知识增量Knowledge Deltav0.1无限进步
简体中文 当前ChineseEnglish 即将推出English日本語 即将推出Japanese한국어 即将推出Korean繁體中文 即将推出Traditional ChineseEspañol 即将推出SpanishFrançais 即将推出FrenchDeutsch 即将推出German

卡片库

398 个词，按八类排成目录。

理解模型Models

AI 是什么

17
人工智能Artificial Intelligence机器学习Machine Learning深度学习Deep Learning生成式人工智能Generative Artificial Intelligence通用人工智能Artificial General Intelligence大语言模型Large Language Model小语言模型Small Language Model基础模型Foundation Model模型Model参数Parameter权重Weight超参数Hyperparameter自然语言处理Natural Language Processing分类任务Classification回归预测Regression窄域人工智能Artificial Narrow Intelligence超人工智能Artificial Superintelligence

模型的结构

19
神经网络Neural Network层Layer激活函数Activation FunctionTransformer 架构Transformer注意力机制Attention自注意力Self-Attention多头注意力Multi-Head Attention编码器Encoder解码器Decoder自回归生成Autoregressive Generation嵌入向量Embedding上下文窗口Context Window词元Token分词器Tokenizer位置编码Positional Encoding混合专家模型Mixture of Experts扩散模型Diffusion Model生成对抗网络Generative Adversarial Network变分自编码器Variational Autoencoder

怎样生成答案

19
推理运行Inference解码策略Decoding Strategy贪心解码Greedy Decoding束搜索Beam Search随机采样Sampling温度Temperature核采样Nucleus Sampling前 k 项采样Top-k Sampling词元分数LogitsSoftmax 函数Softmax随机种子Random Seed推理型模型Reasoning Model测试时计算Test-Time Compute幻觉Hallucination键值缓存Key-Value Cache输入预填充Prefill逐步生成阶段Decode PhaseFlashAttentionFlashAttention长上下文能力Long-Context Capability

训练与微调

23
训练Training预训练Pretraining自监督学习Self-Supervised Learning监督学习Supervised Learning无监督学习Unsupervised Learning微调Fine-Tuning监督微调Supervised Fine-Tuning指令微调Instruction Tuning强化学习Reinforcement Learning人类反馈强化学习Reinforcement Learning from Human FeedbackAI 反馈强化学习Reinforcement Learning from AI Feedback奖励模型Reward Model近端策略优化Proximal Policy Optimization直接偏好优化Direct Preference Optimization参数高效微调Parameter-Efficient Fine-Tuning低秩适配Low-Rank Adaptation量化低秩适配Quantized Low-Rank Adaptation知识蒸馏Knowledge Distillation量化Quantization剪枝Pruning检查点Checkpoint持续预训练Continued Pretraining上下文学习In-Context Learning

训练里的数学

14
训练集Training Set验证集Validation Set测试集Test Set训练轮次Epoch批大小Batch Size训练步Training Step损失函数Loss Function交叉熵Cross-Entropy梯度下降Gradient Descent反向传播Backpropagation学习率Learning Rate优化器OptimizerAdam 优化器Adaptive Moment Estimation正则化Regularization

怎样评价模型

13
准确率Accuracy精确率Precision召回率RecallF1 分数F1 Score混淆矩阵Confusion Matrix概率校准Probability Calibration模型评估Evaluation基准测试Benchmark评测数据污染Benchmark Data Contamination数据泄漏Data Leakage过拟合Overfitting泛化Generalization困惑度Perplexity (metric)

图像、声音和视频

8
多模态Multimodal视觉语言模型Vision-Language Model光学字符识别Optical Character Recognition自动语音识别Automatic Speech Recognition文本转语音Text-to-Speech声音克隆Voice Cloning文本生成图像Text-to-Image Generation文本生成视频Text-to-Video Generation

学会提问Prompting

把话说清楚

15
提示词Prompt系统指令System instruction开发者指令Developer instruction用户指令User instruction指令层级Instruction hierarchy零样本提示Zero-shot prompting少样本提示Few-shot prompting思维链提示Chain-of-thought prompting自洽性采样Self-consistency提示工程Prompt engineering上下文工程Context engineering智能体记忆Agent memory上下文压缩Compaction上下文退化Context rot渐进加载Progressive disclosure

