025–2026 年的一个共同趋势是把长期停留在结构层面的指标与真实的网络动力学打通。 最新进展（2025–2026） 结构指标与动力学的打通。 深圳大学吴宗泽、王向…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "networks",
    "graph",
    "复杂系统与网络科学"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "计算科学与仿真",
     "url": "science/computational-science-and-simulation.html",
     "score": 0.0971
    },
    {
     "title": "气候科学与地球系统",
     "url": "science/climate-science-and-earth-system.html",
     "score": 0.0929
    },
    {
     "title": "统计学与因果推断",
     "url": "science/statistics-and-causal-inference.html",
     "score": 0.0878
    },
    {
     "title": "天文与宇宙学",
     "url": "science/astronomy-and-cosmology.html",
     "score": 0.0796
    }
   ]
  },
  {
   "i": 126,
   "title": "天文与宇宙学",
   "t": "天文与宇宙学",
   "u": "science/astronomy-and-cosmology.html",
   "s": "2025–2026 年的天文学与宇宙学由三股力量推动：詹姆斯·韦布空间望远镜（JWST）持续产出系外行星大气与早期宇宙观测；DESI 的星系巡天为「动态暗能量」提供了强烈但尚未定论的证据，直接挑战 ΛCDM 标准宇宙学模型；引力波天文学则进入大样本统计时代，探测事件总数快速累积。这三条线索共同构成了当前宇宙学「标准模型受挑战、观测精度快速提升」的整体图景。 …",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 6,
   "tags": [
    "dynamical",
    "energy",
    "catalog",
    "天文与宇宙学"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.1007
    },
    {
     "title": "复杂系统与网络科学",
     "url": "science/complex-systems-and-network-science.html",
     "score": 0.0796
    },
    {
     "title": "气候科学与地球系统",
     "url": "science/climate-science-and-earth-system.html",
     "score": 0.0709
    },
    {
     "title": "统计学与因果推断",
     "url": "science/statistics-and-causal-inference.html",
     "score": 0.0511
    }
   ]
  },
  {
   "i": 127,
   "title": "开放科学与科研基础设施",
   "t": "开放科学与科研基础设施",
   "u": "science/open-science-and-research-infrastructure.html",
   "s": "开放科学主张研究成果（论文、数据、代码、协议）可发现、可访问、可互操作、可复用，并通过预印本、开放获取、数据仓储与科研评价改革来落地。2025–2026 年的关键变化是：开放从\"倡导\"走向\"强制要求\"，同时科研诚信与可复现性问题推动基础设施与评价制度的重构。 最新进展（2025–2026） 开放获取强制化。 美国 NIH 修订后的公共获取政策于 2025 年…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "nih",
    "fair",
    "medrxiv",
    "access",
    "dora",
    "open",
    "research",
    "new"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "气候科学与地球系统",
     "url": "science/climate-science-and-earth-system.html",
     "score": 0.0655
    },
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.0594
    },
    {
     "title": "神经科学",
     "url": "science/neuroscience.html",
     "score": 0.0517
    },
    {
     "title": "统计学与因果推断",
     "url": "science/statistics-and-causal-inference.html",
     "score": 0.0466
    }
   ]
  },
  {
   "i": 128,
   "title": "数学前沿",
   "t": "数学前沿",
   "u": "science/mathematics-frontier.html",
   "s": "数学前沿通常由三类事件标记：顶级奖项的颁发（菲尔兹奖等）、长期悬而未决猜想被证明或推翻，以及人工智能开始实质性介入数学研究与证明流程。2025–2026 年，这三条线索同时出现了值得记录的进展：三维 Kakeya 猜想被证明、ICM 2026 颁发新一届菲尔兹奖，以及多个 AI 系统在国际数学奥林匹克（IMO）与研究工作层面取得突破。 最新进展（2025–2…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 6,
   "tags": [
    "lean",
    "mathematics",
    "alphaproof",
    "fields",
    "imo",
    "hong",
    "nexus",
    "seed"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 机构 / 产品",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 11,
   "related": [
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.0642
    },
    {
     "title": "生物学与基因组学",
     "url": "science/biology-and-genomics.html",
     "score": 0.061
    },
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.058
    },
    {
     "title": "材料科学",
     "url": "science/materials-science.html",
     "score": 0.0569
    }
   ]
  },
  {
   "i": 129,
   "title": "材料科学",
   "t": "材料科学",
   "u": "science/materials-science.html",
   "s": "材料科学前沿在 2025–2026 年呈现出「AI 驱动发现 + 自动化实验」与「经典体系持续突破」双线并进的特征：以 Google DeepMind 的 GNoME 为代表的深度学习预测大幅扩张了已知稳定材料的空间，自主实验室（self-driving lab）把发现-验证周期压缩到天级；与此同时，二维材料、金属有机框架（MOF）、超导与量子材料等经典方向…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "gnome",
    "mof",
    "materials",
    "ref",
    "discover",
    "deepmind",
    "mit",
    "google"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 机构 / 产品",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 11,
   "related": [
    {
     "title": "化学与催化",
     "url": "science/chemistry-and-catalysis.html",
     "score": 0.2922
    },
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.169
    },
    {
     "title": "生物学与基因组学",
     "url": "science/biology-and-genomics.html",
     "score": 0.1221
    },
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.0959
    }
   ]
  },
  {
   "i": 130,
   "title": "气候科学与地球系统",
   "t": "气候科学与地球系统",
   "u": "science/climate-science-and-earth-system.html",
   "s": "气候科学以地球系统为对象，涵盖大气、海洋、冰冻圈、陆地生物圈及其与人类活动的相互作用，通过观测、再分析、数值模式与统计归因来理解气候变化的成因与后果。当前该领域的三条主线是：耦合模式比较计划 CMIP7 的推进、以 AI 气象预报为代表的数据驱动方法崛起，以及临界点（tipping points）与极端事件归因研究的深化。 最新进展（2025–2026） 观…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "weather",
    "climate",
    "aurora",
    "earth",
    "change",
    "气候科学与地球系统"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心科学与关键概念",
    "关键进展：临界点与归因",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "复杂系统与网络科学",
     "url": "science/complex-systems-and-network-science.html",
     "score": 0.0929
