Imagine having a personalised AI collaborator at your side who helps you do your best thinking. After sharing your documents with your personal AI, it becomes an instant expert in those sources so you can read, take notes, and collaborate with it to refine and organise your ideas.
This has been the core functionality behind Google’s NotebookLM, an AI-powered note-taking and research assistant, since it launched in early access in July 2023. NotebookLM is designed to help you synthesise information from your own notes and documents. Integrating a language model (hence the “LM” in NotebookLM), the assistant aims to provide personalised assistance in understanding, summarising, and generating insights from user-provided content.
NotebookLM combines the capabilities of Google’s most advanced large language model, Gemini 1.5 Pro, with a note-taking interface, enabling you to interact with and gain insights from your documents. There’s a great deal to unpack in what NotebookLM can do for you, so it’s worth spending some time to review in depth what Google has created.
In this post, I want to talk about a new feature in NotebookLM that captures the imagination, a feature that elevates its functionality far beyond creating simple text summaries.
The Remarkable ‘Audio Overview’
NotebookLM can create AI personas that can discuss a topic you’re interested in by having a conversation about it, the recording of which is then available in an audio format that makes it very easy to create a podcast.
I tried the new ‘Audio Overview’ feature Google has enabled in NotebookLM, which is just one in a range of updates and new features.
Frankly, my jaw dropped when I listened to the audio recording that was created. I expected to hear a conversation that would most likely be discernable as something created by artificial intelligence following some kind of script, no matter how credible the generated voices sounded.
Instead, what I got was something that sounded as far removed from AI as it’s possible to go. The conversation sounded unquestionably like that of two actual people in unscripted, natural, discourse, just like you and I having a chat.
The seven-minute audio recording is embedded below. But before you jump to it, let me briefly explain what I did and how.
Last week, I published a post on this blog titled “It’s Time To Be Honest About Security Alerts.” The post discusses how companies like Microsoft and Spotify need to take a more proactive approach to security alerts. The post criticises the dismissive tone of current security alerts, which often attribute multiple login attempts to innocent typos. I argued that these alerts downplay the very real possibility of a malicious attack and instead should acknowledge the seriousness of the situation.
I shared this post with NotebookLM and selected the “Audio Overview” option for the AI assistant to create a podcast-style discussion with simulated voices about the content of the post.
Here’s how that works:
- AI Personas: NotebookLM created two AI personas, a man and a woman, called “hosts”. These hosts have distinct voices and conversational styles, contributing to a dynamic and engaging dialogue. Currently, it’s available in US English.
- Content Analysis: Using advanced language understanding, NotebookLM analyses the shared document, identifying key themes, arguments, and supporting evidence.
- Dialogue Generation: It then constructs a conversation between the two AI hosts where they discuss these key points, raise questions, express opinions, and challenge each other’s viewpoints.
- Natural Language Generation: The conversation is generated using natural language processing techniques, aiming for a fluent, coherent, and engaging discussion that closely resembles a real-life conversation between two people.
- Podcast-Style Format: The final output is delivered in a WAV uncompressed audio file in a podcast-style format, complete with natural-sounding voices and even subtle pauses and intonations, enhancing the illusion of a genuine human conversation.
I’ve taken the original seven-minute WAV audio and created an MP4 video to make it available on YouTube to embed here and share. Take a listen and see for yourself what this NotebookLM tool is capable of in this simple, straightforward example:
(If you don’t see the YouTube embed above, listen to it on YouTube.)
Mind-blowing, wouldn’t you say?
This is just the Audio Overview recording: if I’d added an intro and outro, some branding, and perhaps some theme music and saved it as an MP3 file, I’d have a podcast episode of under ten minutes. Adding some further narrative could get it to 15 minutes or more. (That’s in the area of what Shel and I record for the short-form For Immediate Release podcast episodes we create. We could have fun with that idea!)
Real World Examples
I’ve been thinking of ways organisations can embrace NotebookLM and see measurable benefits from doing so. There are so many examples I can think of!
Let me continue the focus on audio with these two examples:
1. Marketing Content Creation
Scenario: A marketing team needs to create content quickly for different platforms, like blog posts, podcasts, or social media summaries, from a single source, such as a product whitepaper or case study.
