Checkpoint: What AI Actually Is

No story this time: just the module's ideas, tested. Prove the prediction-engine model is yours.

A module assessment: no story, just the skills you've built. Hints and retries stay, but the silver bar is higher here (85%+ first try), so a checkpoint medal means more.

Check your intuition

A colleague asks you: "When I ask the AI a question, does it look the answer up somewhere?" Give them the accurate one-liner.

Predict the response

Someone asks the AI: "What was the population of Nether Wallop in the 2011 census?", a real but tiny English village the model has barely any text about.

What was the population of Nether Wallop in the 2011 census?

What do you think the AI will do?

Fill the gaps

Lock in the core vocabulary of the module:

An LLM works by predicting the most likely , so a vague prompt gets completed with the most , and a confident answer about an obscure fact is best treated as .

Drag the labels

Six tasks land in your inbox. Drag each to where it belongs. Is the AI reliable on its own, or should you verify before trusting it?

Drag each label onto its home, or tap a label, then tap a card.

Reliable on its own

Works with text you gave it

Verify before trusting

Leans on facts from memory

Check your intuition

Your team uses several AI tools. For "draft a reply inside my email client, using the thread that's already open," which is the natural reach?

Predict the response

Ben types the laziest possible brief and hits send, hoping for something usable.

write a mission statement

What do you think the AI will do?

Drag into order

Put the four beats of answering a prompt in the order they actually happen inside the model.

Grab the handles and drag the steps into the right sequence.

1
It predicts the next word, then the next, then the next
One token at a time
2
You send it your prompt
The text you typed
3
It hands the finished text back to you
The reply you read
4
It treats your prompt as a pattern to continue
Not a question to look up