NotebookLM
Updated for version: 3-11-2024
Accessible via: https://notebooklm.google/
Requires login via: Google
Page contents
Ratings
Accuracy / Quality ★★★★☆
Flexibility / Features ★★★★☆
Data security / Privacy ★★★★☆
Pros/cons
Pros
- NotebookLM “Grounds” in specific, user provided sources, rather than more general sources (Retrieval-Augmented Generation)
- Easily find connections between topics in sources
- Ability to generate podcasts summarizing your provided sources
- Allows orderly management of notes and sources
Cons
- Can only interpret text/audio
- 50 file upload maximum per notebook
- Hallucinations are possible, despite user-supplied sources
- Quality of the source material uploaded heavily influences output quality
Description
NotebookLM is a research and study-assistant chatbot created by Google. It is different from other Large Language Models as it especially focuses on documents uploaded by the user, using so-called Retrieval-Augmented Generation (RAG). The model is trained on many existing text files, but can be “grounded” in your provided text. After this grounding process, the language model applies the connections gained from training to the user provided sources. This way the model is able to generate responses based on your source material, rather than basing it on irrelevant information from all over the internet. This helps you to gain fast insights and knowledge of large amounts of textual data.
In this guide, we will give an example of how to use NotebookLM to summarize content and 'interact' with the content. Moreover, we will let NotebookLM create a podcast on the background information we provide.
Features and examples
Basic questions and tasks
After creating a new notebook, NotebookLM will directly ask you to upload sources. NotebookLM supports Google Docs, Google Slides, PDF, text and markdown files, web URLs, copy-pasted text and YouTube URLs of public videos. Therefore, in many cases you might need to change your file type from the Office format (Word, PowerPoint) to Google’s file format by uploading it to your Google Drive.
Note: NotebookLM can only be used for textual inputs and not for numerical inputs (like Excel documents) yet.
After uploading documents, a dashboard appears. In this dashboard, you can do multiple things: On the left side of the screen you can (de)select and upload up to 50 sources you wish to work with. By clicking on these sources, the source will be opened alongside a summary and a list of key topics.
You can invite others to contribute to your notebook through the “Share” button on the top right of the screen.
On the bottom of the page, there is a chatbox where you can ask questions and generate new ideas. NotebookLM helps you to get started with some example questions. The answer provided includes references to the specific uploaded sources where NotebookLM retrieved the information from.
You can ask NotebookLM to make quizzes on your provided content. This is most suited for knowledge questions, not for reasoning-level questions.
Notes
You can save interesting answers to your notes, by clicking on the “Save to note” button on the bottom-right corner of the response. Moreover, you can create your own notes by clicking the “Add note” button on the top left of the page.
By selecting one or more notes, you can ask questions or give tasks to NotebookLM for these specific notes. This way, you can ask NotebookLM to make quizzes, find connections and do many other tasks.
You can upload your quiz questions to Claude to make an interactive quiz.
Notebook guide and podcast generation
In the Notebook guide, you can find several examples of how the tool can help you interpet sources, such as making a timeline or make quizes. Moreover, in this tab, you can generate a downloadable audio file in the form of a “deep-dive podcast”. You can customize the podcast by providing a prompt, for example to specify the tone of the podcast, the level of depth of the topics discussed, provide some additional contextual information (e.g. "take into account that the audience is familiar with the format"), or to point towards a specific topic within the content to be discussed (e.g. "highlighting" a specific part of the uploaded content).