    },
    {
     "title": "计算科学与仿真",
     "url": "science/computational-science-and-simulation.html",
     "score": 0.0793
    },
    {
     "title": "天文与宇宙学",
     "url": "science/astronomy-and-cosmology.html",
     "score": 0.0709
    },
    {
     "title": "开放科学与科研基础设施",
     "url": "science/open-science-and-research-infrastructure.html",
     "score": 0.0655
    }
   ]
  },
  {
   "i": 131,
   "title": "物理学前沿",
   "t": "物理学前沿",
   "u": "science/physics-frontier.html",
   "s": "物理学前沿在 2025–2026 年围绕几条主线推进：大型对撞机（LHC Run 3）在 13.6 TeV 能量下持续检验标准模型；高精度测量（如 μ 子反常磁矩）达到空前精度并迫使理论计算重新审视；凝聚态领域在常压超导温度上刷新纪录；以及量子效应在宏观电路中的实验确认获得 2025 年诺贝尔物理学奖。此外，人工智能开始直接参与前沿理论物理的计算工作。 最新…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "muon",
    "pressure",
    "atlas",
    "updates",
    "record",
    "search",
    "claude",
    "new"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "材料科学",
     "url": "science/materials-science.html",
     "score": 0.169
    },
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.1475
    },
    {
     "title": "化学与催化",
     "url": "science/chemistry-and-catalysis.html",
     "score": 0.1073
    },
    {
     "title": "天文与宇宙学",
     "url": "science/astronomy-and-cosmology.html",
     "score": 0.1007
    }
   ]
  },
  {
   "i": 132,
   "title": "生物学与基因组学",
   "t": "生物学与基因组学",
   "u": "science/biology-and-genomics.html",
   "s": "2025–2026 年，生物学与基因组学的推进主要沿三个方向：单细胞与空间组学把「图谱化」从组织推进到全身体、全胚胎尺度；长读长与组装技术把人类基因组做到近乎无缺口，并首次实现单细胞系来源的完整端粒到端粒（T2T）二倍体基因组；AI 结构预测（AlphaFold 3）与实验数据（NMR/cryo-EM）的耦合，使蛋白质结构与互作预测更贴近真实构象。这些进展共…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "alphafold",
    "news-release",
    "alphafold3",
    "broad",
    "dna",
    "publications",
    "deepmind",
    "faq"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 机构 / 产品",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 11,
   "related": [
    {
     "title": "化学与催化",
     "url": "science/chemistry-and-catalysis.html",
     "score": 0.1358
    },
    {
     "title": "材料科学",
     "url": "science/materials-science.html",
     "score": 0.1221
    },
    {
     "title": "神经科学",
     "url": "science/neuroscience.html",
     "score": 0.1086
    },
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.0889
    }
   ]
  },
  {
   "i": 133,
   "title": "神经科学",
   "t": "神经科学",
   "u": "science/neuroscience.html",
   "s": "神经科学在 2025–2026 年的主线是「连接组（connectome）与脑图谱的规模化」：从完整果蝇脑到完整的雄性果蝇中枢神经系统，连接组研究不断突破规模与跨脑区连通的限制；与此同时，RNA「条形码」等新技术把神经元连接追踪变成测序问题，大幅提升通量与单突触分辨率。在脑机接口（BCI）方向，双向、可恢复感觉反馈的临床系统与无线高密度植入器件成为焦点，中国…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "university",
    "bci",
    "brain",
    "mcb",
    "california",
    "news-events",
    "mit",
    "group"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 机构 / 产品",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 11,
   "related": [
    {
     "title": "生物学与基因组学",
     "url": "science/biology-and-genomics.html",
     "score": 0.1086
    },
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.0654
    },
    {
     "title": "数学前沿",
     "url": "science/mathematics-frontier.html",
     "score": 0.0518
    },
    {
     "title": "开放科学与科研基础设施",
     "url": "science/open-science-and-research-infrastructure.html",
     "score": 0.0517
    }
   ]
  },
  {
   "i": 134,
   "title": "统计学与因果推断",
   "t": "统计学与因果推断",
   "u": "science/statistics-and-causal-inference.html",
   "s": "统计学提供从有限样本推断总体、量化不确定性与设计实验的工具；因果推断则聚焦\"干预的效果\"而非单纯关联。Rubin 的潜在结果框架与 Pearl 的结构因果模型（含因果图与 do-演算）是两大主流语言。实践中广泛使用随机对照试验、目标试验模拟（target trial emulation）以及因果机器学习方法。2025–2026 年的显著变化是 LLM 与自动…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "causal",
    "trial",
    "bias",
    "target",
    "jax",
    "discovery",
    "inference",
    "llm"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心概念与常见陷阱",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "复杂系统与网络科学",
     "url": "science/complex-systems-and-network-science.html",
     "score": 0.0878
    },
    {
     "title": "计算科学与仿真",
     "url": "science/computational-science-and-simulation.html",
     "score": 0.065
    },
    {
     "title": "天文与宇宙学",
     "url": "science/astronomy-and-cosmology.html",
     "score": 0.0511
    },
    {
     "title": "材料科学",
     "url": "science/materials-science.html",
     "score": 0.0492
    }
   ]
  },
  {
   "i": 135,
   "title": "计算科学与仿真",
   "t": "计算科学与仿真",
   "u": "science/computational-science-and-simulation.html",
   "s": "计算科学以数值方法、算法与高性能计算（HPC）求解无法解析处理的数学模型，覆盖分子动力学、计算流体力学（CFD）、电子结构计算、数字孪生等方向。2025–2026 年最显著的趋势是\"AI 与物理模型融合\"：机器学习势函数、神经算子与代理模型被引入传统求解流程，同时 E 级（exascale）超算为大规模仿真提供了算力底座。 最新进展（2025–2026） E…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "dft",
    "neural",
    "framework",
    "gpu",
    "计算科学与仿真"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "复杂系统与网络科学",
     "url": "science/complex-systems-and-network-science.html",
     "score": 0.0971
    },
    {
     "title": "量子信息科学",
     "url": "science/quantum-information-science.html",
     "score": 0.0846
    },
    {
     "title": "气候科学与地球系统",
     "url": "science/climate-science-and-earth-system.html",
