Why Was The Previous Stage Not Enough?
Understand the bottleneck first, then see why a new technology appeared.
Chapter 00
Overview: Why did AI not suddenly become large models?
AI's main thread is not the sudden appearance of one model. It is the result of rules and search, knowledge engineering, probability and statistics, representation learning, Transformer, pretraining and post-training, RAG, and agents repeatedly addressing earlier bottlenecks.
Why did AI not suddenly become large models?
Historical Spine
From Rules To Systems
Rules And KnowledgeExplainable, But Hard To Scale
Search and expert systems showed that rules can express reasoning, while exposing combinatorial explosion and the cost of maintaining exceptions.
Data And RepresentationAble To Learn, But Dependent On Examples
Statistical learning and neural networks shifted the problem toward data, features, and representations, allowing models to generalize from examples.
Foundation Models And SystemsPowerful, But In Need Of External Structure
Transformers support scaled pretraining and post-training shapes cooperative behavior; RAG, tools, memory, and evaluation then organize models into modern AI systems.
How To Read
Four Things To Notice In Every Chapter
Use One Action To Gain One Intuition
Each demo teaches one core aha moment instead of packing real-system complexity into an introductory explanation.
What Did The New Technology Solve And Leave Behind?
RAG, agents, and evaluation are not endpoints. They continue to address factuality, controllability, and reliability.
Simplification note
This Is A Learning Map, Not A Complete AI Encyclopedia
The overview deliberately keeps the main thread and representative technologies while omitting many branches, people, papers, and engineering details. Later chapters use interactive demos to develop one core mechanism intuition for each era.
References
Concept Check
Test The Core Intuition With One Question
This check supports reflection and never blocks the next chapter. Records stay on this device.