Human in the Looping with NotebookLM

DALL-E Humans in the loop

I’m an AI optimist; I’ve said that here many times and I expect I’ll keep saying it often. What this means for me is:

• I don’t believe AI will replace us all, steal all our jobs in the next, say, 5 years

• Some types of jobs may have more risk of that, I’m focused on my job as a “knowledge worker” and expert in cybersecurity.

• The idea that AI in copilot mode, or ‘centaur or cyborg’ mode as Ethan Mollick phrases it, is something I buy into. This has us using AI for the tasks that we as humans either cannot do well or find to be low-level recurring tasks we’d rather not have to bother with, while we work on the tasks where humans are more effective than AI.

• The word expert is big piece of this. We need expert level knowledge, or close to it, so that we can effectively gauge when the responses and results AI tools give us are valid and useful vs when they are inaccurate or not useful for whatever reason.

This way of working with AI, with people still in the mix, is often referred to as the need for humans in the loop. If you’ve been kind enough to read my ‘stuff’ here with any sort of regularity, you know that I’ve been trying out a lot of being the human in the loop, working in copilot mode, with various GenAI tools.

Over the last week or so I’ve been doing a lot of that with Google’s NotebookLM. Last weekend I did that with a long overdue cybersecurity book recommendation. Give that one a look if you want to see an overview of how NotebookLM works. This week I saw various newsletters mention that you can add YouTube video links to NotebookLM as sources, as you can with uploading files and website links.

I tried that out this morning with one of my favorite talks on cybersecurity risk management and how to talk to executives about it. It’s from a talk at the ICS Security conference, S4 Events, from three years ago and the speaker is Patrick Miller. This one:

I took lots of notes on this talk three years ago, put some of the guidance to use at work, and still think about the ideas and recommendations from it often. And that puts me in a prime position as an “expert” human in the loop to see how NotebookLM does with its “thoughts” about the talk.

So how did NotebookLM do? Very well I’d say. For starters, its suggested follow-up questions are good ones:

NotebookLM Risk Management & Executive Communication with Patrick Miller

The FAQ note it generated does a great job of highlighting some of the key messages in the talk:

NotebookLM FAQ

The Table of Contents is even better:

NotebookLM table of contents

As a final bit of back and forth with NotebookLM about Patrick Miller’s talk, I asked about a slice of it that still stands out to me - always - three years later, without reviewing any of my notes - words about cyber folks needing to communicate better, remember to lose some of the technical jargon, and includes a Starbucks reference. NotebookLM’s response was good here too:

NotebookLM response

When NotebookLM was first released, and for a number of months afterwards, I was underwhelmed by it. I didn’t like the lack of a mobile app for it (still don’t love that, although the web app is decent) and just didn’t imagine I would be using it a lot. Now I am properly whelmed by it. I find myself wanting to use it more almost every day and I really want to start finding ways to use it in combination with other GenAI apps soon.