     "score": 0.0793
    },
    {
     "title": "化学与催化",
     "url": "science/chemistry-and-catalysis.html",
     "score": 0.0676
    }
   ]
  },
  {
   "i": 136,
   "title": "量子信息科学",
   "t": "量子信息科学",
   "u": "science/quantum-information-science.html",
   "s": "量子信息科学研究如何利用叠加、纠缠与测量等量子特性处理信息，涵盖量子计算、量子通信/量子网络与量子传感/计量三大方向。2025–2026 年，该领域在纠错、网络实用化与精密测量上均取得标志性进展，并因 2025 年诺贝尔物理学奖而受到更广泛的公众关注。 最新进展（2025–2026） 基础与荣誉。 2025 年诺贝尔物理学奖授予 John Clarke、Mi…",
   "dm": "基础科学前沿",
   "d": "science",
   "ud": "2026-09-26",
   "rm": 16,
   "tags": [
    "quantum",
    "pqc",
    "optical",
    "helios",
    "error",
    "ibm",
    "nist",
    "roadmap"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 41,
   "related": [
    {
     "title": "信息论与编码",
     "url": "science/information-theory-and-coding.html",
     "score": 0.2214
    },
    {
     "title": "物理学前沿",
     "url": "science/physics-frontier.html",
     "score": 0.1475
    },
    {
     "title": "化学与催化",
     "url": "science/chemistry-and-catalysis.html",
     "score": 0.099
    },
    {
     "title": "材料科学",
     "url": "science/materials-science.html",
     "score": 0.0959
    }
   ]
  },
  {
   "i": 137,
   "title": "6G 与下一代连接",
   "t": "6G 与下一代连接",
   "u": "emerging/6g-and-next-gen-connectivity.html",
   "s": "6G（第六代移动通信）是继 5G 之后的下一代无线通信网络，核心目标是从「万物互联」向「万物智联」跃迁，将通信、感知、计算与 AI 深度融合，构建空天地一体化网络。ITU 于 2023 年 12 月发布 IMT-2030 框架建议书（Rec ITU-R M.2160），确立 6G 愿景与使用场景（ITU：IMT-2030 for 6G mobile tech…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "gpp",
    "itu",
    "ghz",
    "itu-r",
    "pages",
    "publishers",
    "mhz",
    "d2d"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 20,
   "related": [
    {
     "title": "卫星星座与卫星通信",
     "url": "emerging/satellite-constellations-and-satcom.html",
     "score": 0.1512
    },
    {
     "title": "脑机接口（Brain-Computer Interfaces, BCI）",
     "url": "emerging/brain-computer-interfaces.html",
     "score": 0.1021
    },
    {
     "title": "火箭与航天发射",
     "url": "emerging/rockets-and-space-launch.html",
     "score": 0.0938
    },
    {
     "title": "深空探索与太空经济",
     "url": "emerging/space-exploration-and-commerce.html",
     "score": 0.0879
    }
   ]
  },
  {
   "i": 138,
   "title": "AI 制药（AI Drug Discovery）",
   "t": "AI 制药（AI Drug Discovery）",
   "u": "emerging/ai-drug-discovery.html",
   "s": "AI 制药指把机器学习、生成式模型、蛋白质结构预测与自动化实验室（干湿闭环）等技术，用于靶点发现与验证、分子生成与优化、成药性与毒性预测、临床方案设计等环节，目标是缩短研发周期、降低失败率与成本。其技术形态通常分为三大路线：基于表型的反向工程（如 Recursion 从细胞成像表型反推机制）、基于结构与物理的理性设计（如 Isomorphic Labs、Sc…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "drug",
    "isomorphic",
    "insilico",
    "alphafold",
    "medicine",
    "discovery",
    "clinical",
    "trials"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 29,
   "related": [
    {
     "title": "长寿与抗衰老研究（Longevity and Aging Research）",
     "url": "emerging/longevity-and-aging-research.html",
     "score": 0.1986
    },
    {
     "title": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
     "url": "emerging/gene-editing-and-cell-therapy.html",
     "score": 0.1559
    },
    {
     "title": "合成生物学与生物制造（Synthetic Biology and Bio-Manufacturing）",
     "url": "emerging/synthetic-biology-and-bio-manufacturing.html",
     "score": 0.144
    },
    {
     "title": "固态电池与下一代电池（Solid-State & Next-Gen Batteries）",
     "url": "emerging/solid-state-and-next-gen-batteries.html",
     "score": 0.0756
    }
   ]
  },
  {
   "i": 139,
   "title": "人形机器人（Humanoid Robotics）",
   "t": "人形机器人（Humanoid Robotics）",
   "u": "emerging/humanoid-robotics.html",
   "s": "人形机器人（humanoid robot）指具备类人形态（头部、双臂、双足或轮式底盘）并能在人类环境中执行通用任务的机器人，通常被视为具身智能（embodied intelligence）的主要载体。2025–2026 年，该赛道从原型演示进入小批量量产与真实场景验证阶段。业界普遍把 2026 年称为人形机器人的“量产元年”，其标志是特斯拉 Optimus …",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 14,
   "tags": [
    "robotics",
    "isaac",
    "humanoid",
    "ipo",
    "figure",
    "tesla",
    "vla",
    "robots"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 39,
   "related": [
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.1648
    },
    {
     "title": "工业机器人与自动化（Industrial Robotics and Automation）",
     "url": "emerging/industrial-robotics-and-automation.html",
     "score": 0.1608
    },
    {
     "title": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
     "url": "emerging/autonomous-delivery-and-logistics-robots.html",
     "score": 0.113
    },
    {
     "title": "脑机接口（Brain-Computer Interfaces, BCI）",
     "url": "emerging/brain-computer-interfaces.html",
     "score": 0.1043
    }
   ]
  },
  {
   "i": 140,
   "title": "卫星星座与卫星通信",
   "t": "卫星星座与卫星通信",
   "u": "emerging/satellite-constellations-and-satcom.html",
   "s": "卫星星座与卫星通信指以大量低轨（LEO）卫星组网提供宽带互联网接入、手机直连（Direct-to-Cell / D2D）等服务的产业。相较传统地球静止轨道（GEO）卫星，LEO 巨型星座凭借低时延、高带宽与规模部署能力，已成为商业航天最具现金流的赛道之一。业界普遍认为，卫星互联网的收入模式正从「技术验证」转向「运营验证」，Starlink 所代表的卫星互联网…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "starlink",
    "leo",
    "satellites",
    "spacex",
    "ast",
    "d2c",
    "bluebird",
