Your first steps into the world of AI assistants
What is an LLM?
The Simple Explanation
An LLM (Large Language Model) is a computer program that's really good at understanding and generating human language. Think of it as a super-powered autocomplete that's read most of the internet.
When you type a message to ChatGPT or Claude, here's what happens:
You type: "What's a good recipe for banana bread?"
↓
The LLM predicts the most helpful response
word by word, based on patterns it learned
↓
You get: A complete recipe with ingredients and stepsWhat Makes Them "Large"?
| Model | Parameters | Analogy |
|---|---|---|
| ------- | ----------- | --------- |
| Small | ~7 billion | A well-read college student |
| Medium | ~70 billion | An expert in many fields |
| Large | ~400+ billion | A team of specialists |
Parameters are like the "brain cells" of the model - more generally means smarter, but also more expensive to run.
What Can LLMs Do?
Great At
| Task | Example |
|---|---|
| ------ | --------- |
| Writing | Emails, essays, stories, social media posts |
| Explaining | "Explain quantum physics like I'm 10" |
| Summarizing | Turn a 20-page report into bullet points |
| Translating | Between languages, or from jargon to plain English |
| Brainstorming | Generate 10 ideas for a birthday party |
| Coding | Write, explain, or debug code |
| Research | Answer questions with explanations |
| Formatting | Convert messy notes into organized documents |
Not Great At
| Limitation | Why |
|---|---|
| ------------ | ----- |
| Current events | Training data has a cutoff date |
| Math | Can make arithmetic errors (seriously!) |
| Facts about obscure topics | May confidently make things up |
| Counting | "How many r's in strawberry?" often wrong |
| Personal memory | Doesn't remember past conversations* |
| Accessing the internet | Only knows training data (unless given tools) |
*Some apps add memory features, but the base models don't remember you.
The Golden Rule
Always verify important information. LLMs are assistants, not oracles.
Where to Try LLMs (Free Options)
No Account Needed
| Service | URL | Notes |
|---|---|---|
| --------- | ----- | ------- |
| DuckDuckGo AI | duckduckgo.com | Click "Chat" - completely anonymous |
| Perplexity | perplexity.ai | Good for research, cites sources |
Free with Account
| Service | URL | Free Tier |
|---|---|---|
| --------- | ----- | ----------- |
| ChatGPT | chat.openai.com | GPT-3.5 unlimited, GPT-4o limited |
| Claude | claude.ai | Generous free tier |
| Gemini | gemini.google.com | Free with Google account |
| Copilot | copilot.microsoft.com | Free with Microsoft account |
My Recommendation for Beginners
Start with Claude.ai - it's free, no tricks, and designed to be helpful and honest. Create an account and just start chatting.
Your First Conversation
Just Talk Naturally
You don't need special commands. Just type like you're talking to a helpful friend:
You: I'm planning a trip to Japan for 10 days in April.
I like food, history, and avoiding crowds.
Can you help me plan?
Claude: I'd love to help you plan your Japan trip! April is
beautiful - you might catch cherry blossom season...
[continues with detailed suggestions]Try These Starter Prompts
For everyday help:
- "Explain [topic] like I'm a complete beginner"
- "I need to write an email to [person] about [topic]. Help me sound professional but friendly"
- "Give me 5 dinner ideas using chicken and vegetables"
- "Summarize this article: [paste article]"
For learning:
- "I want to learn [skill]. Create a 30-day plan for a complete beginner"
- "What are the most important concepts to understand about [topic]?"
- "Quiz me on [topic] with 5 questions"
For creative projects:
- "Help me brainstorm names for my [business/pet/project]"
- "Write a short story about [topic] in the style of [author]"
- "I'm stuck on [creative problem]. Give me 10 unconventional ideas"
How to Get Better Responses
Be Specific
| Vague (worse) | Specific (better) |
|---|---|
| --------------- | ------------------- |
| "Write about dogs" | "Write a 200-word blog post about why golden retrievers make great family pets, aimed at first-time dog owners" |
| "Help with my resume" | "I'm a marketing manager with 5 years experience applying for senior roles. Review my resume and suggest improvements" |
| "Explain AI" | "Explain how AI image generators work to someone who's never coded" |
Give Context
Less effective:
"How do I fix this error?"
More effective:
"I'm building a website with React. When I click the submit
button, I get this error: [paste error]. Here's my code:
[paste code]. I'm a beginner and not sure where to start."Ask for the Format You Want
"Give me a bullet-point summary"
"Respond in a table format"
"Keep it under 100 words"
"Explain step by step"
"Give me pros and cons"Common Beginner Mistakes
Mistake 1: Trusting Everything
LLMs can be confidently wrong. They may:
- Invent citations that don't exist
- Give outdated information
- Make up statistics
Fix: Verify important facts. Ask "Are you sure about that?" or "What's your source?"
Mistake 2: Giving Up After One Try
First response not great? Don't give up!
Fix: Say things like:
- "That's not quite what I meant. I want..."
- "Can you make it more [casual/formal/detailed]?"
- "Give me a different approach"
Mistake 3: Being Too Brief
Bad: "Write me a story"
Good: "Write a 500-word mystery story set in a small
coastal town, with a twist ending. The main
character should be a retired detective."Mistake 4: Sharing Sensitive Information
Don't enter:
- Passwords or API keys
- Social security numbers
- Private medical information
- Confidential business data
Why: Your conversations may be used to improve the model.
Understanding "Temperature"
Some interfaces let you adjust "temperature" - this controls randomness:
| Temperature | Effect | Best For |
|---|---|---|
| ------------- | -------- | ---------- |
| Low (0-0.3) | Predictable, focused | Facts, code, math |
| Medium (0.5-0.7) | Balanced | General conversation |
| High (0.8-1.0) | Creative, varied | Brainstorming, fiction |
If you don't see this option, don't worry - defaults work fine.
How Much Does It Cost?
Free Tiers Are Generous
For casual use, you'll likely never pay:
- ChatGPT free: Unlimited GPT-3.5
- Claude free: ~100+ messages/day
- Gemini: Unlimited for most uses
Paid Tiers
| Service | Price | What You Get |
|---|---|---|
| --------- | ------- | -------------- |
| ChatGPT Plus | $20/month | GPT-4, faster, more features |
| Claude Pro | $20/month | More usage, priority access |
| Gemini Advanced | $20/month | Gemini Ultra, Google integration |
API Pricing (for developers)
If you build apps, you pay per token:
- GPT-4: ~$10-30 per million tokens
- Claude: ~$3-15 per million tokens
- Open source: Free (but you run the hardware)
A million tokens ≈ 750,000 words - that's a LOT of conversation.
Key Terms Glossary
| Term | Simple Definition |
|---|---|
| ------ | ------------------- |
| LLM | Large Language Model - the AI that powers chatbots |
| Token | A chunk of text (~4 characters). LLMs think in tokens |
| Prompt | What you type to the AI |
| Context window | How much the AI can "remember" in one conversation |
| Hallucination | When the AI makes something up |
| Fine-tuning | Training an AI on specific data for special tasks |
| Temperature | Controls how creative/random responses are |
| System prompt | Hidden instructions that set the AI's behavior |
What's Next?
Now that you understand the basics:
- Go try it! Open Claude.ai or ChatGPT and have a conversation
- Experiment with different types of requests
- Learn the deep details in our course based on Karpathy's video
Ready to understand how LLMs actually work under the hood? Continue to the Deep Dive section →
This guide is original content to help beginners. The Deep Dive modules are based on Andrej Karpathy's excellent video lecture.