How NotebookLM Helps: Using NotebookLM, the team could upload a lengthy whitepaper and generate an audio conversation between AI personas discussing the product’s key features and market impact. This podcast-style discussion can then be published on company websites and social media platforms as original content, providing a new, engaging way to reach audiences who prefer audio content.
- Example: A B2B SaaS company uploads a detailed whitepaper on a new software feature and generates a conversational AI podcast where two AI hosts explain the feature in layman’s terms, making it easier for potential clients to grasp its benefits.
- Benefit: The audio format is quicker to produce, broadens audience reach (targeting listeners as well as readers), and positions the company as innovative by using cutting-edge AI tools for content creation.
2. Internal Communication and Training
Scenario: A large company frequently distributes lengthy policy updates, compliance instructions, or training manuals to employees, often resulting in low engagement.
How NotebookLM Helps: By uploading these documents to NotebookLM, a company could generate an AI-driven audio discussion that distills key points from complex documents into a podcast-style format. The AI personas could engage in a conversation about the document’s most important sections, turning the static material into an engaging dialogue. Employees could listen to this while commuting, allowing them to absorb essential information more easily.
- Example: A human resources department could upload a new code of conduct document and create an AI-generated podcast where two AI personas discuss the major updates and implications for employees.
- Benefit: This format could cater to employees who prefer learning via audio, making it easier to ensure compliance and understanding across teams.
On the Cusp of a New Horizon
With NotebookLM, we’ve had a glimpse of how AI can really change the world with the implementation of generative AI audio technology like this. I’m filled with excitement and fear, all at the same time!
I’m excited at the possibilities this technology offers in a more immersive and engaging way to learn from documents, catering to auditory learners and those who prefer a conversational approach to information consumption.
It makes complex information more accessible by presenting it in a conversational format, potentially benefiting individuals with reading difficulties or those who prefer audio content.
The natural back-and-forth between the AI hosts can stimulate critical thinking, encouraging users to consider different perspectives and engage more deeply with the material. I was astounded when I first listened to the hosts’ conversation extending into new areas beyond what I originally wrote.
Just like humans would do.
But there are concerns.
Despite advances like this, large language models can sometimes generate inaccurate or misleading information, especially when dealing with complex or nuanced topics. Note the disclaimer at the foot of every NotebookLM screen: “NotebookLM may still sometimes give inaccurate responses, so you may want to confirm any facts independently.”
You need to check the accuracy of any audio, just as you do now with content generated when using chatbots like ChatGPT. There isn’t yet the capability to download a transcript of the output from Audio Overviews that would aid the verification process, but Google says it’s on their to-do list.
Privacy and data security should be front of mind regarding how user data and uploaded documents are handled and stored, particularly for sensitive information. This is true of all AI-generated content activity.
As AI tools become more advanced, concerns about their impact on jobs and the workforce are inevitable. Businesses must navigate the ethical considerations of content creation. And ensuring that AI-generated content is fact-checked and free from bias will be crucial to maintaining trust with all stakeholders.
Meanwhile, take advantage of the availability of NotebookLM to explore the full capabilities of this amazing tool. See how easy it is to generate an audio podcast. And brace yourselves for a flood of AI-generated podcasts in the coming weeks and months as people experiment and share their outputs and thoughts. The price of accessibility!
Game-Changing Innovation
Google NotebookLM is pushing the boundaries of AI by transforming how we interact with our own content. Whether you’re looking to streamline internal communication, enhance customer engagement, or create compelling marketing content, this tool offers powerful new possibilities.
The key will be balancing innovation with caution, especially regarding privacy, ethics and accuracy concerns.
How do you see the role of AI-generated audio in your business?
Related Reading:
- The Rise of Shadow AI is a Double-Edged Sword for Corporate Innovation (25 July 2024)
- Not Using AI Could Become a Matter of Negligence (30 April 2024)
- A Balanced Approach to AI Integration: Lessons From Klarna’s Customer Service Success (4 March 2024)
(The image at top was created with Adobe Firefly.)