    "itu"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 39,
   "related": [
    {
     "title": "火箭与航天发射",
     "url": "emerging/rockets-and-space-launch.html",
     "score": 0.2655
    },
    {
     "title": "6G 与下一代连接",
     "url": "emerging/6g-and-next-gen-connectivity.html",
     "score": 0.1512
    },
    {
     "title": "深空探索与太空经济",
     "url": "emerging/space-exploration-and-commerce.html",
     "score": 0.1337
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0828
    }
   ]
  },
  {
   "i": 141,
   "title": "合成生物学与生物制造（Synthetic Biology and Bio-Manufacturing）",
   "t": "合成生物学与生物制造（Synthetic Biology and Bio-Manufacturing）",
   "u": "emerging/synthetic-biology-and-bio-manufacturing.html",
   "s": "合成生物学（synthetic biology）以工程化理念设计和改造生物系统，通过标准化生物元件、基因线路与底盘细胞，实现化学品、材料、能源、食品与医药的定向生产；生物制造（bio-manufacturing）则是其产业化形态，用生物基工艺替代传统石化或化学合成路径。2025–2026 年，该领域呈现「工具层与 AI 设计能力快速成熟、平台型公司商业模式承…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "biology",
    "synthetic",
    "esm3",
    "baker",
    "dna",
    "alphafold",
    "lab",
    "market"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "AI 制药（AI Drug Discovery）",
     "url": "emerging/ai-drug-discovery.html",
     "score": 0.144
    },
    {
     "title": "长寿与抗衰老研究（Longevity and Aging Research）",
     "url": "emerging/longevity-and-aging-research.html",
     "score": 0.1172
    },
    {
     "title": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
     "url": "emerging/gene-editing-and-cell-therapy.html",
     "score": 0.098
    },
    {
     "title": "数字健康与医疗器械（Digital Health and Medical Devices）",
     "url": "emerging/digital-health-and-medical-devices.html",
     "score": 0.0962
    }
   ]
  },
  {
   "i": 142,
   "title": "固态电池与下一代电池（Solid-State & Next-Gen Batteries）",
   "t": "固态电池与下一代电池（Solid-State & Next-Gen Batteries）",
   "u": "emerging/solid-state-and-next-gen-batteries.html",
   "s": "下一代电池主要指以固态电解质替代液态电解液的全固态电池（solid-state battery），以及钠离子电池、锂金属负极、硅负极等新化学体系。全固态电池以能量密度高、本征安全性好（不易燃）、可支持超快充为卖点，核心难点在于固态电解质的离子电导率、固-固界面阻抗、枝晶抑制与规模化制造；钠离子电池则以钠资源丰富、低温性能与成本优势，率先在储能与入门级动力场景…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "solid-state",
    "batteries",
    "battery",
    "finally",
    "production",
    "design",
    "固态电池与下一代电池"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "电动汽车与交通电动化",
     "url": "emerging/evs-and-transport-electrification.html",
     "score": 0.2037
    },
    {
     "title": "新能源与气候科技",
     "url": "emerging/new-energy-and-climate-tech.html",
     "score": 0.1888
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0872
    },
    {
     "title": "AI 制药（AI Drug Discovery）",
     "url": "emerging/ai-drug-discovery.html",
     "score": 0.0756
    }
   ]
  },
  {
   "i": 143,
   "title": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
   "t": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
   "u": "emerging/gene-editing-and-cell-therapy.html",
   "s": "基因编辑指通过核酸酶或碱基编辑工具对基因组进行定点改造，主要包括 CRISPR-Cas9 等核酸酶编辑、碱基编辑（base editing）与先导编辑（prime editing）。细胞治疗则以 CAR-T 等改造自体/异体细胞为核心。2023 年底，首款 CRISPR 基因编辑疗法获批，标志该领域从科研走向商业化；2025–2026 年，行业重心转向两个方…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "therapeutics",
    "vivo",
    "editing",
    "scd",
    "therapy",
    "cell",
    "vertex",
    "about-us"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 32,
   "related": [
    {
     "title": "长寿与抗衰老研究（Longevity and Aging Research）",
     "url": "emerging/longevity-and-aging-research.html",
     "score": 0.1811
    },
    {
     "title": "AI 制药（AI Drug Discovery）",
     "url": "emerging/ai-drug-discovery.html",
     "score": 0.1559
    },
    {
     "title": "数字健康与医疗器械（Digital Health and Medical Devices）",
     "url": "emerging/digital-health-and-medical-devices.html",
     "score": 0.1105
    },
    {
     "title": "合成生物学与生物制造（Synthetic Biology and Bio-Manufacturing）",
     "url": "emerging/synthetic-biology-and-bio-manufacturing.html",
     "score": 0.098
    }
   ]
  },
  {
   "i": 144,
   "title": "小型模块化反应堆与核能（Small Modular Reactors & Nuclear）",
   "t": "小型模块化反应堆与核能（Small Modular Reactors & Nuclear）",
   "u": "emerging/small-modular-reactors.html",
   "s": "小型模块化反应堆（Small Modular Reactor，SMR）通常指单机功率较小（业界常以 IAEA 的约 300 MWe 上限为口径）、采用工厂预制、模块化运输与现场组装的裂变反应堆，旨在以更低的单位造价与更短的现场建造周期提供低碳、可调度的基荷电力。与之并列的还有“先进反应堆”与第四代堆（如钠冷快堆、高温气冷堆、熔盐堆、微堆），以及把成熟大型机组…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "reactor",
    "nuclear",
    "smr",
    "doe",
    "nuscale",
    "vernova",
    "bwrx-300",
    "licensing"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 13,
   "related": [
    {
     "title": "新能源与气候科技",
     "url": "emerging/new-energy-and-climate-tech.html",
     "score": 0.1641
    },
    {
     "title": "聚变能源（Fusion Energy）",
     "url": "emerging/fusion-energy.html",
     "score": 0.1072
    },
    {
     "title": "氢能与绿色燃料（Hydrogen & Green Fuels）",
     "url": "emerging/hydrogen-and-green-fuels.html",
     "score": 0.0739
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0509
    }
   ]
  },
  {
   "i": 145,
   "title": "工业机器人与自动化（Industrial Robotics and Automation）",
