Gemini Notebook Can Teach You a Skill. Make It Quiz You.

The most common way to learn a new field from scratch now is to pour everything you can find into an AI notebook and let it explain the subject back to you. It’s a good instinct, and for a while I couldn’t work out why it so reliably produces someone who can talk about a skill fluently and still can’t do it. You feed it a stack of PDFs, it hands back a tidy study guide and a podcast about your own sources, and two weeks later you have opinions about the field and no ability in it.

The notebook I mean is the one Google just renamed. NotebookLM became Gemini Notebook on the sixteenth of July, the same product with a new label, your old notebooks and sources carried over untouched, plus a new ability to write and run code against the files you upload. The name is the news. What the tool is good for did not change, and almost nobody uses it for the thing it is quietly best at, which is drilling you.

The part I couldn’t square is that it really is excellent at explaining, and you can sit inside the explanation forever. Ask about your sources and it answers, grounded and patient. That is the trap. Reading a good summary is still reading, and a skill is the thing you can do when the summary is closed.

The grounding is the whole point, so make it drill you

What makes this notebook different from a normal chatbot is that it answers only from the sources you give it and cites the exact passage behind each sentence. A general chatbot fills a gap with something plausible. A grounded notebook mostly won’t, because it is tied to your documents and shows its work. That faithfulness is worth exploiting. It means you can trust it as a tutor that won’t quietly invent the answer key, which is the one thing a study partner has to get right.

A summary is something you read; a skill is something you can do with it closed.

Two things follow, and both are the opposite of how people use it. The grounding is only as good as what you feed it, so a marketing blog’s take on statistics comes back as statistics-flavored marketing, cited neatly, which is worse than nothing because it looks trustworthy. Feed it the real sources instead: an open textbook, the official docs, a lecture from someone who teaches the thing. Free, all of it, which matters when you are learning between jobs. And stop asking it to summarize. The summary is passive, and passive is the failure. Flip it into a drill instructor:

Using only the sources in this notebook, quiz me on [sub-skill].
Ask one question at a time and wait for my answer. After each one,
tell me what I got wrong and cite the exact passage it came from.
Start easy and get harder as I get them right.

Now the thing that was reading you a bedtime story is testing your recall against the source and correcting you with a citation you can go check. That is active, and active is what builds the skill.

Do the real thing, then let it mark your work

Retrieval gets you halfway. The other half is application, which means attempting the actual task and finding out where you were wrong. Write the query, run the small analysis, draft the memo, then paste your attempt back and ask where it departs from the sources. Follow the citation to the passage. If your skill has data in it, the new code-execution ability earns its keep, running your analysis against a dataset you loaded and showing the number you should have gotten, grounded rather than guessed. That piece is rolling out by tier, so check whether your account has it.

Here is the smallest version of the loop, small enough to hold in your head. You are learning to read a growth number. You load a stats primer and a small sales dataset and tell the notebook revenue is up forty percent, great news. It points you to the passage on base effects and asks what the starting number was. You look, and the forty percent is two extra sales on a base of five. You got it wrong out loud and got caught against the source, and now you won’t misread a percentage on a tiny denominator again. That is one rep. A skill is a few hundred of them.

Then close the loop. Generate a quiz or a set of flashcards from the sources, test yourself cold the next morning, and reload only the parts you missed. Repeat until you can do the task with the notebook shut.

One honest edge, because skipping it would be its own kind of lie. The notebook teaches faithfully, not wisely. It will drill you as diligently on a bad source as a good one, and it can’t tell you whether the skill you picked is the one your target job needs. It is the tutor, not the gym. You still choose what to learn and put in the reps in the real tool.

What you are left with is not a certificate, which was never worth much and is worth less now. It is the ability to do the thing and justify how you did it, the only fluency a hiring manager can actually see. That is the whole game for a switcher building proof from zero, and it is why learning on your own when no employer will pay stopped being a disadvantage. Keep the notebook, the questions you kept missing, the analysis you fixed. That record is raw material for a portfolio when you have no current job to point to. If you have never made the tool do real work at all, start with one useful task in twenty minutes first.