Day 1 — What is AI? What is Claude?

Time: ~45 min · Date: Sat Apr 25

Why this matters

You're going to spend a lot of time with Claude. Knowing what Claude actually is — and what it isn't — saves you from being either over-trusting (it'll do anything!) or over-cautious (it'll steal my job!). Both extremes are wrong. Reality is more useful.

The 60-second mental model

Claude is a language model. You give it words; it gives you words back. Under the hood, it's predicting the most likely next chunk of text based on a huge amount of training data and the conversation you're having.

That's it. No magic.

But "predicting the next chunk of text" is shockingly powerful when:

• The training data was the internet, books, documentation, and code.
• The model is large enough to have learned patterns across all of that.
• You give it good context.

What it can do:

• Write, summarize, edit, translate, explain
• Analyze documents, reports, contracts
• Draft emails, descriptions, social posts
• Answer factual questions (with caveats)
• Walk through reasoning step by step
• Read and write code (this is where Claude shines)

What it can't do:

• Truly know whether something is true (it's confident even when wrong)
• Look up real-time data unless given a tool to do so
• Remember you between sessions (unless you give it memory)
• Replace human judgment on anything that matters

The Claude products

This trips people up. There are several things called "Claude."

You'll use the Claude.ai chatbot today and Claude Code starting May 6.

A few important truths

1. Claude can be wrong. Confidently. Always sanity-check anything that affects a real client or transaction.
2. What you tell it stays in that conversation. Claude doesn't share your messages with other users. Anthropic (the company) does train on some conversations unless you opt out — settings in claude.ai → Privacy.
3. Claude is not Google. It's not searching the internet by default. It's drawing on training data that has a cutoff date. For "what's happening right now," use search.
4. Better input → better output. Vague question, vague answer. Specific question with context, useful answer. (We'll cover this tomorrow.)

Exercise (~25 min)

Go to claude.ai. Sign up if you haven't. Free tier is fine.

Ask Claude these three things:

1. "What questions should I ask a seller in a listing presentation?"
2. "Rewrite this email to be warmer: 'Per our conversation, attached is the listing agreement. Please sign and return at your earliest convenience.'"
3. "I'm buying my first investment property. What's the difference between cap rate and cash-on-cash return?"

For each one, notice:

• Did the answer feel useful or generic?
• Did anything in the answer seem wrong or worth verifying?
• What would you change about the question to get a better answer?

Don't grade Claude. Grade your prompts. That's the muscle.

Recap

You should now be able to answer:

• What's the difference between Claude.ai and Claude Code?
• What's one thing Claude is great at?
• What's one thing Claude can't do?
• Why does the way you phrase a question matter?

Going deeper (optional, ~50 min)

If today's lesson clicked and you want more — here's a curated hour. Skip any of it.

Watch (~25 min)

• Search YouTube: "3Blue1Brown but what is a GPT" — beautiful visual explanation of how language models actually work, no math required.
• Search YouTube: "Anthropic Claude tutorial" — the company's own intro videos.

Read (~25 min)

• Anthropic homepage — see how the company positions Claude.
• Anthropic docs — skim "Welcome" and "Models overview." You don't need to understand all of it. You'll come back later.
• Any article from anthropic.com/news — see what they ship.

Done?

Tick this:

• Day 1 complete

Next: Day 2 — Talking to AI well