The Only Generative AI I Want Is One Trained On My Personal Life
Services like Google's NotebookLM and Fabric are building AI experiences around analyzing your links, files, and personal notes.
That last year of AI hype has largely been driven by the allure of general knowledge chatbots. Experiments masquerading as services called Microsoft Bing, ChatGPT, and to a lesser extent, Google Bard, that attempt to answer just about any question or request with a natural language response, whether that's providing factual information you might dig up in a search through Google, or generating entirely “new” text.
There are real concerns to be raised about the use of these large language models (LLMs), but also dozens of exciting ways that they can be used. Chatbots and other AI features have spread to apps even quicker than Instagram Stories clones ever did, likely because the potential for these features as parsers and generators of information is so great.
The reality of living and working online is that we’re constantly bombarded with new information in the form of articles, videos, screenshots, and PDFs. No one has time to read or watch it all, but if there was a place to save it for later, that could also help you understand what you’ve saved and generate new ideas? That could really change how we use computers. And that’s something made possible by generative AI.
The AI Notebook
Google and Microsoft have given enterprise customers a preview of what that could look like with 365 Copilot and Duet AI. Two chatbots that, on top of being able to generate new slides, documents, and spreadsheets, are meant to be able to answer questions about all of the files and text you haven’t read, too. What’s more interesting to me though, is what happens when you apply the same technique to personal information.
Alongside Duet AI, updates to Android, and the launch of the Pixel Tablet, Google demoed a new AI-powered experiment at I/O 2023 that does just that. At the time, Google called the experiment Project Tailwind, and described it as a sort of research assistant and notebook all rolled into one. Tailwind was supposed to analyze the documents and files you shared with it and generate a “study guide” around the information, with the ability to answer questions, create new content based on the material you uploaded, and even find connections between documents. Perfect for students, haggard writers, and just about anyone who fits the definition of a “knowledge worker.”
The beta version of the service began rolling out to a select group of testers in July as NotebookLM. This early version of the service only accepts imports from Google Docs, and based on hands-ons, caps how many sources you can upload, but still seems like a potentially promising way to sort through the waves of information we’re expected to deal with.
Fabric, “an AI-native workspace and file explorer” I’ve picked up while waiting to get a chance to try NotebookLM, strikes at a similar idea. You can upload just about anything to Fabric, (documents, PDFs, images, videos, and links), create notes, and the “digital library of Alexandria” attempts to make it all searchable using natural language. Things you upload can be sorted into spaces and individual folders, and you can share links so you or multiple other people can interact with your content at the same time, multiplayer-style, with colorful cursors and comments.
The real magic of Fabric, which is essentially a more interactive Google Drive, is how the company is using AI. The Fabric Assistant is always a click away in the bottom right corner. You can ask the assistant to summarize a document, ask specific questions about its contents, and get suggestions of other things you’ve uploaded that could be related. Both services tackle the idea in different ways, but seem to agree on the idea that it’s helpful to have a guiding hand while looking through something you don’t remember saving a month ago.
Insights on demand
Context can be hard to come by for the average person online, and there’s a real possibility that the proliferation of generative AI across the web and apps is the end of that context. If I can get Google to give me a direct answer to a question, why would I bother checking out other links?
What’s exciting about what I’ll call “AI notebooks” like NotebookLM, Fabric, and other similar services, is that their first priority isn’t removing that context. Really, their job is to help you create it, directing you to things you already have that you can dive deeper into, and highlighting how the things you’ve collected matter together.
It’s a better, more realistic implementation of AI as a tool, rather than the be-all replacement for how we’ll find and generate knowledge in the future. And it’s one that feels like it could genuinely change the way we experience all of the things we do with our computers in the next year, rather than in the next 10.