Google NotebookLM: AI Note-taking

Ask a powerful language model to understand documents you upload.

NotebookLM

In July 2023, Google had released an experimental version of its LLM that could work on a smaller set of your own documents, instead of the billions of public documents the large models are trained on. In December 2023, Google updated it to work with its latest LLM, Gemini. Also in the same release, Google added features and cleaned up the user interface.

So this new NotebookLM (LM stands for Language Model) is free and useful enough to deserve a serious trial from you. It has the feel and AI power of the larger models (e.g. ChatGPT, Bard), but it restricts its answers to the limited set of documents you have uploaded.

Here is my review. I created a small sandbox of documents about NotebookLM itself, so I could ask it questions about itself!!

Brace yourself. This is a long post, but written in clear language and moving fast.

Key Terminology:

Google is using a terminology that can be unfamiliar to us. So here are the key terms and the hierarchy of features in NotebookLM:

• NotebookLM Product

  • Multiple Notebooks, each of which has

    • Source Documents

      • Document’s text

      • Source Guide - Automatic Summary

      • Source Topic list - automatic

    • Chat is a collection of

      • Prompts

        • The user writes a prompt

        • The system suggests a prompt

      • AI system’s response (answer)

    • Notes

      • “Saved Response”: The system’s response (answer), converted to a note

      • “Written Note”: A free-form note written by the user

  • Help pages

  • Discord community forum

  • Share a notebook with others by email

Getting started: clean user interface

To start, head over to the NotebookLM web app. The first dashboard screen presents you with any notebooks you had created earlier.

Notebook playground

Selecting the “New Notebook” opens up an empty playground for that notebook.

Bringing in source documents is as easy as uploading text documents (not Microsoft Word, though) or copy-paste text from any source, including web pages:

I sourced 3 documents, all 3 based on web pages created by Google. Since there is no place to ask NotebookLM to visit a URL, I had to go to each URL in the browser and cut and paste the relevant text. NotebookLM automatically redacted any images or videos.

Here is one document I made from a NotebookLM blog Google had posted:

I ended with just 3 source documents, all from the web:

Start the Chat: Create a document

Now that we have source documents, let’s chat. I thought I would just tell it what I wanted it to do. Include enough details and see what it does.

I need your help in writing a detailed blog post on NotebookLM. It should have (1) an explanation of why NotebookLM is interesting (2) a review of key steps needed in using it (3) key use cases and (4) a conclusion telling the reader whether they should start using NotebookLM or not.

Answers came back in the typical chat-engine conversational format:

And here is what it generated. In other words, NotebookLM generated this document, the blog post it recommends me to post. Note the response below does stick pretty close to what I asked it to do in my prompt:

NotebookLM: Revolutionizing Information Synthesis with AI-Powered Notetaking

NotebookLM, a revolutionary product from Google, is an AI-driven notebook designed to enhance the way we process and understand information. By combining the power of language models with your existing content, it offers a groundbreaking tool for faster insights and creative idea generation.

Why NotebookLM is Interesting:

1. Personalized AI Assistant: Unlike traditional AI chatbots, NotebookLM can be "grounded" in specific sources, creating a personalized AI assistant tailored to your needs.

2. Efficient Information Synthesis: NotebookLM helps you synthesize information from various sources, saving you time and effort in consolidating knowledge.

3. Dynamic Idea Generation: It can generate creative ideas based on uploaded content, aiding in brainstorming and problem-solving.

4. Responsible AI Development: Google approaches NotebookLM's development responsibly, prioritizing user feedback and deploying safety measures.

Key Steps in Using NotebookLM:

1. Grounding the AI: Select Google Docs or upload relevant documents to ground NotebookLM in your desired content.

2. Summary Generation: Obtain a summary of your uploaded content, along with key topics and questions for deeper understanding.

3. Question and Answer: Ask specific questions about your documents, allowing NotebookLM to provide insightful answers.

4. Idea Generation: Request NotebookLM to generate new ideas based on your content, fostering creativity and exploration.

5. Fact-Checking: Always verify the AI's responses against your original sources to ensure accuracy.

Key Use Cases:

1. Research and Academia: Summarize complex research papers, clarify concepts, and generate study guides.

2. Content Creation: Develop outlines, compose short videos, or craft compelling pitches for investors.

3. Professional Knowledge Work: Extract insights from reports, brainstorm ideas for projects, and generate meeting notes.

Conclusion: Should You Start Using NotebookLM?

If you're seeking a tool to enhance your information synthesis and idea generation, NotebookLM is worth exploring. While still in its experimental phase, its potential to revolutionize how we interact with information is evident.

1. Early Adoption: Joining the waitlist grants you early access and the opportunity to contribute to its development.

2. Continued Refinement: Google's commitment to user feedback and responsible deployment suggests ongoing improvements.

3. Potential for Transformation: NotebookLM's innovative approach to AI-powered notetaking holds the potential to transform your learning and productivity workflows.

Ultimately, the decision to use NotebookLM depends on your specific needs and willingness to embrace cutting-edge technology. If you're intrigued by the prospect of an AI-driven assistant for information synthesis and idea generation, NotebookLM is worth trying.

You can Summerize all your Source Documents as a collection

Let’s find out what we get if we prompt it to just summarize the 3 documents as a collection:

Can you briefly summarize the source documents?

