I recently learned about NotebookLM, a personalized AI research assistant powered by Google's Gemini 1.5 Pro. In this post I will show how I used NotebookLM to generate the audio file that you can listen to by playing the following audio control.

What is NotebookLM?

Have you ever wanted to research a topic without spending hours pouring over dozens of websites, PDFs, videos and other documents? Now you can using NotebookLM. The basic idea behind NotebookLM is that you supply it with the information you want it to "learn" and then you can ask NotebookLM to summarise the information or answer questions about it.

Getting Started

The following web page is used to provide NotebookLM the content with which it is to be trained. 

As you can see, content can include:

1. Uploaded documents (text files, PDFs, Markdown and MP3 Audio files).
2. Documents from Google Drive.
3. Links to web pages and YouTube Videos.
4. The Clipboard.

There is a limit of 50 different content sources with up to 500,000 words each.

GenHelm has a feature whereby any web page can be rendered as text, so I used this to create a series of text files from GenHelm's marketing site and product help pages. After uploading the content you can ask NotebookLM to create Study Guides, FAQ pages or other content summaries. You can also just ask NotebookLM questions and it will attempt to answer them while referencing the supplied content.

By far the most impressive feature of NotebookLM is its ability to create a "Deep Dive" conversational audio clip like the one linked to above. I am really blown away by how it is virtually impossible to tell that this was AI generated and not produced in a studio with live actors (or GenHelm experts). To generate the Deep Dive audio, you just click on the Generate link shown below. Unfortunately, you can't tell it how long the clip should be or what to focus on. 

The Deep Dive is mostly accurate, however, some features and capabilities are overstated. The most frustrating limitation was that I could not get it to pronounce "GenHelm" correctly. The audio pronounces it using a hard "G" as in "get" rather than a soft "G" as in "generate". I tried addressing this by including a file that indicates that GenHelm is pronounced "JenHelm" but this did not correct the problem.

In summary, I think NotebookLM is an amazing tool and can be very useful for quickly getting up to speed on any topic for which you can supply NotebookLM the content it needs. Anyone looking to add a "Deep Dive" audio to their website should give NotebookLM a try.