Have you ever opened a technical knowledge base, stared at a dense list of concepts for three minutes, and wondered: "Where do I even start?"
You're not alone. AI now involves hundreds of concepts—from LLM to RAG, LoRA to RLHF, Transformer to Agent. Each has its own manuals, tutorials, and papers. But what's missing is the relationship between them: what should you learn after this one?
Easy AI's newly launched concept map is built to answer exactly that question.
A Subway Map That Answers Three Questions
As the project documentation states, the map is designed to answer three questions: where to start, how to learn, and what you'll learn. A traditional knowledge list is like a dictionary sorted alphabetically; a subway map's logic is: follow the line, stop by stop, to your destination.
The map organizes all of Easy AI's knowledge into 7 color-coded learning lines, each a complete path from beginner to advanced. Starting from LLM, following one line takes you systematically through a whole domain. For example, the prompt engineering line goes Prompt → System Prompt → Few-shot → Chain of Thought—four stations forming a complete beginner-to-advanced learning loop.
Boustrophedon Folding: Fitting an Infinite Line on a Finite Screen
The most interesting piece of engineering is the layout algorithm. If a learning line has 20 concepts laid out left-to-right, the screen can't fit them. Easy AI's solution is boustrophedon folding—a Greek-derived word meaning "moving back and forth like plowing a field."
Concretely: each row holds at most 6 stations; when full, the line folds to the next row and reverses direction. Row 1 goes left-to-right, row 2 right-to-left, row 3 left-to-right again... Curved arcs smooth out each turn, so the whole line looks like a coiled snake—compact but not overly long. The algorithm is implemented entirely with hand-written SVG + React, with no third-party graphics library, which is very performance-friendly.
Six Relationship Types: Wiring Knowledge Together
Subway lines alone aren't enough. Horizontal relationships between concepts—like needing LoRA before understanding LoRA rank, or quantization being a compression step before deployment—must also be visualized, or learners still can't see the web-like structure.
The map uses 6 colored edge types to connect concepts:
- Prerequisite (green) — understand A before B
- Mechanism (coral) — A is a part inside B
- Parallel (purple dashed) — same level, horizontally comparable
- Solution (dark green) — A solves B's problem
- Contrast (orange dashed) — alternatives or commonly confused pairs
- Cross-topic bridge (blue dashed) — connects two major themes
Hub Nodes: Main Gates and Side Doors
Some concepts are drawn especially large—like LLM, Transformer, Agent, LoRA, and Quantization. These hub nodes are the field's core entry points, like major transfer stations on a subway map that nearly every line passes through. Whichever line you study, you'll eventually hit these hubs, so they're visually enlarged for quick orientation.
There's also a thoughtful touch: the LLM hub node carries a glowing "recommended starting point" hint. For newcomers who have no idea where to begin, this dramatically reduces decision cost.
Why This Matters
The concept map isn't a nice-to-have. AI knowledge is exploding faster than humans can read linearly. Facing a flat, unordered list every time you open a knowledge base creates heavy cognitive load. A structured map essentially helps learners build a cognitive map—not just memorizing facts, but knowing where each fact sits in the knowledge system.
It's like learning geography: you can memorize every city's name, but only when you open a map and see the mountains, rivers, and oceans between them do you truly understand the world. Easy AI's map is doing that for AI knowledge.
Just as importantly, the data layer is cleanly designed. Node information all derives from a single data source; edges are maintained separately in dedicated files. To add a concept, you add an entry to the data source, then add related edges. In development mode, automatic validation flags issues to the console: any concept missing from a learning line? Any edge referencing a nonexistent concept?
This means as the Easy AI knowledge base grows, the concept map updates automatically—it won't become an outdated map.
Final Thoughts
Easy AI has always done something simple but hard: explaining complex AI concepts in a way ordinary people can understand. The concept map is the next step—not just explaining clearly, but showing where each concept belongs.
If you're learning AI, open the map, find your starting point, and follow one line. Knowledge isn't a well—it's a web. Knowing your position in the web matters more than memorizing more points.
Original post tags: easy-learn-ai, daily update, memory, Xiaokai.