Singularity.Kiwi

AI progress, with honesty.
🎓 Free Interactive Learning

Ai-Edu

Not a textbook. An actual training arena.

6 Challenges
~10 Minutes
0 Badges
0 / 6
1 🎭 Give the AI a Role ~2 min Start →
🎯 What you'll learn: When you give an AI a role, it changes everything — tone, depth, and relevance.

Most people start prompts like this:

"Explain how neural networks work."

That's fine. But watch what happens with a role:

"You are a high school science teacher. Explain how neural networks work to a 15-year-old."

The result is more accessible, more structured, and more useful — because the AI knows who it's being.

💡 Why it works: AI models were trained on text from people in many roles. When you specify one, it matches the tone and depth of that persona.

Your turn: Rewrite this prompt by adding a role. Who should the AI be?

"Give me 5 tips for learning to code."
2 📏 Add Constraints ~2 min Start →
🎯 What you'll learn: Vague prompts = vague answers. Constraints shape the response into exactly what you need.

Without constraints, AI defaults to its most generic voice — long, bland paragraphs.

"Write about climate change."

Add constraints and you get something actually useful:

"Explain climate change in 3 paragraphs for a general audience. Use only IPCC facts. No jargon. End with one action someone can take."

Constraints are like guardrails — they keep the AI on track.

💡 Common constraints: length limits ("under 200 words"), audience ("for a 10-year-old"), tone ("formal"), exclusions ("no jargon"), structure ("5 bullet points").

Your turn: Add at least two clear constraints to this vague prompt:

"Tell me about prompt engineering."
3 🧠 Think Step by Step ~2 min Start →
🎯 What you'll learn: Adding "think step by step" dramatically improves AI reasoning on logic, math, and analysis tasks.

Straight questions sometimes get wrong answers, especially on logic:

"How many R's are in the word 'strawberry'?"

The AI might say "2" (it sees the first R at position 2 and stops). But with chain-of-thought:

"How many R's are in the word 'strawberry'? Let's think step by step."

Now the AI works through: S-T-R-A-W-B-E-R-R-Y → positions 3, 8, 9 → answer is 3.

💡 The trick: Chain-of-thought forces the AI to "show its working" before answering. It catches its own mistakes along the way.

Your turn: Apply chain-of-thought to this classic trick question:

"A bat and ball cost $1.10. The bat costs $1.00 more than the ball. How much is the ball?"
4 📋 Control the Format ~2 min Start →
🎯 What you'll learn: By default AI outputs paragraphs. Specifying a format gives you exactly the structure you want.

Same question, no format specified:

"Compare Python and JavaScript."

You'll get paragraphs. Hard to scan. Now add format:

"Compare Python and JavaScript in a table with columns: Feature, Python, JavaScript. Include 5 rows."

Instant, scannable, useful. You skip the fluff and get what you asked for.

💡 Format ideas: table, bullet points, numbered steps, JSON, code example, FAQ, one-line summary, checklist.

Your turn: Rewrite this to request a specific format (try a table or comparison chart):

"Compare three AI image generators: DALL-E, Midjourney, and Stable Diffusion."
5 ⚠️ Avoid Hallucinations ~2 min Start →
🎯 What you'll learn: AI confidently makes things up. Defensive prompting prevents this.

This prompt is dangerous because it assumes a fact exists:

"What year was the AI Safety Accord signed, and which countries signed it?"

The AI may invent a year, countries, and details — all sounding perfectly confident but completely wrong.

Defensive version:

"What year was the AI Safety Accord signed, and which countries signed it? If you're not sure about any detail, say 'I don't know' instead of guessing."

Now the AI can admit uncertainty instead of fabricating.

💡 The rule: If the answer matters, add a guard. "Say 'I don't know' if unsure" is the single most important prompt hack.

Your turn: Make this prompt hallucination-proof. Add at least one safeguard:

"What was New Zealand's exact GDP growth rate in Q3 2023?"
6 ⭐ The Master Prompt ~3 min Start →
🎯 What you'll learn: Combine everything — role + task + context + format + constraints + hallucination guard — into one killer prompt.

A master prompt has all the ingredients. Here's what a complete one looks like:

"You are a patient tutor (role). A student who's never coded before wants to understand variables (context). Explain what a variable is in Python with a real-world analogy (task). Use 2 short paragraphs and include a code example (format + constraint). If you're not 100% sure about a detail, say so (guard)."

