Appendix A: Glossary

This glossary includes core concepts and common terminology in the AI programming field, helping you quickly understand technical documentation and industry discussions.

Basic Concepts

English Term	Chinese Translation	Brief Description
AI (Artificial Intelligence)	Artificial Intelligence	Technical field enabling machines to simulate human intelligent behavior
ML (Machine Learning)	Machine Learning	Subset of AI, training models through data rather than explicit programming
DL (Deep Learning)	Deep Learning	Machine learning approach based on neural networks
LLM (Large Language Model)	Large Language Model	Large-scale neural network models trained on massive text data
NLP (Natural Language Processing)	Natural Language Processing	Technology enabling computers to understand and generate human language
Transformer	Transformer Architecture	Core architecture of modern LLMs, based on attention mechanism
Token	Token	Basic unit of text processing by models, typically word or subword
Embedding	Embedding/Vectorization	Process of converting text to numerical vectors
Fine-tuning	Fine-tuning	Continued training on pre-trained models for specific tasks
Pre-training	Pre-training	Initial training stage on large-scale data
Context Window	Context Window	Maximum number of tokens a model can process at once
Neural Network	Neural Network	Computational model mimicking biological neurons
Training Data	Training Data	Dataset used to train models
Model Parameters	Model Parameters	Number of learnable weights in a neural network

Models and Inference

English Term	Chinese Translation	Brief Description
GPT (Generative Pre-trained Transformer)	Generative Pre-trained Transformer	OpenAI's LLM series (GPT-4.1, GPT-5, etc.)
Claude	Claude	Anthropic's LLM series (Claude Sonnet 4.6, Opus 4.6)
Gemini	Gemini	Google's multimodal LLM series (Gemini 2.5, 3.1 Pro)
DeepSeek	DeepSeek	Open-source LLM with excellent performance (V3.2, R1)
Temperature	Temperature Parameter	Controls output randomness, 0=deterministic, 2=high randomness
Top-p (Nucleus Sampling)	Nucleus Sampling	Sample from candidate words with cumulative probability reaching p
Max Tokens	Max Tokens	Maximum length limit for generated responses
Inference	Inference	Process of using trained model to generate output
Reasoning Model	Reasoning Model	Models with enhanced logical reasoning capabilities (e.g., o1)
Chain-of-Thought (CoT)	Chain-of-Thought	Technique for models to show reasoning process step by step
Few-Shot Learning	Few-Shot Learning	Guiding model with few examples in prompt
Zero-Shot Learning	Zero-Shot Learning	Having model execute task without providing examples
Multimodal	Multimodal	Can process multiple inputs like text, images, audio
Vision Model	Vision Model	Models that can understand and analyze images

Development Concepts

English Term	Chinese Translation	Brief Description
Prompt Engineering	Prompt Engineering	Technique of designing effective prompts to get desired outputs
Context Engineering	Context Engineering	Optimizing input context to enhance model performance
Vibe Coding	Vibe Coding	Programming collaboratively with AI through natural language
AI-First Development	AI-First Development	Software development paradigm with AI at its core
Copilot	Copilot	General term for AI-assisted programming tools
Function Calling	Function Calling	Capability for LLM to call external functions or APIs
Tool Use	Tool Use	Enabling AI to use external tools to extend capabilities
Structured Output	Structured Output	Forcing model output to conform to specific JSON Schema
Streaming	Streaming	Returning generated results progressively rather than waiting for completion
SSE (Server-Sent Events)	Server-Sent Events	HTTP protocol for implementing streaming responses
System Prompt	System Prompt	Initial instructions defining AI role and behavior
User Message	User Message	Input content sent to AI by user
Assistant Message	Assistant Message	Response content returned by AI
Token Limit	Token Limit	Maximum token count per request

Agent Related

English Term	Chinese Translation	Brief Description
AI Agent	AI Agent	AI system capable of autonomous decision-making and task execution
ReAct (Reasoning + Acting)	Reasoning-Action Pattern	Framework where agents alternate between reasoning and action
Multi-Agent	Multi-Agent	Multiple agents collaborating to complete complex tasks
Sub-Agent	Sub-Agent	Specialized agents called by main agent
Handoff	Handoff	Transfer of task control between agents
Computer Use	Computer Use	AI directly operating computer interface (e.g., Claude)
Agentic AI	Agentic AI	AI systems with autonomy
Planning	Planning	Agent's ability to formulate task execution plans
Memory	Memory	Agent's ability to store and retrieve historical information
Reflection	Reflection	Agent's ability to evaluate and improve own behavior
Autonomous Agent	Autonomous Agent	Can complete tasks independently without human intervention
Tool Agent	Tool Agent	Agent specialized in calling external tools

Protocols and Infrastructure

English Term	Chinese Translation	Brief Description
MCP (Model Context Protocol)	Model Context Protocol	AI context standard protocol proposed by Anthropic
A2A (Agent-to-Agent)	Agent-to-Agent Protocol	Agent communication standard proposed by Google
ANP (Agent Network Protocol)	Agent Network Protocol	Agent interconnection standard proposed by OpenAI
RAG (Retrieval-Augmented Generation)	Retrieval-Augmented Generation	Generation technique combining external knowledge bases
Vector Database	Vector Database	Specialized database for storing and retrieving vector data
Semantic Search	Semantic Search	Search based on semantic similarity rather than keywords
Chunking	Chunking	Splitting long documents into smaller chunks for processing
Reranking	Reranking	Re-sorting retrieval results by relevance
API Gateway	API Gateway	Unified management and routing of API requests
Load Balancing	Load Balancing	Distributing requests across multiple service instances

Production

English Term	Chinese Translation	Brief Description
Guardrails	Guardrails	Mechanisms limiting model output range and quality
Evaluation	Evaluation	Measuring model performance and output quality
Observability	Observability	Ability to monitor and debug AI systems
Prompt Caching	Prompt Caching	Caching common prompts to reduce cost and latency
Batch API	Batch API	Asynchronously batch processing requests to reduce cost
Prompt Injection	Prompt Injection	Attacks where malicious users try to manipulate model behavior
Jailbreak	Jailbreak	Techniques to bypass model security restrictions
Rate Limiting	Rate Limiting	Limiting API call frequency to prevent abuse
Cost Optimization	Cost Optimization	Strategies to reduce AI application operating costs
Model Versioning	Model Versioning	Managing deployment and switching of different model versions
A/B Testing	A/B Testing	Comparing effectiveness of different models or prompts
Fallback	Fallback	Backup plan when main service fails
Monitoring	Monitoring	Real-time tracking of system running status and metrics
Logging	Logging	Recording request, response, and error information

Recommended Learning Resources

Resource	Description	Link
learn-claude-code	Build a Claude Code-like agent from scratch in 11 sessions	GitHub
hello-agents	Comprehensive guide to building AI agents from zero (Chinese)	GitHub
OpenAI Cookbook	Official OpenAI code examples and best practices	GitHub
Anthropic Docs	Claude API documentation and prompt engineering guide	docs.anthropic.com
LangChain Docs	Popular LLM application framework documentation	python.langchain.com
MCP Specification	Model Context Protocol official specification	modelcontextprotocol.io

---

Last updated: 2026-02-20