Is NotebookLM the ChatGPT Moment for Google?

The experimental product ✨ has been getting consumer attention trending on TikTok and Twitter in late September, 2024.

Photo by Claus Grünstäudl on Unsplash

Hey Everyone,

Google thinks it’s hard to go from information to insight and that NotebookLM can help us with that.

(☝—💎For a limited time get a Yearly sub for just $5 a month✨— ☝).

While Generative AI is not going to produce a personalized tutor or teacher anytime soon, it’s fun to imagine what it could become. The Google product that is now called NotebookLM was actually first launched in 2023 as Project Tailwind, and has since been rebranded and expanded and yet still somehow manages to feel like an experimental tool in a kind of purgatory realm of a preview.

Simon Willison calls NotebookLM ‘effectively an end-user customizable RAG product.’ I can see what he is getting at. It’s a bit like a B2C demo of the sorts of products that will soon become possible. What does this lead to? I think all of us can feel the potential of such products or appreciate how multimodal text-to-speech like this feels.

But is NotebookLM as a consumer phenomenon or reaction, its ChatGPT moment for Google DeepMind?

What surprised me is how much it was trending recently on places like Twitter/X recently, where NotebookLM and mostly the Audio Overview feature was getting a lot of seemingly organic shares in places moreover like TikTok in many different languages and places in the world this past week. I was listening to share after share about how wild or impressive this is.

Here is an example:

Consumers react to Audio Overviews on TikTok and Social Media

NotebookLM is Quite Powerful says Andrej Karpathy

Keep in mind it is still in its experimental phase. I don’t know if it enhances note-taking as much as it just has a novel take on summarization and synthetic content. What would a multi-modal RAG B2C product actually look like right?

One of the key product people for NotebookLM is Raiza Martin. On a serious note it’s not exactly clear what NotebookLM could be used for in alternative use cases. The best thing about NotebookLM at this stage in its development is its interface.

Google 🔨 Nails the NotebookLM’s Interface

While the performance leaves a lot to be desired, I do very much like the interface. Upon booting up a new Notebook you see your sources that you have uploaded on the left and presented with a very functional intuitive interface:

Experts and Authors are Involved

The editorial director of NotebookLM is Steven Johnson. You can ironically listen to his insights an actual podcast here.

Read the About Page of his Newsletter.

Steven Johnson

has his own Newsletter too, in fact he’s a famous author (14 books). The Verge quoted him on how he joined the project saying: “They reached out,” Johnson says when I ask how he got involved with Google, “and said, ‘hey, you’ve been dreaming of this ideal software tool that helps you organize your thoughts and helps you write and helps you formulate connections and brainstorm. We think we can do it now.” Johnson signed up, and has been at Google since the summer of 2022.

A Very Experimental Preview of NotebookLM (Notebook Language Models)

He made some odd tactical claims recently about NotebookLM on how to remember everything (like Quotes) you’ve read. We also know that more controls are coming to the generated podcasts. But let’s face it, they aren’t really podcasts, the Audio Overviews are just synthetic generalizations of fairly dubious quality. Let’s not pretend otherwise!

Still, you might over time find use cases in your studies, work or in your own research and content management to use the product. I actually find the quick text summaries NotebookLM can generative from a website, blog or YouTube come to be fairly good. Billed as a note-taking research assistant, Google Labs is trying to be useful. It certainly improves the functionality of Google Suite and Google Workspace as a whole. NotebookLM adds a bit of synthetic multimodal bling at least in terms of summarization.

At Least Google Labs is Trying 🧪

They call it a Research Tool for understanding things.

What can you do with NotebookLM?

I’ve been thinking more about this trending AI tool, and it’s practical applications.

Read our Guide.

How Good are Audio Overviews and is it really “Wild”?

1. Create Audio Overviews (podcasts) from YouTubes (this is actually potentially useful, although many sources are blocked)

2. Create Audio Overviews (podcasts) from documents, blogs or websites

3. Note Taking & Homework Synthesis: Document Upload and Analysis: Users can upload various types of documents, including PDFs, Google Docs, and websites. NotebookLM analyzes these documents to become an expert on the content, allowing it to provide tailored responses and insights based on the uploaded material.

