Useful AI features in your studies

In this guide, I'll share my top tips on the most useful features found in most AI services, things like Learning Mode, Document Management, and Deep Research. NOTE! Some of the AI features I recommend in this resource may work differently, be limited, or be missing altogether depending on the AI service you use.

Learning Mode

The learning mode is one of the best tools out there. It enables what university is actually about: learning and being prepared.

One behavior that many people find annoying is that AI assistants usually just spit out an answer or start a task, even when it's not really what you're looking for. Learning mode solves that. It sets the assistant to discuss with you according to the Socratic method: asking questions, holding dialogue, finding out what you can do, and moving forward from there.

Do you know what you need to learn? Use the learning mode.
It finds out where you are and adapts the help to your level.

Don't know what you need to learn? Use the learning mode.
It discusses with you, finds out what is a good way forward and adapts the help to your level.

Regardless, learning mode is often the perfect way to get help from AI when you need to learn something.

💡 The learning mode is called different things in different AI assistants, but you'll almost always find it in the same place: click on the plus symbol or the corresponding icon next to the message box. There, choose how the AI assistant should respond. Sometimes it can be found under "More" or similar.

Text editing mode

The text editing mode solves one of the most annoying problems with AI assistants. If you ask it to make a small change to a text, it will often rewrite the entire paragraph and with that come changes you did not ask for.

When you enable it, you get access to a built-in text editor (think Word or Google Docs) that you can also edit completely manually, or let the AI assistant update a specific part of the text. It enables what is otherwise one of the most difficult problems with AI in studying and writing, to get the text exactly the way you want it.

Where can I find the Copilot feature?
Below the generated answer is a symbol with a pen. Click on it to start editing the text.

Document management

One of the most important things when using an AI assistant is that it has access to the right information. This is what document management is all about. There are two main ways to manage documents in an AI assistant. Pasting into chat and creating a knowledge base.

	Paste to chat	Knowledge base
How does it work?	The text is pasted directly into the chat, just as if you had written it yourself.	You first put the document into a digital "archive" and every time you ask the AI, a digital "archivist" will retrieve the relevant info so that the AI can use it.
Positive consequenses	Guaranteed that the information is used, it is directly in the conversation.	The information is stored permanently. You can have a lot more material saved, as the assistant only retrieves what is relevant.
Negative consequenses	Negative consequence	Not 100% guaranteed that all information is used on every occasion.

How to use a knowledge base?
NotebookLM is meant to be a tool based on this particular technology.
But in a regular AI assistant, the two features below, the project feature and custom agents, are the easiest ways to use knowledge bases.

Where can I find the Copilot feature?
Copilot does not have access to the "Knowledge Base feature", but it does have access to another valuable feature. It can access your documents that you have in Sharepoint and other Microsoft 365 apps.

The project function

The project feature allows you to gather multiple conversations in one place, like a common room. All chats you open within the project share the same material and the same memory.

There are two features you set up when you create a project:

Tell
the AI assistant who you are, what the project is about, and how you want help. It is automatically applied to all chats within the project.

Knowledge base
Upload course literature, notes, or articles once. It's available in all conversations within the project.

The collective memory is the strongest card of the project function. If you come up with something important in one conversation, that insight can carry over to the next. Do you have a course or topic you return to? Create a project for it.

Where can I find the Copilot feature?
The feature is missing from the version of Copilot that ORU provides.

Tailor-made agents

A custom-made agent serves as a template for a preset chat. Like the project function, there is "customized instruction" and "knowledge base", but the purpose is different.

Every chat starts with exactly the same conditions, like a fresh copy of the template. The chats do not share any memory; They are their own, independent conversations.

This makes agents perfect for situations where you want the AI to take on a role: an examiner providing feedback, a counterpart to practice a case against, or a course-specific assistant you want to share with your group/class. You package the template once, then anyone can open a chat with it without having to set anything up themselves.

Project Feature vs Custom Agent

	Project Function	Tailor-made agent
Customized instruction	Yes	Yes
Knowledge base	Yes	Yes
Shared memory	Yes, shared between all chats.	No, each chat is self-contained.
Able to share	No	Yes, can be packaged and shared.
Best suited for	Courses and projects you work on over time	Role-play, practice, feedback, or when you want to share a pre-set assistant.

If both are available and you don't have a specific need for shareability or role-playing, select the project feature. The shared memory makes it more powerful for everyday work.

Where can I find the Copilot feature?
Agents are available in the version of Copilot provided by ORU, but lack the Knowledge Base feature.

Web search

AI assistants generate their responses based on their training, basically an educated guess about what's right. It doesn't really know what's actually true. Web browsing solves a large part of the problem. Basically all AI assistants today can search for information online, and in most cases when you need something fact-related or current, it's a very good idea to enable it.

