AI vs Machine Learning
AI is the broader goal and problem domain. Machine learning is just one implementation method, not the whole story.
AI, Machine Learning & LLM Map
This page doesn't start with formulas, but first builds a stable "technology map". Without a map, learning Token, Attention, Transformer, and RAG later would be like memorizing scattered terms; with a map, you'll know where each concept fits in the entire technical chain.
Click a Layer, See the Details
Click a card on the left, and the right side will show what problems this layer solves, its typical tasks, and its relationship to subsequent LLM learning.
Learning Tip: When encountering new terms in the future, first ask yourself whether it's about data representation, model structure, training methods, inference strategies, or application systems. Classifying it first will make understanding much faster.
Recommended Learning Path
1
Start with the Map
Understand which layers you'll learn to avoid being overwhelmed by terminology at the start.
2
Then Representation
Understand how text becomes tokens, vectors, and matrices before diving into Attention.
3
Then Transformer
Dive into core mechanisms like QKV, positional encoding, Mask, and Transformer Block.
4
Finally Systems
Understand RAG, Tool Use, Agent, evaluation, safety, and deployment.
4 Most Confused Concept Pairs
Machine Learning vs Deep Learning
Deep learning is an important branch of machine learning, specialized in learning complex representations from high-dimensional data.
Deep Learning vs Large Models
LLMs aren't a separate world; they're a scaled pre-trained model approach built on deep learning.
Model Principles vs Product Systems
Capabilities like "can retrieve", "can use tools", and "can answer like an assistant" often come from the combined effect of model and system, not just a single network structure.
Common Misconception: When many people hear "LLM technology", they only think of Transformer or Attention. But that's just the model structure layer - real deployment capability also depends on training, inference, and system layers.
What to Learn After This Page
If You're Completely New
Recommended next step: "Characters, Tokens & Vectors" to understand text representation first.
If You Want to Try Interactive Labs First
You can go directly to the Transformer self-attention lab, then come back to complete the roadmap and foundational pages.
One-sentence summary: LLMs aren't isolated points; they're a technical map unfolding layer by layer from data representation, model structure, training, inference to application systems.