   "t": "工业机器人与自动化（Industrial Robotics and Automation）",
   "u": "emerging/industrial-robotics-and-automation.html",
   "s": "工业机器人指面向制造业的自动化操作系统，主要包括多关节机器人、SCARA、直角坐标、并联（Delta）机器人与协作机器人（cobot）。自动化则进一步涵盖产线集成、机器视觉、运动控制与“黑灯工厂”（关灯无人化生产）等。国际机器人联合会（IFR）于 2026 年 9 月 24 日发布《2026 年世界机器人报告》，显示全球在用工业机器人总量首次达到 500 万…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 6,
   "tags": [
    "robots",
    "universal",
    "amr",
    "now",
    "new",
    "工业机器人与自动化"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 10,
   "related": [
    {
     "title": "人形机器人（Humanoid Robotics）",
     "url": "emerging/humanoid-robotics.html",
     "score": 0.1608
    },
    {
     "title": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
     "url": "emerging/autonomous-delivery-and-logistics-robots.html",
     "score": 0.1213
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0734
    },
    {
     "title": "电动汽车与交通电动化",
     "url": "emerging/evs-and-transport-electrification.html",
     "score": 0.0678
    }
   ]
  },
  {
   "i": 146,
   "title": "数字健康与医疗器械（Digital Health and Medical Devices）",
   "t": "数字健康与医疗器械（Digital Health and Medical Devices）",
   "u": "emerging/digital-health-and-medical-devices.html",
   "s": "数字健康（digital health）涵盖远程医疗、可穿戴健康监测、数字疗法、医疗软件与 AI 医疗器械等。其监管核心是「作为医疗器械的软件」（Software as a Medical Device, SaMD）：美国 FDA 依据预期用途对临床诊断与治疗类 AI 软件按医疗器械管理（治疗类监管最严），纯行政类工具通常不受器械监管但仍受隐私等规则约束（H…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "fda",
    "medical",
    "novo",
    "scripts",
    "health",
    "watch",
    "devices",
    "list"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 23,
   "related": [
    {
     "title": "脑机接口（Brain-Computer Interfaces, BCI）",
     "url": "emerging/brain-computer-interfaces.html",
     "score": 0.165
    },
    {
     "title": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
     "url": "emerging/gene-editing-and-cell-therapy.html",
     "score": 0.1105
    },
    {
     "title": "合成生物学与生物制造（Synthetic Biology and Bio-Manufacturing）",
     "url": "emerging/synthetic-biology-and-bio-manufacturing.html",
     "score": 0.0962
    },
    {
     "title": "长寿与抗衰老研究（Longevity and Aging Research）",
     "url": "emerging/longevity-and-aging-research.html",
     "score": 0.0905
    }
   ]
  },
  {
   "i": 147,
   "title": "新能源与气候科技",
   "t": "新能源与气候科技",
   "u": "emerging/new-energy-and-climate-tech.html",
   "s": "新能源与气候科技涵盖光伏、储能、电动汽车与电池供应链、电网与虚拟电厂、绿氢、核能（含小型模块化反应堆 SMR 与聚变）、碳捕集（DAC）以及相关政策框架。2025–2026 年呈现明显的\"冰火两重天\"：光伏与储能成本持续下探、装机快速扩张；聚变与 SMR 迎来资本与工程的密集里程碑；而绿氢则因需求与融资瓶颈出现大规模项目取消。同时，AI 数据中心带来的电力需…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "dac",
    "cfs",
    "climeworks",
    "iea",
    "nuscale",
    "nuclear",
    "fuel",
    "hydrogen"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/公司（含官方链接）",
    "关键数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 33,
   "related": [
    {
     "title": "聚变能源（Fusion Energy）",
     "url": "emerging/fusion-energy.html",
     "score": 0.2467
    },
    {
     "title": "碳捕集与气候适应（Carbon Capture & Climate Adaptation）",
     "url": "emerging/carbon-capture-and-climate-adaptation.html",
     "score": 0.2302
    },
    {
     "title": "氢能与绿色燃料（Hydrogen & Green Fuels）",
     "url": "emerging/hydrogen-and-green-fuels.html",
     "score": 0.2276
    },
    {
     "title": "电动汽车与交通电动化",
     "url": "emerging/evs-and-transport-electrification.html",
     "score": 0.2209
    }
   ]
  },
  {
   "i": 148,
   "title": "无人机与低空经济（Drones and Low-Altitude Economy）",
   "t": "无人机与低空经济（Drones and Low-Altitude Economy）",
   "u": "emerging/drones-and-low-altitude-economy.html",
   "s": "低空经济指以低空空域为依托、以常态化低空飞行活动为牵引，辐射带动相关领域形成的综合性经济业态，核心载体包括消费级/工业级无人机与 eVTOL（电动垂直起降飞行器，electric Vertical Take-Off and Landing）。2024 年两会期间低空经济首次被写入《政府工作报告》，2024 年 12 月国家发展改革委成立低空经济发展司统筹顶层…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "aviation",
    "xwzx",
    "market",
    "无人机与低空经济"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 18,
   "related": [
    {
     "title": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
     "url": "emerging/autonomous-delivery-and-logistics-robots.html",
     "score": 0.172
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.1441
    },
    {
     "title": "深空探索与太空经济",
     "url": "emerging/space-exploration-and-commerce.html",
     "score": 0.1359
    },
    {
     "title": "氢能与绿色燃料（Hydrogen & Green Fuels）",
     "url": "emerging/hydrogen-and-green-fuels.html",
     "score": 0.0934
    }
   ]
  },
  {
   "i": 149,
   "title": "氢能与绿色燃料（Hydrogen & Green Fuels）",
   "t": "氢能与绿色燃料（Hydrogen & Green Fuels）",
   "u": "emerging/hydrogen-and-green-fuels.html",
   "s": "氢能被视为难减排行业（钢铁、化工、重型交通、航运航空）脱碳的重要载体。按制取方式分为灰氢（化石燃料制氢）、蓝氢（化石制氢 + 碳捕集）、绿氢（可再生电力电解水）等；绿色燃料则指以绿氢为原料合成的绿氨、绿色甲醇以及可持续航空燃料（SAF）。2025—2026 年，行业从“大规模宣布项目”转向“以最终投资决策（FID）、绑定承购协议与政府支持为准”的现实阶段：一…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "hydrogen",
    "iea",
    "ccus",
    "green",
    "xwzx",
    "gnsz",
    "fid",
    "gdxw"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 14,
   "related": [
    {
     "title": "新能源与气候科技",
