Behind the product: NotebookLM | Raiza Martin (Senior Product Manager, AI @ Google Labs)

Oct 10, 2024 Episode Page ↗
Overview

Ryza Martin, Product Lead for Google's NotebookLM, details the AI product's journey from a 20% project to a successful tool. She shares how its small team operates like a startup, surprising use cases, and the vision for AI-powered content creation.

At a Glance
22 Insights
48m 58s Duration
12 Topics
5 Concepts

Deep Dive Analysis

Introduction to NotebookLM's Audio Overviews

The Genesis and Small Team Behind NotebookLM

Development of the Audio Overviews Feature

Google Labs' Technology-First Product Development Approach

The 'Content Studio' and Audio Model for Podcast Generation

Surprising and Hilarious Use Cases of Audio Overviews

Operating with a Startup Culture within Google Labs

NotebookLM's User Growth and Changing Demographics

Future Vision: AI Editor Surface and Mobile Experience

The Role and Influence of Steven Johnson on the Team

Addressing AI Misuse and Red Teaming at Google

Future Directions and User Feedback for NotebookLM

Audio Overviews

A feature within NotebookLM that generates an AI-powered audio summary, often in a podcast-like format, from any uploaded source material like URLs or documents. It allows users to consume information via voice modality rather than just text.

Google Labs Operating Model

A distinct approach within Google, characterized by a focus on starting with technology rather than a problem, moving fast with fewer processes, and fostering a startup-like environment. It allows teams to iterate quickly and build in public, often with small teams.

Content Studio

The 'secret sauce' behind NotebookLM's ability to generate high-quality, engaging audio outputs. It's an internal system designed to take an opinionated approach to content, making it relatable and interesting, and powers features like the Notebook Guide and audio generation.

AI Editor Surface

Raiza Martin's long-term vision for NotebookLM, where users can have an AI interface that allows them to fully remix any input (video, audio, emails, social media) into any desired output (blog post, tutorial video, chatbot). It aims to make information malleable across different mediums.

Red Teaming AI

A critical process at Google involving dedicated teams that rigorously test AI products for potential misuse, safety concerns, and unexpected behaviors. This helps identify and address vulnerabilities before or after launch, continuously adding to test cases.

?
What is NotebookLM and its Audio Overview feature?

NotebookLM is an AI-powered research tool from Google Labs that generates AI-generated audio overviews, often in a podcast format, from any uploaded source material like documents or URLs, allowing users to interact with content in a new modality.

?
How did NotebookLM originate and what was its initial team size?

NotebookLM started as a '20% project' within Google Labs, initially led by Raiza Martin with one engineer and later joined by Steven Johnson. It grew from a small project focused on interacting with content using LLMs.

?
How does Google Labs operate differently from other parts of Google?

Google Labs prioritizes starting with technology and finding applications, operates with fewer processes, moves faster, and fosters a startup-like environment. This allows teams to build and ship products more quickly and publicly.

?
What technology powers NotebookLM's audio feature and makes it sound so good?

NotebookLM uses Gemini 1.5 Pro as its base model, combined with a powerful voice/audio model. The 'Content Studio' is the key component that helps shape the audio output to be engaging and relatable, based on extensive listening and refinement.

?
What are some surprising or unusual use cases for NotebookLM's Audio Overviews?

Users have uploaded their resumes or performance review notes to hear AI hosts 'hype them up,' created multi-episode podcast series from Wikipedia articles (like 'Histories of Mysteries'), and even generated insightful analyses from documents containing only repeated words like 'poop' and 'fart' or 'chicken'.

?
What is the long-term vision for NotebookLM?

The long-term vision is an 'AI editor surface' that allows users to fully remix any input (video, audio, text) into any desired output format (blog post, video, chatbot). More tactically, the team is focused on developing a mobile experience and more magical, delightful control over outputs.

?
What is Steven Johnson's role on the NotebookLM team?

Steven Johnson, a New York Times best-selling author and journalist, acts as a key partner and 'product' for the team. Raiza observed his unique research and knowledge de-densification workflows to inform how the technology could empower everyday users.

?
How does Google address potential misuse and ethical concerns with AI like NotebookLM?

Google employs extensive red teaming efforts, with large teams dedicated to testing for as many potential unsafe scenarios as possible. They continuously add to test cases and would pull back features if they were deemed unsafe, while also observing public reaction to unexpected outputs.

1. Observe Users for Insights

To gain crucial product insights, spend meaningful time observing users, watching them perform tasks, and discussing their feelings and experiences, regularly and intentionally.

2. Adopt Startup Operating Model

To foster innovation and move fast within a larger organization, create an environment with fewer processes, where product managers, engineers, and designers collaborate closely and implement simultaneously.

