Level 0 · Understand — the lesson
Understand (The model)
Before you climb, look at the machine: what a token is, where the memory lives, why it's fluent even when it's wrong — and every surface Claude offers, from chat to the API.
The line This level is about understanding, not doing. Nothing here touches your tools or data — that starts at Level 2. But every rung above leans on what's on this page.
Play the models below — each one you try climbs a floor of your tower.
Everything is tokens — here's the split
6 tokens · 4 words. One short sentence, six tokens. Your message, your files, the tool results and the reply are all counted this way — in and out, every turn.
The window is its whole memory — add turns and watch the edge
Each turn is re-read every time. Keep adding — the window only holds so much.
Watch it invent — then watch grounding fix it
When will the Jakarta site permits clear?
Permits like these typically clear in 4–6 weeks — yours should be approved around March 14.
⚠ Confident, specific — and invented. It can't see your filing, so it predicted a plausible answer instead of a true one.
Portal says: "Under review, est. 8–10 weeks, filed Jan 20." When will it clear?
Per the portal, 8–10 weeks from the Jan 20 filing puts it late March to early April. I can't tighten that without a newer status.
✓ Same model — now it reasons over facts you handed it, and flags what it still doesn't know.
Same model, more reach — tap a rung
The raw model, reasoning over what you paste.
Unlocks Draft, explain, or think a decision through.
Do it now — 5 small steps
One step = one paste into Claude (the mobile app is perfect). Your progress saves on this device.
-
Step 1 of 5 · See your words as tokens
Tokens are invisible until you look once — then pricing, speed and limits all make sense.
Explain what a token is, using this exact sentence as the worked example: "The quarterly forecast shifted." Show roughly how it splits into tokens, and why token count (not word count) is what drives cost and memory.
You should see: Your own sentence, split into the units Claude actually reads.
-
Step 2 of 5 · Feel the context window
The 'memory' is just the conversation — test it.
Without scrolling up (you can't): what were the first and most recent things I said in this conversation? Then explain what your context window is and what happens to a chat that outgrows it.
You should see: An honest recap — and the limits of it — straight from the machine.
-
Step 3 of 5 · Find the cutoff
Knowing where knowledge ends tells you when to demand search.
What is your knowledge cutoff? Give one kind of question I should never trust you on without web search, and one where the cutoff barely matters. Explain the difference.
You should see: The boundary between knowledge and guesswork, stated plainly.
-
Step 4 of 5 · Catch a hallucination on purpose
Seeing the failure mode once, safely, builds the verify reflex for good.
Without using web search: give me one precise-sounding statistic about my industry. Then critique your own answer — which parts might be fabricated, and how exactly would I check them outside this chat?
You should see: Fluency and accuracy pulled apart in front of you.
-
Step 5 of 5 · Map your loadout
Every chat has a different reach — know yours before you climb.
List the tools and abilities you actually have in this chat right now — files, web search, connectors, artifacts, anything else. For each one: a single thing it makes possible that a plain chat can't do.
You should see: Your real capability surface, not the theoretical one.
The ideas behind the steps — tap a takeaway to unpack it.
How it actually works
Takeaway It's next-token prediction — always fluent, not always right.
Claude predicts the next token, over and over. That one trick at scale produces reasoning, code and judgment — and the failure mode above: it always writes a plausible continuation. Fluency is built in; accuracy you check.
Its memory is the window
Takeaway Claude knows exactly what's in the window — nothing more, and only for this chat.
Everything is tokens (see the split above) and they live in a finite window re-read every turn. Fill it and the oldest context falls out — long chats drift. No memory between chats by default. So: one task per chat, key facts first.
The instruction stack
Takeaway Write the durable instructions once; spend your message on the task.
Instructions layer, top wins:
| Layer | Set by |
|---|---|
| System prompt | Anthropic / the app |
Standing (a Project, an AGENTS.md) |
you, once |
| Your message | you, each turn |
The standing layer is why you don't re-explain a project every chat — and the whole mechanism behind AGENTS.md at Level 4.
One model, more reach
Takeaway Same model, more reach at each rung — that's the whole climb.
The ladder above isn't smarter models — it's the same mind with more reach at each rung, from plain chat to the raw API. One more dial worth knowing: pick the model tier + thinking budget to match the stakes of the task, not out of habit.
Is this the right level for your task?
You've seen the machine. What do you want to get better at first?
Pick the closest match — you'll get a verdict.
The capstone — the whole level in one prompt
One prompt that turns Claude into your foundations tutor — it teaches each concept against your own usage, then quizzes you.
You're my tutor on how Claude actually works. Teach me, one concept at a time, checking my understanding before moving on: 1. Tokens — what they are, why they're the unit of cost and memory. 2. The context window — what you hold, what falls out, why long chats drift, and why there's no memory between chats by default. 3. Training and the knowledge cutoff — and why ungrounded answers can be confidently wrong (hallucination). 4. The instruction stack — system prompt vs standing instructions (Projects, AGENTS.md) vs my message. 5. The capability surface — chat, search, files, artifacts, projects, connectors (MCP), skills/slash commands, Cowork, Claude Code, and the API — one line each on what it adds. 6. Model tiers and thinking budgets — how to match the model to the task. Use concrete examples from how I'm talking to you right now. After each concept, ask me one quick question to confirm I've got it before continuing.
Self-check
✓ You've got this level. On to Level 1: Ask →
You've leveled up when…
You're ready when the words tokens, window, cutoff and grounding feel like tools, not jargon — and when a confident answer makes you ask "grounded in what?" before you act on it.
Next: Ask (Level 1) — put the machine to work in a plain chat, and build the verify reflex where mistakes are cheap.
Next · Level 1: Ask →