Can a Deep Dive support learning?

Google's new NotebookLM is a novel approach to interfacing with generative AI. Not quite an agent, not a chat..but one feature that caught my eye was the Deep Dive conversation.

As a teacher, the critical test of using new tech is "Does it help learning?". Let me share a shallow and a deeper use of NotebookLM I have tried in the classroom and the results thus far, then what my next idea is.

NotebookLM is Beta, which means it might be buggy but (perhaps only for now) it is free (another criterion met! ).

Right from the start, the "Notebook" approach is evident. As you can see, I have 3 notebooks. A notebook contains three sections: Sources - documents you upload, Chat (quite different from a regular chatbot) and Studio - this is the really interesting part. It will generate a conversation between two presenters (male and female, with tolerable American accents). As well as documents like a study guide, FAQs, etc.

Google's NotebookLM- inside a notebook

The "Deep Dive" conversation is based on the documents you have uploaded and completely surprised me with the insights, depth, and how it brought relevant examples and ideas in that were not part of the source document. The biggest surprise to me was how easy it was to listen to with a flawless conversational style - no robotic voices. Ok, actually the biggest surprise was the scientific accuracy of what they said! (more on that later)

A Quick and Dirty example 

Starting a new module Yr11 Physics class: we had started Module 3, tIQ1 but after camp interruptions, I was looking for an idea to recap what we had done and look ahead as well. I simply extracted the Module 3 part of the NESA Year 11 Physics syllabus and uploaded that as the only source. Without any further modifications, It generated a conversation between the two hosts. The students were actually engaged for the 10 minute section I have played them. Don't believe to me, try it for yourself (the whole purpose of this post)! However, if you want to try a sample before signing up, click the player on the above to hear what it created in the 7 minutes (2 to create, 5 to render the audio) I had before the bell rang.

Customising the Deep Dive.

Without any instructions, it will just do its thing. But let's say that you wanted it to (as I did) go deep into Inquiry Question 1, and then "tease" the topics coming in the remaining inquiry questions. I did try creating a document, added to the sources which explicitly told it this. Does not work. However, in the Studio pane, next to "generate" is a button labelled "customise". Ok, my excuse is that this button disappears once you have generated the discussion, you must delete the conversation first to see it. Using "customise" is just like writing any genAI prompt - tell it what you want it to do. Again, I was positively surprised with the scientific accuracy. One host was a little weird pronouncing the wave equation, but the female host said it correctly. The new ideas and examples they added (beyond the source document) were excellent. Here is my customising prompt:

You can listen to the customised result using the player on the right.

Next ideas

My Stage 5 science class have started on Homeostasis and feedback loops, with the rest of the new "Diseases" topic to go. When they return from camp next week, I am planning on generating a Deep Dive that goes into detail on what they have covered and then tease the rest.

This time, I am going to have students visualise a y-axis (front to back of classroom) and an x-axis (left to right of classroom). They stand in the classroom according to: "how useful is this to consolidate your learning?" (y-axis) and "how accurate is the science?" (insert your domain here) in the conversation (x-axis). I use this idea to get their reflection and feedback in many contexts, so they are familiar with the activity. A great starter to a class discussion, part of that exploring where they thought the AI was wrong (Oh, how my students love pointing out errors). I'll report back on how it went - subscribe to this blog if you would like to get that post.

The Meta Lesson

In both examples where I have used it (so far), my goal was (of course) the science learning - to get them thinking. My second goal was the meta lesson -how could you use genAI ethically to support your learning? We chatted about other subjects / topics where they could use this to create short, easy listening "podcasts" as an alternative study activity. Regardless of our personal opinions about genAI, it will be part of their future. We need to equip them with adaptable skills AND good examples of using genAI well. I explain "well" this way to my students, "Your aim should be to use AI more like a pushbike than an Uber".

Sitting in the back of an Uber (probably on your phone) is to completely outsource the navigation, effort, and experience. Whereas a pushbike magnifies your effort so that you go further and faster than you could without, but you are still the one doing and learning. I am always humbled by maturity of my students on topics like this, along with their thirst for guidance and ideas. They are as apprehensive (and excited) about the future as any of us teachers and we are vital role models with a duty to prepare them as best as we can.

More in Google NotebookLM

Lot's more inNotebookLM that has potential for learning...but that is for another post.