🍳Analogy: Think of it like cooking. AI is all of cooking. ML is a specific technique (like stir-frying). Deep Learning is a sophisticated version (like mastering wok hei). Generative AI is a chef that invents entirely new recipes.
Part 2 — Under the Hood
How Generative AI Actually Works
It's not magic — it's pattern recognition at massive scale. Here's the 3-minute version.
📖
1. Training
AI reads billions of pages — books, websites, code, conversations. It learns patterns in language: what words tend to follow other words.
🧩
2. Tokenization
Text gets broken into small pieces called "tokens" — not whole words, but chunks. "Understanding" might become "Under" + "stand" + "ing".
🎯
3. Prediction
Given a sequence of tokens, AI predicts the most likely next token. It's the world's most sophisticated autocomplete — one word at a time.
⚡
4. Generation
Repeat step 3 thousands of times. Each predicted token gets added, then AI predicts the next one. That's how it writes paragraphs, code, and poems.
💡Key insight: AI doesn't "understand" like humans do. It recognizes patterns incredibly well. That's why it sounds confident but sometimes makes things up — it's predicting what sounds right, not checking facts.
🤔Hallucinations explained: When AI generates text that sounds plausible but is factually wrong, that's a "hallucination." It's not lying — it's predicting tokens that pattern-match well but don't correspond to reality. Always verify numbers, dates, and specific claims.
Part 3 — How AI Reads
Tokenizer Playground
AI doesn't read words — it reads tokens. Type anything below and see how AI breaks it apart.
Type or paste text below
🟢
Common Words
Frequent words like "the", "is", "and" are usually a single token — cheap and fast.
🟣
Subword Pieces
Less common words get split: "understanding" → "under" + "standing". More tokens = more cost.
🔵
Characters
Rare words or other languages may split into individual characters — the most expensive per word.
🟡
Special Tokens
Punctuation, numbers, and special characters get their own tokens. Spaces are often merged with the following word.
💰Cost insight: AI pricing is per-token. The same message in Thai (สวัสดีครับ) uses ~3x more tokens than English ("Hello") — meaning multilingual content costs more. This matters for ASEAN businesses operating across languages.
Part 4 — The Creativity Dial
Temperature: Control AI's Creativity
Temperature controls how "creative" vs "predictable" AI's responses are. Drag the slider to see the difference.
Prompt: "Write a one-line summary of today's meeting about Q3 budget"
0.0
T=0.0 — Factual & Predictable
The Q3 budget meeting covered revenue targets of $2.4M, a 12% increase in marketing spend, and a hiring freeze in engineering until August.
🧊
Low (0.0–0.3)
Use for: Compliance reports, data extraction, factual summaries, anything where accuracy matters most.
⚖️
Medium (0.5–0.7)
Use for: Email drafting, report writing, general Q&A — balanced between creative and reliable.
🔥
High (0.8–1.0)
Use for: Brainstorming, creative writing, generating alternatives — but verify everything. Not for production.
👔Analogy: Temperature is like asking a colleague for advice. Low temperature = your cautious compliance officer (safe, predictable). Medium = your reliable project manager (balanced). High = your creative strategist (exciting ideas, but double-check the facts).
Part 5 — What AI Can Actually Do
AI Use Cases — For Every Role
AI isn't just for engineers. Here's what it can do for you today — and what it can't.
📋Summarize
Condense a 30-page report into bullet points. Turn a 1-hour meeting recording into a 5-line summary.
All Roles
✍️Draft
Write first-draft emails, proposals, SOPs, status reports. You edit and refine — AI handles the blank page.
All Roles
🗂️Classify
Sort documents by type (invoice, receipt, contract). Route support tickets to the right team automatically.
Operations · Finance
🔍Extract
Pull vendor names, amounts, dates, and line items from unstructured documents like invoices and contracts.
Finance · Compliance
🌏Translate
Convert content across ASEAN languages — English, Bahasa, Thai, Vietnamese, Filipino — while preserving tone.
Regional Teams
💻Code
Generate scripts, automate repetitive tasks, build formulas, debug errors. Turn plain-English instructions into working code.
Tech Teams
📊Analyze
Identify patterns, anomalies, and trends in data. Flag unusual transactions. Generate insights from spreadsheets.
Analytics · Risk
What AI Can't Do (Yet)
Guarantee factual accuracy — Always verify numbers, dates, and specific claims. AI predicts what sounds right, not what IS right.
Replace human judgment — Complex decisions (hiring, strategy, risk appetite) still need human context, ethics, and accountability.
Access your systems automatically — AI doesn't know your internal data unless you explicitly give it. It needs integration.
Learn from your conversations — Each chat session starts fresh. AI isn't getting smarter from talking to you (unless you set up fine-tuning).
Part 6 — AI on AWS
The Services That Matter
You don't need to build these. You just need to know they exist — so you can ask the right questions.
🧠
Amazon Bedrock
Access the best AI models (Claude, Titan, Llama) through a single API. Build AI-powered workflows without managing infrastructure.
For: Teams building AI workflows
💬
Amazon Q Business
An AI assistant that answers questions from your company's documents, wikis, and data. Like a smart internal search engine.
For: All roles
👨💻
Amazon Q Developer
AI coding assistant that generates code, finds bugs, and explains complex systems. Works directly in your IDE.
For: Developers
🎉
PartyRock
Build AI-powered apps with zero code and zero AWS account needed. Free playground to experiment with AI — try it today.
For: Anyone curious
🎯Key message: You don't need to be a developer to benefit from AI on AWS. Amazon Q Business is for everyone — ask it questions about your company's policies, procedures, and data, and it answers in plain language.
Part 7 — A Taste of Prompting
3 Tips to Talk to AI Better
Most people talk to AI like a search engine — and get bad results. Here's how to get great results instead.
🔑The fix: Don't search — brief. Talk to AI like you're briefing a smart colleague who's never worked at your company before.
Be Specific
Tell AI exactly what you want — format, length, audience, and purpose. ❌ "Write about Q3 results" ✅ "Write a 3-bullet executive summary of Q3 results for our board deck. Keep it under 100 words. Focus on revenue growth and market expansion in SEA."
Give Context
Tell AI who you are, what this is for, and any constraints. ❌ "Summarize this document" ✅ "I'm a finance manager preparing for a leadership review. Summarize this 20-page audit report into 5 key findings. Flag anything that needs immediate action."
Iterate, Don't Restart
AI gets better in conversation. Ask for a draft, then refine. ✅ "Make it shorter" → "Add bullet points" → "Make the tone more formal" → "Add a call-to-action at the end"
🎬Coming in Session 2: We go deep on prompt engineering — structured techniques, Chain-of-Thought, personas, and a live prompt competition with AI Judge scoring. Today is just the appetizer.
Part 8 — Your Action Items
Next Steps & Resources
Try one AI task this week. Then dive deeper with these free courses on AWS Skill Builder.
📝
This Week
Try one AI task at work — summarize meeting notes, draft an email, or ask AI to explain something complex in simple terms.
📚
Self-Paced
Start a Skill Builder course below. The 10-minute courses are perfect for a coffee break. The learning plans guide you step by step.
🎯
Session 2
Join us next time for "Talk to AI: Live Prompt Engineering Workshop" — hands-on techniques that make AI actually useful for your role.