     "url": "emerging/new-energy-and-climate-tech.html",
     "score": 0.2276
    },
    {
     "title": "碳捕集与气候适应（Carbon Capture & Climate Adaptation）",
     "url": "emerging/carbon-capture-and-climate-adaptation.html",
     "score": 0.1486
    },
    {
     "title": "电动汽车与交通电动化",
     "url": "emerging/evs-and-transport-electrification.html",
     "score": 0.1097
    },
    {
     "title": "无人机与低空经济（Drones and Low-Altitude Economy）",
     "url": "emerging/drones-and-low-altitude-economy.html",
     "score": 0.0934
    }
   ]
  },
  {
   "i": 150,
   "title": "深空探索与太空经济",
   "t": "深空探索与太空经济",
   "u": "emerging/space-exploration-and-commerce.html",
   "s": "深空探索与太空经济涵盖载人航天（月球与火星任务、空间站）、无人探测（月球、火星、小行星样本返回）、在轨服务与制造，以及由此形成的商业市场规模。2025–2026 年，载人绕月飞行、月球采样与火星样本返回成为主线，同时近地轨道空间站正从政府主导的 ISS 向商业化空间站过渡，政府在轨预算因国防需求而结构性增长。ESA《Report on the Space E…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "nasa",
    "spacenexus",
    "space",
    "esa",
    "economy",
    "morgan",
    "cgtn",
    "china"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "火箭与航天发射",
     "url": "emerging/rockets-and-space-launch.html",
     "score": 0.2182
    },
    {
     "title": "无人机与低空经济（Drones and Low-Altitude Economy）",
     "url": "emerging/drones-and-low-altitude-economy.html",
     "score": 0.1359
    },
    {
     "title": "卫星星座与卫星通信",
     "url": "emerging/satellite-constellations-and-satcom.html",
     "score": 0.1337
    },
    {
     "title": "6G 与下一代连接",
     "url": "emerging/6g-and-next-gen-connectivity.html",
     "score": 0.0879
    }
   ]
  },
  {
   "i": 151,
   "title": "火箭与航天发射",
   "t": "火箭与航天发射",
   "u": "emerging/rockets-and-space-launch.html",
   "s": "火箭与航天发射是商业航天与卫星星座部署的基础环节，核心议题包括可复用火箭技术、单位载荷发射成本、全球发射频次与商业发射市场份额。可复用一级火箭通过回收与翻新显著摊薄成本，被认为是过去十年发射价格下降的主要驱动力之一。ESA《Report on the Space Economy 2026》数据显示，2025 年全球共进行 324 次轨道发射（其中 315 次…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 11,
   "tags": [
    "spacex",
    "rocket",
    "starship",
    "glenn",
    "launch",
    "esa",
    "space",
    "starlink"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 22,
   "related": [
    {
     "title": "卫星星座与卫星通信",
     "url": "emerging/satellite-constellations-and-satcom.html",
     "score": 0.2655
    },
    {
     "title": "深空探索与太空经济",
     "url": "emerging/space-exploration-and-commerce.html",
     "score": 0.2182
    },
    {
     "title": "6G 与下一代连接",
     "url": "emerging/6g-and-next-gen-connectivity.html",
     "score": 0.0938
    },
    {
     "title": "碳捕集与气候适应（Carbon Capture & Climate Adaptation）",
     "url": "emerging/carbon-capture-and-climate-adaptation.html",
     "score": 0.0696
    }
   ]
  },
  {
   "i": 152,
   "title": "电动汽车与交通电动化",
   "t": "电动汽车与交通电动化",
   "u": "emerging/evs-and-transport-electrification.html",
   "s": "电动汽车（EV）与交通电动化指以动力电池与电驱动系统替代内燃机的道路交通变革，涵盖纯电动（BEV）、插电式混动（PHEV）乘用车，以及重卡、轻商用车、巴士等商用车型，并延伸至充电与换电网络、电池与补能技术。按动力类型口径，中国官方统计口径为「新能源汽车（NEV）」，包含 BEV 与 PHEV。IEA《Global EV Outlook 2026》指出，202…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "iea",
    "catl",
    "executive-summary",
    "daily",
    "electric",
    "tesla",
    "china",
    "global"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "新能源与气候科技",
     "url": "emerging/new-energy-and-climate-tech.html",
     "score": 0.2209
    },
    {
     "title": "固态电池与下一代电池（Solid-State & Next-Gen Batteries）",
     "url": "emerging/solid-state-and-next-gen-batteries.html",
     "score": 0.2037
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.1318
    },
    {
     "title": "氢能与绿色燃料（Hydrogen & Green Fuels）",
     "url": "emerging/hydrogen-and-green-fuels.html",
     "score": 0.1097
    }
   ]
  },
  {
   "i": 153,
   "title": "碳捕集与气候适应（Carbon Capture & Climate Adaptation）",
   "t": "碳捕集与气候适应（Carbon Capture & Climate Adaptation）",
   "u": "emerging/carbon-capture-and-climate-adaptation.html",
   "s": "碳捕集、利用与封存（CCUS）及二氧化碳移除（CDR，含直接空气捕集 DAC）被视为实现净零排放的补充手段；气候适应则指通过基础设施、预警、保险与政策提升社会对已发生气候影响的抵御能力。二者在 2025—2026 年呈现两条相反的轨迹：CCUS/CDR 从技术示范走向有限商业化，但部署时间表普遍后移、成本仍未显著下降；气候适应则因极端天气加剧而融资缺口持续扩…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "climeworks",
    "ccus",
    "dac",
    "solar",
    "carbon",
    "capture",
    "mammoth",
    "carbon-removal"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "新能源与气候科技",
     "url": "emerging/new-energy-and-climate-tech.html",
     "score": 0.2302
    },
    {
     "title": "氢能与绿色燃料（Hydrogen & Green Fuels）",
     "url": "emerging/hydrogen-and-green-fuels.html",
     "score": 0.1486
    },
    {
     "title": "火箭与航天发射",
     "url": "emerging/rockets-and-space-launch.html",
     "score": 0.0696
    },
    {
     "title": "聚变能源（Fusion Energy）",
     "url": "emerging/fusion-energy.html",
     "score": 0.0633
    }
   ]
  },
  {
   "i": 154,
   "title": "空间计算与 XR",
   "t": "空间计算与 XR",
   "u": "emerging/spatial-computing-xr.html",
   "s": "空间计算（Spatial Computing）与扩展现实（XR，含 VR/AR/MR）指将数字内容与物理空间融合、并通过眼动、手势、语音等自然交互方式操作的计算形态。2025–2026 年，行业出现明显的结构性分化：一方面，MR/VR 头显（如 Apple Vision Pro、Samsung Galaxy XR、Meta Quest）聚焦生产力与沉浸体验，…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "galaxy",
