Google NotebookLM — The Future of AI

I’ve been playing around with Google NotebookLM and I love the concept.

It is currently “only available in the US” but I found a workaround.

Before I jump into my review, let me tell you how to use Google NotebookLM in Canada, Europe, or any other country outside the USA.

I use the Google Chrome browser, and I just downloaded the free extension called Browsec VPN.

Just set your location to the USA, and you are good to go to use Notebook LM.

What is NotebookLM?

It is a website where you can upload a bunch of files that are specific to whatever topic you are working on. Then Google uses its own LLM — Gemini Pro — to analyze your own documents.

It let’s you upload up to 20 files for each notebook. If you want more information than that, you can combine documents into one file.

For example, if you were using 40 articles as a reference, then you could cut and paste the articles on put 2 in each Google Doc.

What I love about NotebookLM

I think this format will be the future of how we interface with AI. When we are using AI, we often want to ask questions about only a subset of data, not the entire internet, and not about the entire corpus of data the LLM was trained on.

For example, let’s take the article I’m currently writing for example. I can use NotbookLM to upload several articles, notes, and documents that I personally chose as a reference.

There is a lot of garbage on the internet, and I want to be able to direct the AI to focus on ONLY the information I provide, and answer everything based off of only that.

I did in fact, use it for this very article.

Here is a screenshot of the interface, and how I used it:

I uploaded 3 documents I prepared on this topic. Then I prompted the following:

Based on what I uploaded, give me 5 bullet points to include in my article about what is useful about NotebookLM

Then it generated these 5 points:

• Privacy: NotebookLM does not share or store your sources with anyone, so you can rest assured that your data is safe and private.
• Collaboration: NotebookLM allows you to share your notebooks with others, making it easy to collaborate on projects and get feedback from others.
• Source Grounded AI: This means that NotebookLM can only draw on authoritative, private, and trusted information that you provide it with.
• Citations: NotebookLM provides citations for all of the information it generates, so you can easily verify the accuracy of the information.
• Large context window: NotebookLM can process up to 4 million words of text, which gives it a much larger context window than any other AI tool on the market.

Those are pretty good, so I will leave them unchanged.

What I don’t like about NotebookLM

It reminds me of making a custom GPT with ChatGPT. I like the interface of NotebookLM better than a GPT, but the concept is similar. You can upload your own work in there, and ask the model to use whatever you uploaded.

The problem, is that this sounds better in theory than it is in reality.

The models don’t seem to learn from your uploaded data as well as you may hope.

For example, if you upload 20 long essays that you wrote, and said, “write a new essay based on my writing style”, the LLM won’t be able to mimic your writing style very well.

It can read the information you give it, but it doesn’t use its full power to “train” on it.

You can ask it summarize what you said, and ask it to find information, etc, but it is not the same as “training” on your data.

Another problem is that it still does make mistakes or hallucinates. You need to fact check everything.

Overall, NotebookLM is a really cool concept and very much worth using if you are working on a new project. I highly recommend experimenting with it.

I think it will get a lot better (and all major LLMs will have something like this soon, ChatGPT, Claude, etc.

Here is what needs to happen for Google NotebookLM to get insanely useful, and what I predict will happen within the next 2 years:

1. The models will get a lot better — better logic, better reasoning, less error prone.
2. The context windows will get larger (You will be able to share more than 20 documents)
3. This will become multimodal — You’ll be able to upload video, audio, and pretty much any format. You will also be able to get any format as an output
4. The models will be able to actually “train” on your uploaded information. This will be the biggest ones.

Sign up to our emails to learn more about AI: AI Growth Guys Newsletter

Check out our new YouTube Channel

Follow us at our website: AI Growth Guys

Set up an AI consulting call with me: Andrew Best Minnect