Concept Lesson No AI connection required

What is AI? Your Very First AI User Manual

Before diving in, we need the right "map." No code in this lesson—we're just building a clear, accurate mental model of AI to keep you on track and avoid detours as you learn.

The One-Sentence Definition

AI = Programs that learn patterns from massive amounts of data, letting machines do things that used to require human intelligence.
Click the blue words below to have your AI TA explain them → Large Language Model, Neural Network, Token

AI is just a math function

It isn't magic, and it isn't conscious. It takes an input (your text), runs billions of math operations, and produces an output (the answer). Think of it as an incredibly complex calculator—fixed input, fixed calculation, fixed output.

How does AI "think" of an answer? — A game of probability autocomplete

AI isn't "thinking," it's calculatingwhat the most likely next word is. Pick a sentence starter below to see the probability distribution of the next words through the model's eyes 👇

The model only predictsone word at a time. Once picked, it adds it to the end of the sentence and predicts the next one, chaining them together. A 1,000-word article is generated exactly this way, one word at a time. The essence of "Generative AI" is just this continuous game of guessing the next word.

A Brief History of AI (6 milestones you should know)

1950
The Turing Test
Alan Turing asked, "Can machines think?" — If you can't tell whether you're chatting with a human or a machine, the machine passes the test.
1956
The Birth of "Artificial Intelligence"
The Dartmouth Workshop coined the term "Artificial Intelligence." Researchers believed they could build a machine that thinks like a human within 10 years.
1980s
Expert Systems
Hardcoding human experts' rules into software—if a patient has symptoms A+B, diagnose them with disease C. Useful, but brittle; slight rule changes caused the system to break.
2012
The Deep Learning Boom
AlexNet crushed the competition in an image recognition contest. Neural networks + big data + GPUs sparked the modern era of AI.
2017
The Transformer Architecture
Google's paper "Attention Is All You Need." It's the foundation for all of today's Large Language Model , including GPT, Claude, and Gemma.
2022
The LLM Explosion
ChatGPT launched, hitting 100 million users in 2 months. AI shifted from an expert tool to a tool for everyone. DeepSeek, Gemini, and Claude emerged back-to-back, with capabilities iterating rapidly.
2025–2026
The Era of Multi-Agent Collaboration (Present)
AI isn't just answering questions anymore; it's proactively completing tasks. One person directing multiple AIs working together, is like having an on-call team—you are a one-person company.

AI ≠ Search Engine

Many people using AI for the first time think: "I'll just ask it a question, like a search engine." But they're actually very different.

Search Engines (Google/Bing)

  • Finds existing webpages, doesn't generate new content
  • Matches keywords, returns a list of links
  • Answers are fixed, written by others
  • Can't understand context
  • Can't reason or analyze

Large Language Models (ChatGPT/Claude)

  • Generates answers in real-time; each time might be different
  • Understands meaning, doesn't just match words
  • Can reason, analyze, and create
  • Remembers conversation context (to an extent)
  • Can output in your specific format

What is AI good at?

Before you start exploring, parents and kids can get to know AI's "superpowers." It's not magic, and it's not just a calculator. Once we figure out its quirks, we can have a lot more fun with it.

Read text → Think → Write text
Today's mainstream AI (ChatGPT / Claude / Gemini / DeepSeek) is basically this loop. Everything it can do stems from these three steps.

🎨 The magic AI is great at

  • Wildly imaginative creation: Draw a cat wearing sunglasses, or make up a fairy tale about a Mars expedition.
  • Tireless translation and summarization: Summarize a thick storybook into three sentences, or translate a Spanish cartoon into English.
  • Providing inspiration and advice: Don't know where to go this weekend? Stuck on a gift for a friend? Have it list 10 ideas for you.
  • Tireless explaining: Like a knowledgeable tutor, patiently answering all the "whys" in your child's head.

🔎 Where you need to be the gatekeeper

  • 100% accurate facts: It might misremember obscure historical dates or scientific details.
  • Complex logic and math problems: It's a genius at language, but with big math problems, it can sometimes be as careless as a kid just learning to count.
  • Making the final call: AI can give you lots of great advice, but the smartest, most compassionate decisions always need to be made by you and your parents.

💡 Pro Tip: AI is getting more capable every day—it can browse the web, write code, and even edit video. We'll unlock these "advanced gears" in later lessons.

💡 Why gatekeep it yourself? Understanding AI "Hallucinations": AI is like an overly eager buddy who always wants to please you. When you ask it something it doesn't know, it doesn't want to let you down, so it might confidently guess the answer—this is called a **Hallucination**. So, when it comes to important facts, remember to put on your detective hat and double-check if the AI got it right!

Core Concepts (Click to expand)

You'll see these terms pop up repeatedly in the upcoming lessons. Don't worry about memorizing them all right now—just get familiar with them.

How much does AI cost? — Understanding tokens

The AI industry has a universal unit of measurement called token(About 1 Chinese character = 1-2 tokens, 1 English word ≈ 1 token).The words you send to the AI + the words it sends back, are added together and billed by the token.

Model Approximate Price (per million tokens) Features
DeepSeek-V3 ~¥2 Super affordable and capable, great for everyday use
GLM-4.5-air (Zhipu) ~¥3 Strong in Chinese, fast speeds in China
GPT-4o-mini (OpenAI) ~¥4 ChatGPT's budget-friendly version
GPT-4o (OpenAI) ~¥18 Top tier for overall capabilities
Claude Sonnet 4 (Anthropic) ~¥22 Best for coding / long document processing
Local Gemma / Qwen (running on your PC) ¥0 (just the cost of electricity) Speed is limited by your graphics card, slightly less capable
For example:You ask GPT-4o a question: your 200-word input (~300 tokens) + the AI's 400-word response (~500 tokens) = ~800 tokens → roughly ¥0.014.
In other words,the price of a $4 coffee pays for 1,500 chats with GPT-4o; or 10,000 chats with DeepSeek; and if you run a local model, it's unlimited.
Prices fluctuate with the market, so check their official websites. Expensive models are usually smarter but not always necessary—start with the cheap ones and see if they get the job done.

The AI Capability Pyramid: Which level can you reach?

From basic chatting to orchestrating multiple AIs working together—this course will guide you up layer by layer.

We're living in an era of acceleration

The Industrial Revolution took nearly 200 years to move humanity from manual labor to machines. The AI revolution is moving at probably ten times that speed.

Here's a spot-on analogy: learning AI isn't about learning how to build an engine, it's learning how to drive. You don't need to understand how the engine works under the hood, but you do need to know how to hit the gas, navigate, and when to hit the brakes.The people who master the steering wheel are the ones who will truly benefit.

This course is your driver's ed. Lesson one is about getting a feel for the car—how powerful it is, what to watch out for, and how to communicate so it actually understands you.

Lesson Summary

Lesson Summary:
You've pulled back the curtain on AI and stopped seeing it as some omnipotent "black magic." You now see its true nature—a powerful, probability-based word-guessing game. You also know it can "confidently spout nonsense," which means you've learned how to collaborate with a critical eye.

I get it now: AI isn't a god, nor is it just a machine. It's a master of probability-based "word association." I have to understand its boundaries and quirks to really make it work for me.

Deep Dive Path

  • Ask the same question across different AIs (Kimi, ChatGPT, Doubao) → Notice the different language styles of each "brain"
  • Ask the AI about an obscure detail (like the exact source of a quote in a book) → See how confidently it answers, and practice your fact-checking habits
  • Have the AI summarize a long article, then check its accuracy → Practice evaluating its output quality and giving it prompts to improve