Google teases Project Tailwind — a prototype AI notebook that learns from your documents

Google has announced a lot of AI projects at its I/O conference this year, but the one that has me most excited is just a prototype: Project Tailwind.

Essentially, it’s an AI notebook trained on your documents that you can query like a sort of personalized tutor or writing companion. Google framed it as a tool for students, but it could have a lot more potential for anyone who deals with a lot of text in their life.

“Like a real notebook, your notes and your sources power Tailwind,” said Josh Woodward, a senior director of product management at Google. “How it works is you can simply pick the files from Google Drive, and it effectively creates a personalized and private AI model that has expertise in the information you give it. We’ve been developing this idea with authors like Steven Johnson and testing it at universities.”

In an example demo Woodward showed onstage, Tailwind was fed a bunch of study notes and then picked out a number of details, including key topics and suggested questions. A text box allowed Woodward to dig down into this information, creating a glossary of terms for a specific topic, for example.

It’s not clear if Google has any ambition to launch Project Tailwind at this point (the company described it as a “prototype” that had been “put together over the last few weeks” using the company’s new PaLM 2 API), but the concept is incredibly attractive.

As Woodward said, “We realized it’s not just for students. It’s helpful for anyone synthesizing information from many different sources that you choose. Like writers researching an article, or analysts going through earnings calls, or even lawyers going through a case. Imagine collaborating with an AI that’s grounded in what you’ve read and all your notes.”

It’s worth noting that this is a line of inquiry that many companies are pursuing at the moment. After all, AI language models can already be fine-tuned to deal with certain types of data, so why not fine-tune them on the user’s data? Note-taking app Notion is already pursuing this idea, while the open-source AI scene is full of developers fine-tuning leaked language models on the contents of their Google Drive.

There are some big potential challenges, of course. One is compute demands. Language models are expensive to train and fine-tune — who pays the bill for Project Tailwind? The other is reliability. These systems are notorious for making up information. There’s no guarantee they won’t do the same even if they’re just constrained to personal notes.

At any rate, it’s worth investigating. Google says you can sign up to test Project Tailwind as part of its new AI Labs program. We’ll be taking notes.