CourseModule 2
Module 2 of 6

Essential AI Vocabulary Made Simple

1 hour
4 sections

Learning Objectives

  • Understand prompts, tokens, and context windows
  • Recognize hallucinations
  • Know what models and training data mean
Section 1

Prompts - Your Instructions to AI

A prompt is simply what you type to an AI. It's your input, your question, your request.

The quality of your prompt directly affects the quality of AI's response.

Example Comparison

Vague Prompt:

"Write about dogs"

Specific Prompt:

"Write a 200-word blog post about golden retrievers for first-time dog owners, focusing on temperament and exercise needs"

The specific prompt will get a much better response because the AI knows exactly what you want.

Key Insight

Think of prompting like giving instructions to a helpful but literal-minded assistant. The more specific and clear you are, the better results you'll get.

Section 2

Tokens - How AI Reads Text

AI doesn't read text word-by-word like we do. It breaks text into tokens - chunks that are roughly ¾ of a word on average.

Examples

TextApproximate Tokens
"Hello"1 token
"artificial intelligence"2-3 tokens
"ChatGPT"2 tokens
1 page of text~300-400 tokens

Why This Matters

  1. Pricing: AI services often charge per token
  2. Limits: There's a maximum number of tokens AI can process at once
  3. Long words: Unusual words get split into multiple tokens, which can affect quality

Quick Estimate

A rough rule: 1,000 tokens ≈ 750 words

Section 3

Context Windows - AI's Working Memory

The context window is how much text an AI can "remember" at once during a conversation. Think of it like a desk that can only hold so many papers.

Context Window Sizes (2024)

ModelContext Window
GPT-4o~128K tokens (~96K words)
Claude 3.5~200K tokens (~150K words)
Gemini 2.51M tokens (~750K words)

What This Means

  • Long conversations: AI might "forget" things from early in the conversation
  • Large documents: Bigger context windows let AI work with longer texts
  • Be specific: If something important was mentioned earlier, remind the AI

Analogy

Imagine talking to someone who can only remember the last few pages of your conversation. That's roughly how context windows work - the AI processes everything in its window but has no memory beyond it.

Exercises

  • 1Have a long conversation with ChatGPT. After many messages, ask it what you discussed at the very beginning.
  • 2Try uploading a long document to Claude and asking questions about it.
Section 4

Hallucinations - When AI Makes Things Up

Hallucinations occur when AI generates information that sounds plausible but is completely made up.

This happens because AI predicts likely text, not true text. It doesn't have a database of facts - it predicts what words should come next based on patterns.

Famous Examples

  • ChatGPT inventing court cases that don't exist
  • AI generating fake research papers with realistic-sounding titles
  • Creating biographical details about real people that never happened

How to Spot Hallucinations

Watch for:

  • Very specific details (dates, names, statistics) that aren't verifiable
  • Information that seems too perfect or too specific
  • Claims about recent events (AI's knowledge has a cutoff date)

Protection Strategies

  1. Verify important facts with authoritative sources
  2. Ask for sources (but verify those too!)
  3. Use AI for research but cross-reference
  4. Be especially careful with medical, legal, and financial information

The Golden Rule: Trust, but verify.

Exercises

  • 1Ask AI about a specific fact from your professional field. Verify if it's correct.
  • 2Ask ChatGPT to cite its sources for a claim. Try to find those sources online.