Often, the AI assistant chooses to search itself, sometimes you need to force it. You can then click on the plus symbol and select "Web Search", or write directly in the message that it should search online. You often get a source back, but never blindly trust it or the content you receive. It can still be wrong. AI sometimes refers to sources that do not exist, and even if they exist, the content may not reflect what it claims.

Deep Research

Deep Research is web searching on steroids. While most responses from an AI assistant take a few seconds, Deep Research can take up to half an hour. This is because the assistant acts independently on the internet: searching, evaluating, and following tracks. If it finds something interesting early, it can choose to dive deeper on that track instead of continuing broadly. It's not a simple search. It's an exploration. It spends many minutes gathering relevant information and compiles it into a detailed report with sources and links.

Here are three ways to use it well:

1. Map out the need before running
Have a discussion with the AI assistant first. Tell them that you plan to run a Deep Research and that you want to define what you are actually looking for. That way, you can land on a sharp question before you get started, and avoid the assistant spending time finding things that are not relevant.

2. Use it to find sources and open eyes
Deep research is great for finding materials, sources, and information you didn't know existed. Don't see it as a way to write texts, but see it as a way to explore an area and get relevant findings to work on.

3. Use it to create context
Are you missing relevant context, such as course material, but want to discuss a topic with the AI assistant? Run a Deep Research first. I also recommend saving the result and putting it in the knowledge base. Then it is left for future conversations.

Where can I find the Copilot feature?
In the upper right corner there is a menu where the mode "Automatic" is preselected. Click there and instead select "In-depth reflection".

💡 One of my most common ways to use Deep Research: I start a search for a topic I want to learn more about, then enter the results into NotebookLM and generate a podcast I can listen to on the bus.

Transcription

Transcription is an incredibly underrated feature. The idea is simple: you record your voice and it is converted into text. Almost all AI assistants have it, often with a microphone icon.

You can talk to the Assistant instead of typing. When I use AI, especially when I'm studying, I try to share as many thoughts and feelings as possible. What I find difficult, what I understand, what I have questions about, what I'm curious about and what I feel like I'm guessing. Combined with the learning mode, it becomes incredibly good. But getting all those nuances out via text can be heavy, then transcription is superbly smooth.

NOTE! This feature can be unreliable and often has a limit of about ten minutes. I have on several occasions talked for a quarter of an hour, got out lots of thoughts and thoughts, and when it is time to use, everything has either fallen away or been wrong. There are few times I feel as frustrated.

Voice Mode: In this mode, you can talk to the AI and it will respond, with a voice, without you having to press anything. It works differently depending on the tool.

Where can I find the Copilot feature?
Click the Microphone icon in the bottom right corner of the prompting bar.

Long-term memory

The AI systems that power the assistants have no built-in ability to remember information between sessions. The long-term memory feature solves that. It saves small pieces of text about you: who you are, how you want help, what you prefer. This allows the assistant to adapt to future conversations.

It's mainly about information about you and your preferences, not about specific projects. For that, the project function is the right way to go.

But there are weaknesses. Sometimes it saves something that was a temporary wish and treats it as a permanent truth that affects how it behaves in the future. It can give strange results. There are also privacy issues surrounding personal information being stored in this way.

If you want to turn the function on or off, you can usually find it under settings in the AI assistant.

Where can I find the Copilot feature?
Click the three dots in the upper-right corner and select Settings, then Personalizations. Turn on and manage Saved Memories.

Customized System Instruction

Is there a way that the AI assistant behaves that bothers you? Anything you want them to know about you? Anything you wish it did differently? Something that you often instruct about but repeat in every new chat?

Then there is something called a system instruction. It is a text field, usually under settings, where you can type in how the assistant should behave. There are different names in different tools. Look under Settings & personalization. What you type there affects all chats throughout the AI assistant.

Where can I find the Copilot feature?
Click the three dots in the upper-right corner and select Settings, then Personalizations. Enable and edit Custom Instructions.

Secure data management

When you use an AI assistant, you share information. Sometimes it's okay, but sometimes it can be risky. This may involve personal data, copyrighted material or other sensitive information.

In many cases, what you share is used to enable the company behind the assistant to improve its services, including by training AI models. This means that what you write may be available as knowledge in a future version. You don't always want that. I recommend turning that feature off. You usually do this under settings, in the section for privacy or privacy.

But even if you do, there are risks. This is no guarantee that sensitive information will not be leaked. Generally speaking, if you use any of the standard AI assistants, you should expect that what you write may become available to someone else. An easy way to test yourself: would you have been comfortable sharing this information with a stranger? If not, consider whether you really want to share it with a big tech company.

One way this is often tried to be handled is through an isolated cloud. For example, this is what Copilot via Örebro University's student account will enable: that the information is processed on Microsoft's servers and is guaranteed not to be spread. The problem is that you can't trust it one hundred percent. There is also an American law called the Cloud Act, which means that American authorities can, under certain circumstances, retrieve information even from Europe-based servers.