NotebookLM: The best use of AI I have found (so far)

This is a game-changer for extracting data and insights from any sources

We’re looking for a Learning Designer to join our Content Team at Eedi! Read more and apply by clicking here.

Regular readers will know that I have been down an AI rabbit hole in recent months. I wrote about my experimentations with AI to anticipate student misconceptions, write a multiple-choice diagnostic question, and plan a lesson.

However, my new favourite over the last few weeks is NotebookLM from Google. If you need to extract data from documents or are a research geek like me, this will blow your mind.

Use-case #1: Data extraction

For a project I am working on with Eedi, I needed a list of the White Rose Maths small steps from their primary schemes of work. These are housed in separate pdfs and look like this:

Until now, I had been manually copying and pasting each of these lists into a text editor (to remove all the formatting) and then into an Excel spreadsheet where I could work on it. This was pretty time-consuming.

Step-forward NotebookLM. I uploaded all the White Rose Maths pdfs - you can upload several documents - and then typed in the chatbox:

Please give me a list of each of the small steps in each of the blocks

10 seconds later, this came out:

A list that I could copy and paste straight into Excel.

So, if you want to extract data from a document, or several documents), NotebookLM is your new best friend.

Use-case #2: Summarisation

Right, strap yourselves in for this one. If you have ever faced the challenge of extracting the key insights from a lengthy document - perhaps a research paper - then you know how tough it can be. The language may be dry or inaccessible, the signal may be lost amidst the noise, or you don’t have the time required to read, digest and synthesise.

Once again, step forward NotebookLM.

I uploaded Learning verus Performance by two of my favourite former podcast guests, Nicholas Soderstrom and Robert Bjork. This is a seminal paper on memory and desirable difficulties… but it is something of an epic, coming in at a whopping 41 pages. You can download the original paper here.

A few seconds after the upload, NotebookLM produced this:

First, we have a useful summary. I have experimented with a number of papers, and this summary is always a paragraph long, written in the kind of clear accessible text that many research paper abstracts could learn a lot from.

Second, we have suggested questions or the opportunity to pose your own questions in the chatbox. Both of these options open the kind of back-and-forth dialogue anyone who has experimented with ChatGTP will be used to. I tried a question of my own:

What are the top, practical implications from this paper for teachers wanting to improve the learning of their students?

Here is the output:

Practical Implications of the Learning-Performance Distinction for Teachers

The sources offer several practical implications for teachers who want to enhance their students' long-term learning. Here are some of the most important takeaways:

• Don't mistake current performance for learning. Just because a student is doing well in class doesn't mean they are truly learning the material. Performance is what is observable in the moment, while learning refers to the durable and flexible knowledge or skills that persist over time. For example, a student who can flawlessly recite multiplication facts in class may not necessarily be able to apply that knowledge to solve more complex math problems or remember the facts a week later.

• Embrace "desirable difficulties." Certain teaching strategies, while appearing to hinder performance in the short term, can actually promote long-term learning. These "desirable difficulties" include distributing practice, varying practice conditions, and using testing as a learning tool. Although these methods might initially lead to more struggle and errors, they ultimately result in deeper and more lasting learning.

• Space out learning opportunities. Instead of cramming all instruction on a single topic into one lesson, space it out over multiple sessions. The spacing effect, a well-documented phenomenon in cognitive psychology, demonstrates that spacing out study or practice sessions leads to better long-term retention than massing them together. For instance, instead of teaching the concept of fractions in a single week, introduce it one day, review it a few days later, and then revisit it again the following week.

• Vary practice conditions. Encourage students to practice skills and apply knowledge in different contexts and with varied examples. This variability, while potentially leading to more errors during initial practice, enhances flexibility and transfer of learning. For example, when teaching students about different types of essays, instead of having them write only persuasive essays one week and only narrative essays the next, mix up the essay types within and across practice sessions.

• Use testing as a learning tool. Regular testing shouldn't be viewed merely as an assessment tool but rather as an integral part of the learning process. Retrieving information from memory, as is required during a test, strengthens the memory and makes it more resistant to forgetting. Incorporate low-stakes quizzes, retrieval practice activities, and opportunities for students to recall information without looking at their notes.

• Educate students about metacognition. Help students develop an awareness of their own learning processes and equip them to distinguish between effective and ineffective study strategies. Teach them about the learning-performance distinction and why strategies that feel more effortful, like spacing and testing, often lead to superior long-term retention. This metacognitive awareness can help students make more informed decisions about how they study and learn.

The sources provide compelling evidence that a shift in focus from short-term performance to long-term learning can significantly impact educational outcomes. By understanding and implementing the practical implications of the learning-performance distinction, teachers can help their students achieve deeper, more durable, and more flexible learning.

Pretty good, hey?

Best of all, each takeaway has a number that links back to the original source:

So, at any point, you can dive back into the source material to find the relevant section of the document from which the insight comes. This way, you can check if you agree with the takeaway or delve deeper if you choose.

But that is not the most impressive bit. At the click of a button, you can generate a flipping podcast on the uploaded document. And this is not just some robotic reading of the summary. Oh no, this is two super enthusiastic, super human-sounding hosts chatting away completely naturally about what they have taken from the paper.

Have a listen:

The scary thing is, they are better at hosting than me!

This opens up a whole new way to consume content. Upload it to NotebookLM, download the podcast you your phone (I use DropBox for this), and listen to it on your travels. It sure beats wading through multiple pages of sometimes unintelligible texts while hunched over your laptop.

And there is more…

You can also upload a selection of documents (up to 50) on the same theme and thus get insights from several sources on one theme.

And you can upload more than PDF documents. You can upload audio files, link to webpages, or even YouTube videos. And each time NotebookLM will perform the same impressive summarisation functions - podcast included. It will also invite you to ask questions to shape the output to your needs.

A word of caution

Needless to say, we need to be wary of falling into the same trap as when using these AI tools to identify misconceptions, create diagnostic questions, or plan lessons - we should not trust their output blindly or wholeheartedly. That is why I like the references feature so much - I can easily look at the original source and make my own judgement if needed. Moreover, because NotebookLM limits itself to looking at the source(s) you share with it, and only them, the chance of it hallucinating or pulling in other knowledge that is unrelated is not as great as with tools like ChatGTP that are pulling from many, many more sources.

So, for saving time, extracting and synthesising key insights from documents, allowing me to delve deeper, and presenting output in various formats for me to consume, I think NotebookLM is incredible.

How could you use NotebookLM?

Let me know in the comments below!

🏃🏻‍♂️ Before you go, have you…🏃🏻‍♂️

1. … checked out our incredible, brand-new, free resources from Eedi?

2. … read my latest Tips for Teachers newsletter about asking impossible questions?

3. … listened to my latest podcast with Ollie Lovell about a recent lesson he taught?

4. … considered booking some CPD, coaching, or maths departmental support?

5. … read my Tips for Teachers book?

Thanks so much for reading and have a great week!

Craig