结构化输出

5
工具调用Tool calling工具结果Tool result数据结构约定Schema结构化输出Structured outputJSON 模式JSON mode

质量与安全

9
模型与应用评估Evals沙箱Sandbox权限Permissions人在回路Human-in-the-loop护栏Guardrail提示注入Prompt injection越狱提示Jailbreak可识别个人信息Personally identifiable information红队测试Red teaming

委派任务Agents

智能体与工作流

14
推理与行动交替ReAct先计划后执行Plan-and-execute智能体Agent智能体式Agentic工作流Workflow任务编排Orchestration子智能体Subagent多智能体系统Multi-agent system交接Handoff智能体运行框架Agent harness氛围编程Vibe coding智能体编程Agentic codingAI 辅助编程AI-assisted programming提示串联Prompt chaining

技能与插件

5
技能包Agent Skill技能说明文件SKILL.md插件Plugin连接器Connector自定义 GPTCustom GPT

MCP 与连接

13
模型上下文协议Model Context ProtocolMCP 宿主MCP hostMCP 客户端MCP clientMCP 服务器MCP serverMCP 工具MCP toolMCP 资源MCP resourceMCP 提示模板MCP prompt传输方式Transport标准输入输出传输stdio transport可流式 HTTP 传输Streamable HTTP服务器发送事件Server-Sent Events智能体间协议Agent2Agent Protocol智能体名片Agent Card

选好工具Tools

对话助手

7
ChatGPTChatGPTClaudeClaudeGeminiGemini AppsDeepSeekDeepSeek Chat千问Qianwen / Qwen ChatKimiKimi豆包Doubao

读文献、找资料

5
PerplexityPerplexityNotebookLMNotebookLM / Gemini NotebookZoteroZoteroObsidianObsidianOverleafOverleaf

编程助手

7
CodexCodexClaude CodeClaude CodeCursorCursorGitHub CopilotGitHub CopilotVisual Studio CodeVisual Studio CodeWindsurfWindsurfTRAETRAE

模型平台与本地运行

7
Google AI StudioGoogle AI StudioHugging FaceHugging Face魔搭社区ModelScopeOllamaOllamaLM StudioLM StudioOpen WebUIOpen WebUIOpenRouterOpenRouter

工作流与生成

4
ComfyUIComfyUIn8nn8nDifyDifyFlowiseFlowise

接口与成本

17
流式输出Streaming批处理Batch processing提示缓存Prompt caching速率限制Rate limits延迟Latency首个输出等待时间Time to First Token吞吐量Throughput可观测性Observability运行追踪Tracing应用编程接口Application Programming Interface软件开发工具包Software Development Kit服务端点Endpoint接口密钥API key事件回调Webhook本地模型Local model开放权重Open weights开源 AIOpen-source AI

补齐基础Coding

文件与格式

8
MarkdownMarkdownJSONJavaScript Object NotationYAMLYAML Ain't Markup LanguageXMLExtensible Markup LanguageCSVComma-Separated ValuesUnicodeUnicodeUTF-8Unicode Transformation Format, 8-bit正则表达式Regular Expression

语言与运行

13
SQLStructured Query LanguagePythonPython计算笔记本Computational Notebook命令行界面Command-Line Interface终端TerminalShellShell文件路径File Path当前工作目录Current Working Directory环境变量Environment Variable软件包与模块Package and Module软件依赖Software Dependency虚拟环境Virtual Environment脚本Script

Git 与协作

23
GitGit仓库RepositoryGitHubGitHub提交Commit分支Branch合并Merge差异比较Diff合并请求Pull Request补丁Patch统一差异格式Unified diff差异块Hunk工作树Working tree暂存区Staging area / Index克隆Clone获取远程更新Fetch拉取并整合Pull推送Push派生仓库Fork远程仓库配置Remote问题与任务记录Issue项目说明文件README许可证License发布版本Release