And here is the summary. Pretty well written and cleanly bulletized. And brief, as we asked.

Create a Glossary

When I asked, it dutifully created a good list of key terms present in my source documents. Lesson: it does understand the world, and what a glossary means. So it is a versatile Language Model but focused primarily on the limited set of source documents.

Knowledge is limited to Source Documents

To be sure, I wanted to know where all its wisdom came from:

Did you use any source documents other than the 3 I uploaded?

Its response put a smile on my face:

I also asked if it could tell me the mass of the moon, but it demurred.

So it does seem that each notebook plays mostly within the sandbox of the source documents you upload. Even if the mass of the moon is a pretty simple query, the engine chose to find answers only from the source documents, which say nothing about the mass of the moon.

Hence the importance of selecting good sources.

It would be really good, though, if Google could combine its Search prowess and go out on its own and select quality content and turn it into sources. i.e. we, the lazy users, would much rather let the computer find good sources. All we do is ask questions to that limited domain.

Source Document and Source Guide

By the way, source documents can be viewed, added, or deleted from a Notebook, but they cannot be edited. NotebookLM has no editor built in. It is a chat engine, not a document composition engine.

Clicking on a source document’s card gives you access to the full source document, with an automatic document summary and an extracted list of topics the document covers (“SourceGuide”):

Can not create files (PDF, Word, PPT or others)

So let’s see what other end products it can create.

Can you create a presentation slide deck following the blog post outline I gave you earlier?

Response. This is what most chat engines are doing at this time - give a hierarchical bulleted list, in an '“outline” format, which can be pasted into Google Slides or Microsoft PPT outline:

Okay, so we are not getting actual slides here. But maybe these engines will soon be integrated into Google Slides. Microsoft has already started integrating GitHub Copilot into Office applications. We are not too far from fuller integration of AI into our business applications.

Not multi-modal: Can Not handle images. Yet.

This is rather bland-looking, visually. Can you add any pictures, photos, or graphs to liven this up?

Here is the answer:

I was expecting images to show up. But this experimental version is a chat engine at its core. So it tells me what types of images and diagrams would be appropriate for this blog post. Looking forward to the day when its input and output both are multi-modal: i.e. both accept and train on images, podcasts, and videos, as well as include them in output.

Hallucination?

Now ensued an interesting conversation on images. I was just polite and gave my thanks for its efforts. And it seems to “hallucinate” that it does work with images and has provided me images in the answers.

So NotebookLM is just not doing images at this time, although somehow, it thought that it had provided images. Nothing wrong with it - we are at the early stages of this revolution. Hold your horses:

I am sure these “adjacent” capabilities will be coming soon to a chat engine near you.

Do not upload Private or Secure Documents

Since NotebookLM is in an Experimental mode, it is very likely humans as well as AI systems can see your data. What they would do with your data, whether they protect it and how, and whether they associate it with your identity or not seem to me to be open questions. These questions may be covered in Google’s Privacy Policies and User Agreements.

But, if I were you, I would not take that chance until Google creates a supported enterprise-hosted version.

Rating its responses

A rating feature for every response:

which I am sure is useful to keep training the chat engine. Also includes a convenient copy button to take the answer text outside of NotebookLM.

Pin to save a response at the top of the Notebook

Clicking on the “pin” button on every answer

saves that answer as a note, i.e. that particular response in a thread is archived and available later:

Help

There are a few sparse pages of help on Google’s site, just the bare minimum. If you want more, they also have a Discord community you can sign into:

All Notebooks are automatically saved

And, finally, I checked to make sure that NotebookLM does keep an archive of your “Notebooks” - each notebook being a thread focused on a specific topic. So the next time you come to it, the saved notebooks show up on the Dashboard:

So far, it does not appear there is any limit on number of Notebooks you can save, or amount of source documents you can upload. If the product catches on, I am sure those limitations will crop up.

What is in NotebookLM’s future?

And when I asked it about its future direction, it did respond:

I have no idea if these are actual Google plan, or it is what AI has inferred based on its general knowledge about itself.

Application example: Contracts

I can imagine uploading a (not very long) business contract, term sheets, slide outlines, etc, and asking NotebookLM to summarize the contract and ask specific questions about it. A pretty useful enterprise tool where everyone who may need to understand a contract with a client can now ask natural language questions instead of reading the legalese in the contract!

Application example: 10-K investor reports

Ditto. Upload investment documents. Give it time to digest. Tell it what you are interested in and why, and you will get good answers back. Days of reading the full 10-K reports may be over.

Application example: Medical Research

Imagine a medical researcher, uploading past research articles as source documents, and then doing Q&A with the system to understand what those documents are saying. Researcher can also start planning out how they should write their own research paper.

Conclusion

A pretty useful tool, available for free, to let a chat engine learn from documents you select and upload as reference material. Ask questions to that set of documents, save the answers, and copy-paste to any other document format you want (PDF, Word, PPT, Google Docs, Email, etc).

In other words, a powerful LLM (Google Gemini in this case), let loose on your own source/training documents.

Create multiple Notebooks, each with its own set of source documents you upload. A set of source documents that can be added or deleted, so the knowledge captured in any given Notebook is dynamic - it can change with the times.

Very useful tool. I hope that Google will market it widely, get a lot of users, learn a lot, and iterate.