The chatbot turns into the trainer

Steven Johnson is a really meta writer. He writes often about science and expertise, and likes to immerse himself within the issues he’s protecting, even utilizing them to alter the best way he writes books. A few years in the past, a number of months earlier than ChatGPT launched and the AI growth took over the tech world, Johnson received {a magazine} task that despatched him actually, actually deep down the AI rabbit gap. And he by no means got here again up.

Now, along with writing books, Johnson can also be working at Google. He’s a part of the staff constructing a product referred to as NotebookLM — “Pocket book,” because the staff calls it. It’s a note-taking and analysis software: you add paperwork and import net hyperlinks, and Pocket book’s Gemini-powered AI helps you manage issues, extract info, and perceive a topic higher. “They reached out,” Johnson says once I ask how he received concerned with Google, “and mentioned, ‘hey, you’ve been dreaming of this very best software program software that helps you manage your ideas and helps you write and helps you formulate connections and brainstorm. We expect we are able to do it now.” Johnson signed up, and has been at Google for the reason that summer time of 2022.

The product itself first launched in 2023 as Challenge Tailwind, and has since been rebranded and expanded in huge methods. Simply final week, the staff launched Audio Overviews, which generates a podcast — with two chatty hosts, loads of forwards and backwards, and a very outstanding penchant for the phrases “deep dive” and “buckle up” — primarily based on the knowledge you present. It’s fascinating, it’s difficult, and it’s getting higher actually quick.

On this episode of The Vergecast, Johnson joins to debate his fascination with AI, his time at Google, and the current and way forward for NotebookLM. We discuss concerning the difficult points raised by a software like this, and whether or not it’s okay to let an AI do your analysis and homework.

We additionally speak about how to verify a software like NotebookLM is each correct and simply fact-checked, why context home windows are extra essential to the way forward for AI than most individuals understand, and the way usually AI podcast hosts ought to say “like” in dialog. And we speak about Johnson’s personal course of as a author and creator, and the way AI is altering the best way he works.

If you wish to know extra on every little thing we talk about on this episode, listed below are some hyperlinks to get you began: