Original video in Chinese.
Key Takeaway
- NotebookLM is an experimental AI note-taking app released by Google Labs. It combines a chatbot and RAG, changing the logic of traditional note-taking.
- Its core functional areas are the Sources area (add documents), the Chat area (AI-guided questions and answers), and the Notes area (manually/automatically create notes).
- Through features like AI-suggested questions, source citations, and note generation, NotebookLM significantly improves users’ understanding of documents and the efficiency of knowledge retention.
- The product is designed to achieve “Conversational Learning,” letting AI help users digest materials and generate content.
- Although still in an early stage, NotebookLM shows the enormous potential of AI in learning and knowledge management.
This is the best AI learning product I’ve used.
It’s an experimental product Google Labs released not long ago: NotebookLM. They added a chatbot powered by a large language model and RAG to a traditional note-taking app, and the whole product logic changed.
Although NotebookLM is still in an early stage, the foundation is already in place. After using it for a while, I found it extremely helpful for my learning, as well as for knowledge retention and organization.
I strongly recommend everyone try it after watching this video. I’ve already started thinking about how to combine it with the DEVONthink and Obsidian I use every day.
I’ll start from scratch and show you.
Right now, this product is only available to users in the United States. But that’s no issue for us.
After logging in, you’ll see a somewhat bare-bones page like this: notebook creation and selection.
Once you enter a notebook’s detail page, there are just three core areas:
- Sources area
- Chat area
- Notes area
The Sources area is for adding documents. It supports importing from Google Drive, PDF uploads, or pasting text directly.
After a document is uploaded, you can select one document or multiple documents. Based on your selection, the AI will automatically analyze them and provide a Summary and Key Topics. At this point, suggested questions from the AI appear above the chat box in the Chat area.
What’s the benefit of designing it this way?
When we upload material we haven’t read, we often don’t know how to start a conversation with the AI—we know nothing about the content, so of course we don’t know what to ask.
At this point, we can click any of the Key Topics or suggested questions, and the AI will automatically give an answer. Each answer includes citations, referencing the source. Hovering over them will show the original text. If you click, it will take you directly to the corresponding place in the document, so you can also see the context and get a more complete understanding.
It’s worth noting that after you ask one question, the AI-suggested questions will update as well. So even if you don’t type in any questions yourself, you can still follow the AI’s guidance and, with a few clicks, get an initial understanding of a large document.
That’s the significant improvement brought by integrating large language model capabilities into a note-taking app.
The Notes area works the same way.
We can add notes manually, or click the pin button in any chat response to automatically turn it into a note.
When the notes are about done, we can select them all, and the AI will also suggest actions. For example, summarize them, merge them all, or create an outline. Giving commands directly through the chat is also fine.
Built on top of notes, plus chatbot and RAG, the whole note-taking app changes completely. Google calls it: Conversational Learning.
Just like in the demo I showed earlier, you start with a basic question, and the AI guides you through digesting the entire material. In the process, you can also easily create several notes. Finally, using those notes, you can even have the AI help you generate content.
Learning and producing output—that’s Conversational Learning. That’s also what draws me to this product.
Of course, as I said at the beginning, NotebookLM is still very early and has a lot of shortcomings. For example, it lacks the most basic multi-level folder feature. You can’t create subfolders, which is really inconvenient for organizing materials and selecting documents. Also, language support isn’t sufficient; it only supports English. If I ask a question in Chinese, the AI still replies in English. If I upload Chinese documents, the AI still replies in English.
Of course, these are secondary issues. As long as Google keeps improving around this core structure, there’s a very good chance it will build an extremely capable productivity tool.
This should be the Google product I’m most looking forward to.
OK, that’s it for this episode. Remember to like and follow. See you next time!