    "idc",
    "ray-ban",
    "resource-center",
    "specs",
    "band",
    "xreal",
    "vision"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/公司（含官方链接）",
    "关键数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 24,
   "related": [
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0852
    },
    {
     "title": "人形机器人（Humanoid Robotics）",
     "url": "emerging/humanoid-robotics.html",
     "score": 0.0822
    },
    {
     "title": "数字健康与医疗器械（Digital Health and Medical Devices）",
     "url": "emerging/digital-health-and-medical-devices.html",
     "score": 0.0689
    },
    {
     "title": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
     "url": "emerging/autonomous-delivery-and-logistics-robots.html",
     "score": 0.0605
    }
   ]
  },
  {
   "i": 155,
   "title": "聚变能源（Fusion Energy）",
   "t": "聚变能源（Fusion Energy）",
   "u": "emerging/fusion-energy.html",
   "s": "聚变能源旨在模拟太阳内部的核反应，使轻原子核（主要是氘与氚）在极高温高压下结合并释放能量，被普遍视为近似无碳、燃料资源丰富、可提供基荷电力的潜在终极能源。技术路线主要分为两大家族：一是以托卡马克、仿星器、球形托卡马克为代表的磁约束，用强磁场约束上亿摄氏度的等离子体；二是以激光或脉冲驱动靶丸压缩的惯性约束。评价聚变性能的核心指标包括聚变增益 Q（聚变输出能量与…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 10,
   "tags": [
    "iter",
    "nif",
    "cfs",
    "llnl",
    "ignition",
    "fusion",
    "energy",
    "target"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 30,
   "related": [
    {
     "title": "新能源与气候科技",
     "url": "emerging/new-energy-and-climate-tech.html",
     "score": 0.2467
    },
    {
     "title": "小型模块化反应堆与核能（Small Modular Reactors & Nuclear）",
     "url": "emerging/small-modular-reactors.html",
     "score": 0.1072
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0661
    },
    {
     "title": "碳捕集与气候适应（Carbon Capture & Climate Adaptation）",
     "url": "emerging/carbon-capture-and-climate-adaptation.html",
     "score": 0.0633
    }
   ]
  },
  {
   "i": 156,
   "title": "脑机接口（Brain-Computer Interfaces, BCI）",
   "t": "脑机接口（Brain-Computer Interfaces, BCI）",
   "u": "emerging/brain-computer-interfaces.html",
   "s": "脑机接口（Brain-Computer Interface, BCI）通过采集、解码神经信号并转换为对外部设备的控制指令，或对神经系统进行刺激，建立大脑与外部设备之间的直接通信通道。按侵入程度可分为侵入式（implanted/invasive）、半侵入式（endovascular 等）与非侵入式（non-invasive，如头皮脑电 EEG）。2026 年，…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 8,
   "tags": [
    "bci",
    "brain",
    "stories",
    "switch",
    "live",
    "future",
    "脑机接口"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 19,
   "related": [
    {
     "title": "数字健康与医疗器械（Digital Health and Medical Devices）",
     "url": "emerging/digital-health-and-medical-devices.html",
     "score": 0.165
    },
    {
     "title": "人形机器人（Humanoid Robotics）",
     "url": "emerging/humanoid-robotics.html",
     "score": 0.1043
    },
    {
     "title": "6G 与下一代连接",
     "url": "emerging/6g-and-next-gen-connectivity.html",
     "score": 0.1021
    },
    {
     "title": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
     "url": "emerging/gene-editing-and-cell-therapy.html",
     "score": 0.0956
    }
   ]
  },
  {
   "i": 157,
   "title": "自动驾驶",
   "t": "自动驾驶",
   "u": "emerging/autonomous-driving.html",
   "s": "自动驾驶（Autonomous Driving）指车辆在无需或仅需少量人类干预下完成感知、决策与控制的技术体系。按 SAE 分级，L2 为组合驾驶辅助，L3 为有条件自动驾驶，L4 为高度自动驾驶。2025–2026 年，行业同时出现三条主线：一是 Robotaxi（无人出租车）从测试走向规模化商业运营；二是乘用车智驾从模块化方案转向端到端（End-to-E…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 13,
   "tags": [
    "waymo",
    "tesla",
    "fsd",
    "vla",
    "impact",
    "info",
    "safety",
    "company"
   ],
   "sections": [
    "概述",
    "2025–2026 最新进展",
    "核心技术与关键概念",
    "代表性项目/公司（含官方链接）",
    "关键数据",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 39,
   "related": [
    {
     "title": "人形机器人（Humanoid Robotics）",
     "url": "emerging/humanoid-robotics.html",
     "score": 0.1648
    },
    {
     "title": "无人机与低空经济（Drones and Low-Altitude Economy）",
     "url": "emerging/drones-and-low-altitude-economy.html",
     "score": 0.1441
    },
    {
     "title": "电动汽车与交通电动化",
     "url": "emerging/evs-and-transport-electrification.html",
     "score": 0.1318
    },
    {
     "title": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
     "url": "emerging/autonomous-delivery-and-logistics-robots.html",
     "score": 0.1282
    }
   ]
  },
  {
   "i": 158,
   "title": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
   "t": "配送机器人与物流自动化（Autonomous Delivery and Logistics Robots）",
   "u": "emerging/autonomous-delivery-and-logistics-robots.html",
   "s": "配送机器人与物流自动化覆盖两个主要方向：一是面向末端配送的无人配送车（无人城配车、配送机器人），二是面向仓内的仓储机器人（AGV/AMR、货到人系统、自动化立体库）与无人仓。2025–2026 年，中国无人配送车进入规模化拐点，仓储自动化则在经历资本热潮回落后的整合期，头部企业以更清晰的单位经济模型持续部署。 最新进展（2025–2026） - 无人配送车规…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 7,
   "tags": [
    "warehouse",
    "amr",
    "agv",
    "automation",
    "配送机器人与物流自动化"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 16,
   "related": [
    {
     "title": "无人机与低空经济（Drones and Low-Altitude Economy）",
     "url": "emerging/drones-and-low-altitude-economy.html",
     "score": 0.172
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.1282