网络与网页

17
图形用户界面Graphical user interface集成开发环境Integrated development environment前端Frontend后端Backend超文本标记语言HyperText Markup Language层叠样式表Cascading Style SheetsJavaScript语言JavaScript统一资源定位符Uniform Resource LocatorHTTP与HTTPSHTTP / HTTPS端口Port本机回环名称Localhost域名系统Domain Name System域名Domain name服务器Server客户端Client安全远程连接协议Secure Shell容器Container

硬件

6
中央处理器Central Processing Unit图形处理器Graphics Processing Unit显存Video Random-Access Memory系统内存Random-Access Memory统一内存Unified Memory固态存储Solid-State Drive

处理数据Data

数据整理

10
数据清洗Data cleaning缺失值Missing value异常值Outlier变量编码Variable coding宽表与长表Wide and long data合并与匹配Data merging and matching主键Primary key面板数据Panel data数据字典Data dictionary去重Deduplication

统计分析

12
描述统计Descriptive statistics回归分析Regression analysis固定效应Fixed effects双重差分Difference-in-differences工具变量Instrumental variable标准误Standard error统计显著性Statistical significance稳健性检验Robustness checks异质性分析Heterogeneity analysis因果推断Causal inference内生性Endogeneity样本选择偏差Sample selection bias

可视化与报告

5
统计图Statistical chart置信区间Confidence interval结果表格Results table可重复报告Reproducible report数据可视化原则Data visualization principles

让 AI 做分析

5
代码解释器Code interpreter数据分析智能体Data analysis agent让 AI 写 Stata 或 R 代码AI-assisted Stata or R codingAI 分析结果核对Checking AI-generated analysis合成数据Synthetic data

读写文献Literature

检索

8
检索式Search query布尔运算Boolean operators学术数据库Academic database中国知网China National Knowledge InfrastructureWeb of ScienceWeb of ScienceGoogle ScholarGoogle Scholar引文追踪Citation chasing学术搜索引擎Academic search engine

阅读与综述

8
摘要Abstract精读Close reading文献笔记Literature notes证据矩阵Evidence matrix文献综述Literature review系统综述Systematic review元分析Meta-analysis研究空白Research gap

写作与发表

9
论文结构Paper structure研究问题Research question研究贡献Research contribution学术润色Academic language editing学术翻译Academic translation投稿Journal submission审稿意见回复Response to reviewers预印本Preprint开放获取Open access

引用管理

5
参考文献格式Bibliographic reference style引用管理软件Reference management softwareBibTeXBibTeX引文样式语言Citation Style LanguageDOI 与文献元数据DOI and bibliographic metadata

AI 检索与证据

12
检索增强生成Retrieval-Augmented Generation信息检索Retrieval向量数据库Vector database文本分块Chunking重排序Reranking混合检索Hybrid search语义检索Semantic search知识图谱Knowledge graph图谱增强生成GraphRAG证据锚定Grounding引用标注Citation来源链Provenance

科研规范Integrity

AI 使用与署名

8
AI 使用声明AI use disclosure作者资格Authorship criteria期刊 AI 政策Journal AI policy学位论文 AI 规定AI rules for theses and dissertations生成内容标注AI-generated content labelling出版伦理委员会Committee on Publication Ethics人工最终责任Human accountability致谢Acknowledgements

引用与核查

8
虚构文献Fabricated reference引用核查Citation verificationDOIDigital Object Identifier原文核对Checking the original source二次引用Secondary citation抄袭与改写Plagiarism and paraphrasing查重Text similarity checkingAI 检测AI-generated text detection

数据与伦理

8
数据隐私Data privacy知情同意Informed consent伦理审查Ethics review去标识化De-identification敏感数据Sensitive data数据使用协议Data use agreement数据留存与训练Data retention and model training本地处理Local processing

可复现与开放

7
可复现性Reproducibility复现包Replication package研究日志Research log提示词留存Prompt and output recordkeeping预注册Preregistration开放数据与代码Open data and code结果挑选Selective reporting

评审与教学

5
审稿保密Peer review confidentiality同行评审中的 AIAI use in peer review课程 AI 政策Course AI policy作业中的 AIAI use in coursework学术不端Academic misconduct