Every element adds clarity. Miss one, and the AI fills the gap with guesswork.

Your turn: Write a master prompt about machine learning for a beginner. Include all elements:

  • 🎭 Role — who is the AI?
  • 📖 Context — who is the audience?
  • 🎯 Task — what to explain?
  • 📋 Format — how should the output look?
  • 🔒 Constraints — rules for the answer
  • 🛡️ Guard — hallucination protection
🔬

Prompt Builder Lab

Build a structured prompt piece by piece. Watch your quality score grow.
Prompt Quality 0 / 8
🎭 Role 📖 Context 🎯 Task 📋 Format 📋 Format detail 🔒 1+ constraint 🔒 2+ constraints 🛡️ Guard

📝 Your Final Prompt

Fill in the fields on the left to build your prompt...

Need a starter? Pick a template:

📚

Learn the Basics

Quick reads you can chew through in a couple of minutes.
🧠

AI Doesn't Know Anything

▼

AI models predict text based on patterns, not knowledge. They can sound confident while being wrong — this is called hallucination, and it happens more often than most people think.

💡 Key Insight: Treat AI outputs as suggestions, not answers. Always verify claims that matter.
🎯

Anatomy of a Great Prompt

▼

Role + Context + Task + Format + Constraints = better results. Every element adds clarity. Miss one, and the AI fills the gap with guesswork.

💡 Example: "You are a teacher. Explain X to a 10-year-old in 3 paragraphs with one analogy."
⚡

Prompt Patterns That Work

▼

Chain of thought, few-shot examples, reflexion, role-play — patterns that consistently improve output. The key insight: AI performs better when you show it what good looks like.

💡 Pro tip: Add "Let's think step by step" to any reasoning question.
🌐

Beyond Text

▼

Voice, vision, code, and images — each mode has strengths and blind spots. Screenshot errors for debugging, upload images for visual tasks, use voice for brainstorming.

💡 Pro tip: Paste error messages directly — AI is exceptional at debugging code.
🛡️

Ethics & Safety

▼

Privacy, accuracy, bias, security — working with AI responsibly. Never share personal data with AI. Always double-check important outputs. You're responsible for what you publish.

💡 Remember: You're responsible for outputs you use, not the AI.
🛠️

Build & Resources

Set up your own AI system. Find tools and models.
🏠

Build Your AI System

Offline for privacy, online for power
1

Install Ollama

  1. Download from ollama.com
  2. Open .dmg, drag to Applications
Download
2

Download Model

ollama pull qwen3-coder

Recommended: qwen3-coder (5GB) or deepseek-r1:14b (9GB)

3

Start Chatting

ollama run qwen3-coder

Or use LM Studio for a GUI

1

Install Ollama

  1. Download from ollama.com
  2. Run the .exe installer
Download
2

Download Model

ollama pull qwen3-coder

3

Start Chatting

ollama run qwen3-coder

Or use LM Studio for a GUI

1

Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

Docs
2

Download Model

ollama pull qwen3-coder

3

Start Chatting

ollama run qwen3-coder

Access powerful models through your browser:

ChatGPT

GPT-4o free tier

Free

Claude

Long docs, thoughtful

Free

Gemini

Google integrated

Free

Grok

Real-time X data

Premium
⚠️ Check IT policy first. Some organisations prohibit unauthorised software. If approved, follow the Offline instructions.

AI apps for phone or tablet:

📱 Claude

iOS · Android

iOS Android

💬 ChatGPT

iOS · Android

iOS Android

🔮 Perplexity

iOS · Android

iOS Android

✨ Gemini

iOS · Android

iOS Android

📶 Requires internet. Use WiFi for large uploads.

📚

Resources

Tools, models & guides
🦙 Ollama Run models locally 🎨 LM Studio GUI for models ✨ OpenClaw AI assistant 🤗 Hugging Face Model hub

Quick start: ollama pull qwen3-coder then ollama run qwen3-coder

View Full Resources →
⚡

Automate Your System

OpenClaw desktop AI assistant

OpenClaw is a full AI assistant that runs on your computer. It can read files, browse the web, run commands, and automate tasks — all with local models for complete privacy.

Local models File access Web browsing Automation
⚠️ Important: OpenClaw has system access. It can read files, run commands, and make changes. Only install if you understand these capabilities.
🍎 macOS 🪟 Windows 🐧 Linux
Full Setup Guide →
📰

Latest Ai-Edu Articles

Deeper dives on AI education topics.