3. Shape Technology for Users

Actively shape and refine technology to make it more accessible and understandable for people, aiming for a “wow, I get it” moment.

4. Deepen User Experience

When a product gains traction, prioritize deepening the user experience and engagement in parallel with commercialization efforts to build on initial excitement.

5. Provide Malleable Content Formats

Empower users to choose their preferred content format (audio, text, video) based on their mood or context, shifting the dynamic between the person and the knowledge they receive.

6. Envision AI Editor Surface

Imagine a future where an AI editor surface allows users to remix any input (video, audio, text) into any output (blog post, tutorial video, chatbot), transforming how content is created and consumed.

7. Introduce Voice Modality

Introduce voice input and output into technology to change user interaction, improve feelings about the technology, and affect real-time thinking, making the experience more valuable and fun.

8. Design Magical Controls

When adding user controls (like “knobs” or “sliders”), prioritize designing them to feel magical and delightful, rather than just functional, to maintain the product’s unique appeal.

9. Continuously Learn from Users

Actively seek and read user feedback daily on platforms like X or Discord to continuously learn and improve the product, aiming to build the best possible experience.

10. Red Team Products for Safety

Establish dedicated teams to “red team” products, rigorously testing for potential unsafe scenarios across as many areas as imaginable to ensure safety and identify vulnerabilities.

11. Q&A Over Long Docs

When presented with long, dense documents, consider engaging in a direct Q&A session with the expert or author, as it can be a more efficient way to extract information than reading the entire document.

12. Build in Public with Discord

Engage with users and build a community by establishing a Discord server, especially when building a new product, to gather feedback and foster excitement.

13. Prioritize Mobile Experience

Identify and address the gap in mobile experience for products, as it represents the next horizon for development and can significantly change how users interact.

14. Start with Technology (Labs)

When working in a lab environment, start product development by exploring the technology first and then identify its practical applications, rather than beginning with a problem.

15. Generate Audio Overviews

Use NotebookLM to generate AI-powered audio overviews by simply providing a URL or uploading a document, which can be a surprising and delightful way to consume information.

16. Boost Confidence with AI

Upload your resume or performance review notes to NotebookLM to generate an audio overview, which can boost your confidence by hearing AI hosts enthusiastically describe your achievements.

17. Convert Study Materials to Audio

Students can use NotebookLM to convert their study materials into audio guides, making it easier to learn and review content.

18. Stay Curious and Keep Learning

Continuously listen and remain curious to learn and grow, as the podcast promises to keep diving deep and bringing more content.

19. Subscribe to Podcasts

Subscribe and follow podcasts in your favorite app or YouTube to avoid missing future episodes and to support the podcast.

20. Embed Analytics with Xplo

Use Xplo, a fully white-labeled embedded analytics solution, to build and embed dashboards and reports into your application quickly, connecting to any relational database or warehouse.

21. Improve Product with Sprig

Use Sprig, a product experience platform, to capture real-time product experience data, analyze it with AI, and get actionable recommendations to drive revenue, retention, and user satisfaction.

22. Join Sidebar for Peer Growth

For senior tech professionals, join Sidebar to be matched into peer groups for unbiased opinions, diverse perspectives, and raw feedback to advance your career.

I feel like the way that I've always built products is I start from a problem and I think about the solutions you know to solve it solve it for people in a meaningful way and uh in labs in particular we start with the technology.

Raiza Martin

A lot of technology I feel like you have to shape it and bring it closer to people and I think it's like such an interesting thing to iterate on to be like what is the shape right like soon soon you'll get there like if you keep going at it you'll eventually I think land on something that like when people look at it they're like wow I get it right and that's that's always like what we're hunting for.

Raiza Martin

I told him this I was like Steven I think you're the product. I think it's you and I'm going to follow you around I'm going to watch everything that you do and we're going to try to figure out how we use technology to build it.

Raiza Martin

I think there's something really special about you don't know what's going to happen right you just click generate and then you have this like this these two people just hyping you up.

Raiza Martin

I mean please keep trying it try it share your feedback whether you find it useful not useful you think it's annoying tell me these are the things that I've heard you know for the last year that's helped me to make it better.

Raiza Martin
3
Number of engineers when Project Tailwind (NotebookLM's initial announcement) was made Excluding Raiza Martin, a designer, and Steven Johnson.
Approximately 8
Number of engineers until recently (last month) Before significant expansion for future roadmap.
Approximately 3 years
Age of Google Labs Relatively new organization within Google.
Over 60,000
Number of users in NotebookLM's Discord server As of the time of the podcast recording.
14
Number of books written by Steven Johnson He is a New York Times best-selling author and speaker.