The New Google AI-Powered Research Tool is a Killer App
You need to see (or listen to) it to believe it!
For a company that developed the ‘T’ in GPT, Google had been a late-comer to the AI game.
Google seemingly to be caught off-guard when OpenAI released their now legendary ChatGPT just over 2 years ago and has been on back-footing ever since.
But if the recent barrage of AI announcements is any indication, it’s clear that Google has returned, determined to remind everyone that it ‘owned’ the domain from the beginning.
One of these recent AI innovations was one AI-powered research tool called NotebookLM.
Built on top of Google’s Gemini 2.0, NotebookLM can correlate, identify and synthesize information from various type of materials. It can then provide summarization and extract meaningful insights from which users can ask questions in an interactive-chat style interface.
Distilled Knowledge at your fingertips
Unlike other Gen AI bots, NotebookLM’s knowledgebase is limited ONLY to resources that you share inside each ‘notebook’ (a project container).
This reduces the tendency to hallucinate (to make things up), which we have observed happening often with Large Language Model AI.
You can feed the AI not only with:
Textual Documents (e.g. Plain Text, PDF, Google Docs, etc), but also
Audio (mp3 files) and
URL links to websites and YouTube videos!
Let’s give the above sentences a chance to sink in for a moment and imagine what you can achieve if you have combined knowledge from all these different types of resources at your disposal, ready to be queried.
The NotebookLM free account (yes it’s free!) limits the number of materials you can upload or share to 50 in one project.
You can also choose, by checking the box next to each resource, which materials are included in the current session of interaction.
This way you can potentially load different material from different topics in a single notebook, and target your query to only specific resources.
Audio Deepdive
All of the above functionalities are awesome, but they are not ultra special that you cannot get from other LLM Models.
NotebookLM's multimodal synthesis ability is quite impressive, however, this is not the most amazing part of it.
What is even more jaw-dropping is the Audio Overview function!
It is simply a Podcast, an audio conversation between two AI voices (male and female), discussing the resources.
You have to listen to it to really be able to judge and consider the implications of it.
Because if no one tells you that the audio was AI-generated, you would think that you are listening to two people having a normal conversation about things.
The default prompt to generate this audio, which is no prompt at all, will give you between 10 to 20 minutes of audio depending on how much resource it draws the information from.
It will discuss mostly the insights from the materials you choose, but it’s not like some monotonic speech-to-text reading verbatim from a page.
The voice sounds naturally human-like, including natural diction, dynamic volume in the voice, and even some laughter. Conversations are fluid, though I feel that the choice of words is often a bit too cheerful and exuberant.
Oftentimes, it adds its own perspective to the conversation! It even sprinkles some anecdotes!
Of course, there are some glitches, but if I take a portion of the audio without the glitch (I will share some samples below), you probably won’t realise that they are fake!
However, here is where it becomes a bit dangerous, the deeper it tries to find its own insight, the further it is from the topics, and sometimes it will start making things up 😊
Therefore it is important to add additional prompts, to tell it to stay focused on the topic and don’t make things up. Once I added this instruction, I was able to reduce the hallucination.
What also is good about this audio format, is that you can listen to it while you are doing something else. It is such a time-saver.
To me, this audio podcast is one of the most incredible use of AI so far!
It turned AI sceptics (my family) into believers!
Here are some use cases from what I’ve been using this for:
Anti Money Laundering and Human Trafficking knowledge-base
I loaded 49 PDF publications published by world regulatory bodies on the topic of AML and Human Trafficking (it’s my domain expertise). On average, these documents may have tens of pages each, and jam-packed with valuable information.
If I had to read them one by one, it would’ve taken me days. Not to mention the fact that I have to take notes manually, and try to relate and synthesize information among these documents by myself.
Inside this notebook I was able to ask questions from the constraints of the documents I shared. It can give me insights from the combined knowledge of these materials.
All of this information I could get in an instant without reading all those hundreds of pages.
And then there is the Audio Deepdive!
Have a listen to the first 2 mins of it and let me know what you think.
YouTube Videos Summarizer
Not so long ago, I shared about my experience on using Harpa.AI.
It’s a Chrome extension that can summarize a YouTube video for you in text format.
It was a little clunky to use and it didn’t always give me what I wanted, not to mention it was slow, but it was quite impressive nonetheless.
With NotebookLM, you can just share the YouTube’s Video URL and it will take care of the rest.
I love reading books, and often people would make videos recommending books they had read.
I shared 7 YouTube videos from some of my favourite influencers, Ali Abdaal and Codie Sanchez. There are more than 100 books being recommended in these videos and I was too lazy to write them down one by one by hand 😉
So I asked NotebookLM to list all the book titles with the respective writers which are mentioned in these videos.