    },
    {
     "title": "工业机器人与自动化（Industrial Robotics and Automation）",
     "url": "emerging/industrial-robotics-and-automation.html",
     "score": 0.1213
    },
    {
     "title": "人形机器人（Humanoid Robotics）",
     "url": "emerging/humanoid-robotics.html",
     "score": 0.113
    }
   ]
  },
  {
   "i": 159,
   "title": "长寿与抗衰老研究（Longevity and Aging Research）",
   "t": "长寿与抗衰老研究（Longevity and Aging Research）",
   "u": "emerging/longevity-and-aging-research.html",
   "s": "长寿与抗衰老研究（longevity / geroscience）以「衰老是可干预的生物学过程」为前提，试图通过药理学或基因干预延缓衰老、延长健康寿命（healthspan）。其理论框架以衰老标志（hallmarks of aging）为核心，涵盖基因组不稳定、表观遗传改变、细胞衰老、线粒体功能障碍、慢性炎症（inflammaging）等维度，并催生「老年治…",
   "dm": "新兴科技",
   "d": "emerging",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "biosciences",
    "trial",
    "mouse",
    "insilico",
    "prevention",
    "treatment",
    "human",
    "disease"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性项目 / 公司 / 产品（附官方链接）",
    "关键数据与评测结果（附来源）",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 21,
   "related": [
    {
     "title": "AI 制药（AI Drug Discovery）",
     "url": "emerging/ai-drug-discovery.html",
     "score": 0.1986
    },
    {
     "title": "基因编辑与细胞治疗（Gene Editing and Cell Therapy）",
     "url": "emerging/gene-editing-and-cell-therapy.html",
     "score": 0.1811
    },
    {
     "title": "合成生物学与生物制造（Synthetic Biology and Bio-Manufacturing）",
     "url": "emerging/synthetic-biology-and-bio-manufacturing.html",
     "score": 0.1172
    },
    {
     "title": "数字健康与医疗器械（Digital Health and Medical Devices）",
     "url": "emerging/digital-health-and-medical-devices.html",
     "score": 0.0905
    }
   ]
  },
  {
   "i": 160,
   "title": "AI 监管与政策（全球）",
   "t": "AI 监管与政策（全球）",
   "u": "industry/ai-regulation-and-policy.html",
   "s": "AI 监管与政策（全球） 最后更新：2026-09-26 ｜ 领域：人工智能治理 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 截至 2026 年 9 月，全球人工智能监管已从\"原则宣言\"阶段进入\"分阶段执法\"阶段。欧盟凭借《人工智能法案》（AI Act）建立了全球首个以风险分级为核心的法律框架，并已进入通用人工智能（GPAI）义务的执行期；…",
   "dm": "产业与政策",
   "d": "industry",
   "ud": "2026-09-26",
   "rm": 19,
   "tags": [
    "ftc",
    "gpai",
    "summit",
    "zhengce",
    "general-purpose",
    "india",
    "impact",
    "policies"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心概念与关键条款",
    "四、代表性机构与官方链接",
    "五、关键数据（带来源）",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 57,
   "related": [
    {
     "title": "数字权利与监控",
     "url": "society/digital-rights-and-surveillance.html",
     "score": 0.1724
    },
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.1697
    },
    {
     "title": "Web3 与数字资产",
     "url": "industry/web3-and-digital-assets.html",
     "score": 0.1655
    },
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.1639
    }
   ]
  },
  {
   "i": 161,
   "title": "Web3 与数字资产",
   "t": "Web3 与数字资产",
   "u": "industry/web3-and-digital-assets.html",
   "s": "Web3 与数字资产 最后更新：2026-09-26 ｜ 领域：区块链与数字金融 ｜ 说明：信息来源为公开网络资料，详见文末参考来源 一、概述 2026 年的 Web3 与数字资产行业，已从\"加密原生叙事\"转向\"机构基础设施 + 全球监管落地\"的双轮驱动。一方面，比特币、以太坊完成了从\"实验网络\"到\"受监管资产类别\"的身份转换：现货比特币 ETF 规模突破千…",
   "dm": "产业与政策",
   "d": "industry",
   "ud": "2026-09-26",
   "rm": 18,
   "tags": [
    "crypto",
    "mica",
    "rollup",
    "blob",
    "mas",
    "insight",
    "sec",
    "monitor"
   ],
   "sections": [
    "一、概述",
    "二、2025–2026 最新进展",
    "三、核心概念与关键条款",
    "四、代表性案例/机构（带官方链接）",
    "五、关键数据（带来源）",
    "六、趋势与争议",
    "参考来源"
   ],
   "refs": 58,
   "related": [
    {
     "title": "AI 监管与政策（全球）",
     "url": "industry/ai-regulation-and-policy.html",
     "score": 0.1655
    },
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.1037
    },
    {
     "title": "威胁态势 2026",
     "url": "security/threat-landscape-2026.html",
     "score": 0.0963
    },
    {
     "title": "自动驾驶",
     "url": "emerging/autonomous-driving.html",
     "score": 0.0937
    }
   ]
  },
  {
   "i": 162,
   "title": "数字权利与监控",
   "t": "数字权利与监控",
   "u": "society/digital-rights-and-surveillance.html",
   "s": "数字权利与监控（digital rights and surveillance）关注个人在数字化环境中的隐私权、免受无差别监控的权利、数据经纪（data broker）对个人信息的商业化利用，以及加密技术同执法取证需求之间的张力。2025–2026 年，这一领域的主要变化是：立法与司法开始对生物识别监控、数据经纪和端到端加密的强制访问设定更明确的边界，同时不…",
   "dm": "社会与数字权利",
   "d": "society",
   "ud": "2026-09-26",
   "rm": 9,
   "tags": [
    "privacy",
    "drop",
    "delete",
    "protection",
    "information",
    "over",
    "act",
    "数字权利与监控"
   ],
   "sections": [
    "概述",
    "最新进展（2025–2026）",
    "核心技术与关键概念",
    "代表性案例",
    "关键数据与评测结果",
    "趋势与争议",
    "参考来源"
   ],
   "refs": 17,
   "related": [
    {
     "title": "AI 监管与政策（全球）",
     "url": "industry/ai-regulation-and-policy.html",
     "score": 0.1724
    },
    {
     "title": "合规与安全治理",
     "url": "security/compliance-and-security-governance.html",
     "score": 0.1067
    },
    {
     "title": "数据治理与隐私",
     "url": "data/data-governance-and-privacy.html",
     "score": 0.0937
    },
    {
     "title": "AI 安全与对齐（AI Safety and Alignment）",
     "url": "ai/ai-safety-and-alignment.html",
     "score": 0.0922
    }
   ]
  }
 ]
}