🗂️ How To Use NotebookLM?

Hey there!

Happy 2025! Hope you had a wonderful holiday break.

This year, we’re adding something new to our newsletter: prompting tips and reviews of recent generative AI products to share our knowledge with you, one step at a time. But don’t worry, our regular news coverage isn’t going away.

Here’s what we’ll cover today:

• How to Use NotebookLM: Your Personalized AI Resear …

  • Interface Overview

  • Key Features

  • Getting Started

• More Guides: ChatGPT Features

• Now, onto the News: AI Agents Everywhere!

How to Use NotebookLM: Your Personalized AI Research Assistant

Unlike general chatbots like ChatGPT or Gemini, NotebookLM is designed to work with uploaded information sources. It is called a “personalized AI research assistant " and provides specialized features not found in typical conversational models.

Interface Overview

• Sources: Upload up to 50 files (or more with extra capacity). This could be Google Drive documents, text files, web URLs, YouTube videos, or audio. You can enable or disable these sources when asking questions and even convert your Studio notes into reusable references.

• Chat: Summarizes your uploads, answers questions, and links directly to relevant sections so you can quickly explore large documents or multiple sources.

• Studio: Helps you generate audio overviews, timelines, FAQs, briefings, summaries, or study guides based on your sources. It also lets you create and format notes for easy organization.

Key Features

1. Studio

One of the standout features of NotebookLM is Studio. Click a button, and in seconds, NotebookLM generates polished documents that are easy to read and ready to share:

• Audio Overviews (Podcast):

NotebookLM offers a feature that creates audio summaries of your sources. With just a click on the 'Generate” button, you can obtain a downloadable audio file featuring AI-generated hosts discussing the material.

The tool also includes an “Interactive Mode,” which allows you to ask questions and guide the discussion in real-time. By providing simple instructions, this feature can be tailored to focus on specific topics or exclude unnecessary details.

• Document Generation:

With NotebookLM, creating structured outputs like timelines, FAQs, and summaries is straightforward. Upload a document, choose the desired format, and the system generates a concise, well-organized guide based on the key points in your material.

2. Handling Large Contexts

NotebookLM can work with documents containing up to 2 million tokens, which can simultaneously handle massive PDFs and multiple sources.

3. Multimodal Capabilities

NotebookLM can process and analyze PDFs, YouTube videos, and audio files. It can extract transcripts from videos and audio files and then provide summaries or highlight key points. This flexibility makes it adaptable to different types of content.

Getting Started

1. Create or New Notebook: Choose an upload method (PDF, link, etc.).

2. Once your document is processed, NotebookLM provides a quick summary and suggestions for interacting with it.

3. Use Chat for quick Q&A, Studio for document creation, and Audio Overviews for an AI-powered podcast experience.

We’ve put together a guide to help you get started with NotebookLM.

Check it out and drop your thoughts in the comments or poll below. Your feedback means a lot and helps us create better content for you!

More Guides: ChatGPT Features

Before 2024 ended, OpenAI released new ChatGPT features. We’ve covered each in separate guides so you’ll know exactly how to use them:

• How To Use ChatGPT Search?

• How To Use Projects in ChatGPT?

• How To Use ChatGPT's Advanced Voice Mode?

• How To Use Canvas Feature in ChatGPT?

Also, thanks for asking about prompt templates! We’ve compiled several that might help you get more out of generative AI at work or in everyday tasks:

• Language learning: 10 ChatGPT Prompts To Learn Any Language (Faster and Smarter)

• Teaching: 7 AI Prompts for Educators: Transform Teaching with ChatGPT, NotebookLM, Suno AI, and More

• We started a series on Chain-of-Thought prompting (stay tuned!): The Ultimate Guide to Chain of Thoughts (CoT): Part 1

Do you like this kind of content? Let us know!

Now, onto the News: AI Agents Everywhere!

1. Small Language Models → Small Agents (Hugging Face’s SmolAgents)
   Hugging Face’s SmolAgents drastically simplifies AI agent development. With just three lines of code and Hugging Face’s pre-trained model library, you can create lightweight, code-executing agents. Perfect for fast deployment without wrestling with complex environments.

2. Large Language Models → Large Action Models (Microsoft’s LAM)
   Microsoft’s LAM takes AI beyond text responses, enabling it to execute tasks autonomously. Through a four-phase training process (task breakdown, expert-led learning, trial-and-error, and reward-based refinement), LAM performs tasks faster than GPT-4, though it raises questions about security, ethics, and infrastructure needs.

3. NVIDIA’s Keynote: Agents & Cosmos
   In a CES marathon, Jensen Huang introduced Cosmos, a model trained on 20 million hours of dynamic video. NVIDIA’s vision is that AI will move from “generative” to “agentic,” acting as digital co-workers capable of handling logistics, design, and more. With new hardware (Blackwell chips) and software, NVIDIA is leading the charge toward AI that genuinely understands physical reality.

4. Anthropic’s Model Context Protocol
   Anthropic debuted a protocol allowing AI agents to connect safely to external data. While it promises “safety by design,” the real test will be how seamlessly these agents integrate into enterprise systems and whether they can avoid meltdown-level mistakes.

5. OpenAI’s AGI Workforce Agents
   Arriving in 2025, OpenAI’s AGI Workforce Agents aim to tackle everything from data analysis to business decisions. By combining advanced reasoning with iterative problem-solving, they could function as tireless, super-smart interns, though privacy, bias, and overreach concerns remain.

6. Google’s Deep Research Tool
   Launched alongside Gemini 2.0, Deep Research scours 50+ websites, compiles citations, and shows its plan before gathering data—like a virtual grad student on steroids. After a free trial, it costs $20/month. It faces competition from platforms like Perplexity, setting the stage for an “AI research wars” showdown.

7. Google Wants to Simulate the Universe
   DeepMind’s new project aims to build “world models” that master real-world physics. Led by Tim Brooks, it could eventually integrate with Gemini for advanced reasoning across robotics, autonomous vehicles, and hyper-realistic simulations. Think “The Matrix,” but with Google pulling the strings.

Reminder: Register for HackAPrompt 2.0! It Starts Soon

HackAPrompt 2.0, the largest AI Red Teaming competition ever:

• Up to $500,000 in prizes

• Open to everyone (no coding track available!)

• Collaborate with top minds in AI safety

It will be the biggest AI Red Teaming competition ever.

The launch dates are top secret, but we’re going live very soon. Be among the first to know—join the waitlist today!

Thanks for reading! We’d love your feedback to make this newsletter even better. Our goal is to share content that’s valuable and relevant to you. Help us fine-tune future editions by sharing your thoughts on this week’s email—it’ll only take a moment!