Here is the result of the query: Book List
Additionally, I generated the Audio Deepdive, which gave me a 30 mins summary of those recommendations from across 7 videos.
I am sharing the full audio here to show you how it sounds like.
Universal Translator
You are not limited to share English materials, you can upload/share materials written/spoken in other languages!
NotebookLM will internally translate it to English and then you can query the resources using English as usual.
Even the audio deep dive is able to talk about the materials in English (it doesn’t currently have the ability to ‘speak’ in other languages - but this is not hard to do).
What does this mean in a practical sense?!
It means that you can enjoy all those videos, audios and text from other languages and are able to draw insights from it, immediately.
So far I have tried to load:
A Korean singing competition video
A Math lecture in Slovak
A Ted Talk video German
A PDF of my mortgage contract written in Slovak 🙂
All were translated and distilled in seconds, ready for me to draw insights and to be explained in plain English.
The audio deep dive from these was really fun to listen to.
I am sharing the original YouTube Ted Talk in German, and the Audio deep-dive on the talk.
Legal Assistant
Yup, there is no way I would not load one of those legal documents, terms and conditions, or privacy policies, and have NotebookLM decipher them for me.
And it worked, it even worked for non-English materials!
I mentioned above about my mortgage contract (in Slovak).
Then I also tried to load the latest Instagram’s Terms of Use.
The results were just fascinating, but more importantly, it helped me to understand the intricacies of these documents, without me having to read them.
Checkout this Audio deep-dive about Instagram Terms of Use.
It describes in detail what rights do you have of your contents, and what you allow Instagram to use them for.
Oh, and do you know that Instagram is planning to roll out a subscription?
Website Summarizer
This one is very powerful.
You can get summaries and insights just by sharing URL links to some websites.
I tried this with my own 4-series newsletters about the secret of longevity.
One of the predefined prompts is the ‘Briefing Doc’, so I tried this ‘button’, and here is what I got:
The audio deep-dive from this material came up with something new that was not in my original articles. This has helped me to improve my writing.
It’s only 9 minutes long, so I’m sharing the whole thing directly here.
This got me thinking, for Newletter writers or bloggers, I think it’s a really good idea to provide this audio summary as your article’s Audio TL;DR.
Job Search Assistant - Resumé Reviewer - Career Advisor
This one I found it really incredible and especially for people who are currently searching for a job, it is invaluable.
You can upload your resume and you will get valuable feedback ‘immediately’!
You can ask NotebookLM to explain to you the strengths and weaknesses, and ways how to improve your resumé.
You can tailor the query (the prompt) to ask specific questions, for a specific job position for example.
Speaking about job opening positions, you can also upload the whole text of the job description and requirements. NotebookLM is able to compare your resume against the requirements and give you a detailed analysis of how you fare.
Furthermore, you can ask it to assist you with the interview questions. NotebookLM will give you example interview questions that it considers would come up, and you can practice them
Last but not least, the now iconic audio deep dive! If you are too lazy to write questions or too lazy to read the answers (or prefer to listen), then give the audio a specific prompt and it will generate the podcast accordingly.
Below is the audio deep-dive on the job position as a Senior Project Manager.
As a bonus, this use case is not only working for the candidates but ALSO for the hiring managers/interviewers. As an HR person, you can ask the AI to guide you through the candidate’s resume and to highlight everything that you might need to pay attention to the same CV material, but from the point of view of the interviewer.
These audio deepdives are absolute gems! It’s like having your own personal carrer advisor! I urge you to try it yourself.
An absolute fun and exciting
I really have been having fun using NotebookLM, and when I showed it to my family, they were absolutely amazed too, much more than when I showed them the power of ChatGPT.
Google might be 2 years late, but what a come back it had. I believe that this type of AI modalities, implemented in ‘this’ way, is a killer app for AI.
Did I mention that it is FREE?!
We are truly living in a new and exciting time in our live, and I cannot wait for what is going to be next.
Leave comment if you’ve tried NotebookLM and what was your impression so far.
And if you haven’t, I encourage you to try it, I guarantee an Eureka moment.
Again leave a comment what you discover.
By the way, Google is not finished yet, this is just the beginning. I’ve seen more wonderful things from the deep mind, no pun intended, of Google on this front.
It’s upward and forward for them.
Subscribe to this newsletter if you want to keep up with the latest trends in AI and how it can be beneficial for us in a practical way.
Wow, some good usage cases here. I've just used it to help with some staff training. I recorded our session on Google Meet for a transcript, added some supporting articles and some related YouTube videos. Now they can ask Notebook for reminders instead of m. Lazy teaching or using the tools to make things more efficient? 😅
Chris, all the use cases are super rad!
i'll dive deeper into the article shortly