Google’s NotebookLM could let users build custom chatbots from notebooks

If you had high expectations from NotebookLM, you probably need to raise them even more. It turns out we haven't yet discovered all the work-in-progress features, but let's take this opportunity to close that gap. NotebookLM appears to be developing a new feature that would allow users to create chatbots.

This feature would be accessible from the settings menu of an individual notebook, where users could enter the chatbot creation interface, which bears a resemblance to the Gem creation UI seen in the core Gemini product. Users will be able to prompt instructions for a chatbot that would have access to all the sources in that specific notebook, effectively turning the notebook into a dynamic knowledge base. Additionally, the interface would allow users to test chatbot responses and allow them to share a public link with others or potentially embed it into their website!

If this becomes a reality, it would be a significant development, as currently, Gems are designed to be private and aren't shareable, nor do they have the capability to interact with a specific knowledge base. This new feature could help close that gap. It may also resemble Claude projects with custom prompts—after creating and sharing such a chatbot, others would be able to interact with a specific knowledge base alongside all the notes generated by the notebook's author. This represents a substantial leap towards improved collaborative, AI-assisted work.

But that isn’t all we seem to have overlooked. It turns out that, within the main menu where users can create new notebooks, there will also be a drop-down option allowing the selection of an LLM model intended for the project.

Currently, there are some interesting upcoming models in the code. These are probably codenames / temporary, internal aliases, but we've got "HighTC" and "GoldBookLM" - excited to see what they end up being! Interestingly, the code features Med-PaLM, but the newer & better Med-Gemini isn't mentioned in the code.

Here's a summary of each Google AI model from the list:

• Med-PaLM Notebook: A specialized large language model designed for the medical domain, capable of answering medical questions and handling various biomedical data types.
• Gemini (Preview) Notebook: A multimodal, conversational AI model that integrates language, audio, code, and video understanding. The "Preview" indicator could refer to an upcoming model, or one of the experimental ones already available through AI Studio.
• GoldBookLM Notebook: No specific information available.
• Gemini Pro Vision Notebook: An advanced model for image and video understanding, optimized for tasks like object recognition and digital content analysis.
• HighTC Notebook: No specific information available.

This feature is very likely intended for internal use and testing as there’s no indication it will be made available to the general public. However, the chatbot creation feature seems to be aimed more towards public use. These features are not released yet, and as a result, it might all change in the final product.

Update: Additional info from @raiza_abubakar from Google on the topic 👀

4) An agentic and personalized writing workflow, especially using YOUR style and formats, is another type of “transform” that I’m really excited about. Can’t tell you how often I take a pile of research and some haphazard notes to write my POV on it – this streamlines that flow massively. Gemini 1.5 is really good at this and a well-done UX is what’s needed to connect the user to this capability. Not sure if the space is too crowded or kind of tired at this point, so I just need to study a little bit more before we put this in Notebook.

5) Custom chatbots… I have a lot to say. This is pretty widely used internally at Google and literally every day someone pings me to say “This has 10x’d our team’s productivity.” Not joking. In the hacks you’re still looking at the old version so I’m excited for what you all think when the new version launches :)

With the addition of features like Magic Draft and Audio Overview, NotebookLM is evolving into a very powerful AI-based product that holds promise for the future. Whether it becomes a standalone product or eventually integrates into the core Gemini offering in some way, it certainly has potential.

These features were data mined by Bedros, and this entire article was created as a collaboration between TestingCatalog and Bedros—as a piece of joint research.