Earlier this year, Google announced Project Tailwind, a notetaking AI for learning and productivity.
Today, Google Labs announced the continuation of Project Tailwind under a new moniker: NotebookLM.
This experimental project uses a large language model integrated into Google tools, such as Google Docs, essentially acting as a flexible AI assistant.
Crucially, NotebookLM learns from user data, acting as a virtual research assistant capable of summarizing facts, explaining complex ideas, and generating new discussion points – all based on the user’s selected sources.
A key distinction between NotebookLM and conventional AI chatbots is the ability to “ground” the language model in the user’s notes and sources. Google Labs has designed NotebookLM to restrict the model’s access to only the source material users choose to upload.
For instance, if you’re writing a thesis, you could upload your sources and use NotebookLM to sort through them and retrieve information.
NotebookLM’s self-contained nature helps mitigate the risk of “hallucinations” and misinformation.
Beginning today, select users in the US can use NotebookLM in Google Docs for the following tasks:
- Summarize: NotebookLM can auto-generate a summary of a Google Doc, providing key topics, questions, and more.
- Ask questions: Users can explore information about their uploaded documents.
- Generate ideas: NotebookLM offers features for brainstorming and creativity.
As a project, NotebookLM is still in its early stages. In time, Google intends to fulfill 2 primary objectives:
- Co-developing the product with users: Google Labs aims to regularly interact with users to identify strengths and gaps, striving to evolve NotebookLM into a user-driven product.
- Responsible technology rollout: Google plans to adhere to strict safety criteria and incorporate suitable safeguards before broadening user access and introducing new features. This is essential as the AI will work with users’ data, and Google knows this will rouse skepticism. Google clarified that the data collected is not used to train new AI models, ensuring user files and dialogues with the AI remain private. Moreover, users can choose what documents to use with the AI.
More features are forthcoming, and eligible users can sign up for the waitlist on Google’s blog here.