Steven Johnson has written 13 books, on matters starting from a London cholera outbreak to the worth of video video games. He’s been a tv presenter and a podcast host. He’s a keynote speaker who doesn’t should name himself that in his LinkedIn profile. And for over a 12 months now, he’s been a full-time worker of Google, a standing that’s clear when he badges me into the search big’s Chelsea workplaces in New York to point out me what his crew has been creating.
It’s referred to as NotebookLM, and the best manner to consider it’s as an AI collaborator with entry to all of your supplies that sits in your metaphorical shoulder to information you thru your challenge. NotebookLM was soft-launched to a choose group earlier this 12 months however is now obtainable to all as an “experiment”—that’s Google’s low-risk approach to see how the app behaves and the way we behave with the app.
Johnson discovered his approach to Google by means of a lifelong obsession with software program as a “a dynamic thought accomplice,” a software to hurry up and improve the artistic course of. When he was in faculty he grew to become obsessive about HyperCard, Apple’s software program that broke data into chunks and allowed you to navigate an information-space via hyperlinks. It anticipated internet navigation earlier than the net existed. “I fought mightily to show HyperCard into that dream software, however it wasn’t fairly prepared,” he says. He ultimately grew to become an fanatic of Scrivener, a mixture phrase processor and challenge organizer well-liked with ebook authors. (I’m a fan too.)
When Johnson bought entry to OpenAI’s GPT-3 textual content generator in 2021, he acknowledged that AI may degree up a brand new technology of thought instruments. Oh, wait, he mentioned to himself, this factor that has all the time been behind my thoughts is now going to be doable. Situations unthinkable even a 12 months earlier than have been out of the blue on the desk. Johnson didn’t but know that Google not solely had related massive language fashions, however was already engaged on a challenge very a lot according to his considering. In Might 2022, a small crew within the experimental Google Labs division cold-emailed Johnson. They arrange a gathering through Starline, a Google Labs challenge that permits for eerily intimate in-person conferences. “I principally had a dialog with a hologram who mentioned, ‘You already know, this factor you’ve got been chasing your entire life? We are able to lastly construct it,’” Johnson says. He grew to become a part-time adviser to the small crew, at first sharing the workflow of an expert author. “Here is 4 or 5 engineers, this is an precise writer, let’s simply watch him,” is how Google Labs head Josh Woodward sums up the method. Finally Johnson bought concerned within the improvement of the product itself and was sucked in to the purpose of accepting a full-time gig. His title at Google Labs is editorial director.
NotebookLM, initially referred to as Mission Tailwind, begins by creating an information set of your supply materials, which you drag into the software from Google Docs or the clipboard. After the app has digested all of it, you possibly can then ask NotebookLM questions on your materials, because of Google’s massive language mannequin know-how—partly powered by its just-released improve Gemini. The solutions replicate not solely what’s in your supply materials but in addition the broader common understanding of the world that Gemini has. A vital function is that each reply to your queries comes with a set of citations reporting the place precisely the knowledge got here from, so customers can test the accuracy of its output.
Google shouldn’t be the one firm envisioning merchandise that permit individuals create customized knowledge units to discover with LLMs. At OpenAI’s developer day final month, the corporate launched customized mini-GPTs that may be tuned to a particular process. Woodward acknowledges a “core similarity.” However he argues that NotebookLM focuses extra on enhancing a workflow, and is geared to offer superior accuracy in its outputs. Additionally, he says that the OpenAI merchandise have extra of a character, whereas NotebookLM is designed to haven’t any such pretensions.