Local AI vs cloud AI for studying: what happens to your notes

If you're weighing local AI vs cloud AI for studying, here's the short answer: a local model keeps your notes on your own laptop, but it needs a powerful machine and explains things less well; a cloud model explains better and runs anywhere, but what you give it leaves your computer and sits with a company. For most students, with most class material, the cloud is the practical choice. What matters is knowing what you upload and under which terms.

I say this with some experience behind it. For a good while I ran models on my own computer, first out of curiosity and later because I was building a study tool that worked that way. I loved the idea that nothing ever left my machine. And I ended up moving to the cloud for reasons that had little to do with privacy and a lot to do with the laptops people actually own.

This post is so you can decide for yourself, with the good and the bad on each side, without selling you either one.

What local AI and cloud AI actually mean

A language model is a huge file holding what it learned during training, plus a program that runs it. The difference is where that program runs.

• In the cloud. ChatGPT, Claude, Gemini and almost every AI study app. You type in a browser or an app, your message travels to the company's servers, the model works on it there and sends the answer back. Your computer just displays the text.
• Locally. You download the model and run it on your own machine, with programs like Ollama or LM Studio. Your message never leaves the laptop: your own processor is the one thinking.

The question isn't which one is better, it's what you care about most. If you care most about nothing leaving your computer, local wins. If you care most about a good explanation that also works on your phone, the cloud wins.

What actually leaves your computer

When you use a cloud AI, what you type and the files you upload reach the company's servers. That doesn't mean anyone reads them, but it does mean they depend on what that company does with them. Three things are worth checking in the settings and privacy policy of any service:

1. Whether it trains its models on your chats. Many consumer services have a switch in their privacy settings to stop that. Find it and decide.
2. How long it keeps what you upload, and whether you can delete it yourself.
3. Whether you can delete your whole account and take out or erase what's inside.

With local AI, none of those questions apply: the file never leaves your drive. That's the real advantage, and it shouldn't be played down.

Now think about what you're actually uploading. Your macroeconomics lecture slides, a textbook chapter or your notes on the IS-LM model aren't secrets: hundreds of students get the same material. Other kinds of material are different, and that's where the rule changes.

What I wouldn't upload to any cloud AI

Some things deserve a second thought before you paste them into any chat, however good the company's policy is:

• Other people's data. Interviews you ran for a sociology research project, a survey spreadsheet with names, your notes from an internship or a placement at a school. Those people agreed to take part in your project, not to have their answers pass through a third party's server.
• Information from a company where you work or intern. An internal balance sheet, a customer list. What you were given for the job isn't yours to upload.
• Whatever your university or your course prohibits. Some have rules about what you can do with AI and with which material, especially for theses and capstone projects. Read yours first.

For that kind of material, local AI makes a lot of sense, or no AI at all: anonymizing the data first (removing names and anything that identifies someone) is standard practice in research anyway.

What you need to run a model on your own laptop

This is the part the "free private AI" videos skip. A local model takes up memory while it runs, and the bigger it is, the better it reasons and the more memory it needs. Models are measured in billions of parameters: the smallest fit on almost any modern computer; the mid-sized ones already need plenty of memory; the large ones, the ones that feel like what you use in the cloud, don't fit on a student laptop.

In my own tests, a mid-sized model didn't run well on a MacBook Air with 8 GB: explanations took far too long and the whole machine slowed down. A smaller model ran fine, but you could tell the difference in quality. On a MacBook Pro with lots of memory, on the other hand, it worked really well.

Beyond the hardware, there are three hidden costs:

• Battery and heat. While the model is thinking, your processor runs flat out.
• Disk space. Each model takes up several gigabytes.
• Setup. Installing, picking a model, configuring it. It isn't hard, but it isn't opening a web page either.

Free local AI gets paid for by your laptop. If you have a powerful machine, it's a great option. If you study on whatever computer you've got, like almost everyone does, it'll probably frustrate you.

How well a small model explains things

For studying, the quality of the explanation matters more than almost anything else. A small model can summarize a paragraph or reword a definition just fine, but when you ask it to connect two concepts, walk through a supply and demand graph or tell you why an exam answer is wrong, it gets things wrong more often, and it gets things wrong with the same confidence it gets them right.

Big cloud models do this too, just less often. So the golden rule is the same with any AI: ask it to tell you which part of your material each point comes from, and check it, as I explain in how to use ChatGPT to study. And use it to understand and practice, not to think for you, which is the line I draw in whether using AI to study is cheating.

Why I moved to the cloud

The tool I built, Filmina, a web app that explains your lecture slides out loud, started out local: the model and the voice ran on the computer and nothing left it. It worked, but only well on powerful machines, and most students don't have one. That's why it moved to the cloud, and I told the whole story in Filmina in the cloud. Today it needs a connection and an account, and what you upload goes to its servers. What we do take care of: we don't train models on your material, and the terms of the providers we use don't allow them to either. That isn't the same as local, and I'd rather say so plainly.

My takeaway, after trying both roads: for class material, the cloud, with a service whose terms you know; for other people's data, local or nothing.

Frequently asked questions

What is local AI?

It's an AI model you download and run on your own computer, with programs like Ollama or LM Studio. What you type never leaves your machine, but quality and speed depend on your hardware.

Is it safe to upload my notes to ChatGPT?

Class material usually isn't sensitive, so the risk is low. Check in the settings whether your chats are used to train the model, and don't upload other people's data or confidential information.

What computer do I need to run AI locally?

A modern computer with reasonable memory is enough for small models. For models that explain at a level close to the cloud ones you need a lot of memory, which few student laptops have.

Does local AI work offline?

Yes. Once the model is downloaded, it works without a connection, because your computer does all the work. That's one of its real advantages over the cloud.