4. Research Summarization: One of the primary functions of NotebookLM is to summarize complex documents quickly. This feature is particularly useful for students and professionals who need to digest large amounts of information efficiently.

5. Study Resource Creation: The tool can generate various educational resources such as study guides, FAQs, timelines, and briefing documents based on the uploaded content.

6. Shareable Projects: Notebooks created in NotebookLM can be shared with others, allowing for collaborative lesson planning or project work. Users can give others viewer or editor access to their notebooks. It’s not really collaborative yet though per se, but I see the potential.

Case Study - Creating an Audio Overview from a YouTube

I took a YouTube by MIT titled: MIT Professor on How AI & LLMs are Shaping Financial Advice, Analysis, & Risk Management: Part 1

NotebookLM Created an Instant YouTube Summary in text:

This video, featuring MIT Professor Andrew Lo, explores the potential impact of large language models (LLMs) on the financial industry. The conversation focuses on how LLMs can be used to analyze financial reports, identify market patterns, provide financial advice, assess risks, and even detect fraud. Professor Lo emphasizes the importance of human oversight and the need for rigorous ethical considerations to mitigate bias and ensure algorithmic transparency. He highlights the need for increased regulatory oversight and the development of legislation addressing data privacy and security to protect consumers in this evolving landscape.

Creating Audio Overview

It took Google’s tool over seven minutes to transform a 19 minute YouTube into a 12 minute Audio Overview.

• The Audio overview generalizes a lot from the macro perspective (hence “Overviews”)

• Introduces the YouTube expert fairly well

• The Overviews are dumbed down a bit too much (the product lacks an ‘expert’ analysis)

• A lot of the talking points were ultra generic (the outline of the transcript is poorly executed)

• The podcast connects the dots between cherry-picked parts of the uploaded the YouTube that might deviate from the original material or slightly misquote the expert out of context (which can be fairly damaging hallucinations)

• The Audio Overview in this case was not very useful in understanding or summarizing the key points or the transcript of the provided YouTube in my case study which was disappointing.

Not a supplement for the Original Source!

As it stands today NotebookLM doesn’t make me learn, think or brainstorm better contrary to Google’s emphasis that it might.

It’s a barely passable tool in experimental preview, but listening to the exaggerations on social media, you might imagine otherwise! Is social media hallucinating on how good Audio Overviews is? A lot of folk seem mind-blown by Audio Overviews. 🤯😲

Audio Overviews was an “Ah Ha” Moment for Consumers

While the product is very early stage, it has given some consumers an “ah-ha” moment it would appear. Like a five second ah-ha moment. Hard to know if that’s just them trying to game the X or TikTok algorithm or if they are being genuine. The TikTok mentions of NotebookLM sound on the whole, more genuine.

Someone who doesn’t know anything about RAG or multimodal LLMs, might find Audio Overviews impressive. Well, at first. The text-to-speech is at least smooth and the synthetic podcasts seem like they should be listenable. They ought to be, but they aren’t. Sort of like the half-baked demos of AI products and vaguely deceptive previews we have learned to expect from Google in 2023 and 2024.

That it was trending on social media is a bit hard to fathom considering how unfinished the product is, but I think audio overviews just touched on a nerve that AI could replace us, for example in a totally made-up podcast.

Google’s Half-Baked Level of Product Execution in AI Continues

Talking about Audio Overviews also just made a really good social share. Just like the paraphrasing tool Quillbot has different modes, hopefully Audio Overviews will soon get some badly needed ones as well. Google’s ability to execute outside of its money making products is always a little bit suspect, and this time it’s no different.

Anthropic gives me a Prompt-Library that’s useful and Google gives me a half-broken experimental preview of a product. That’s a problem if Google is serious about being a leader in Generative AI anytime soon. Sergey Brin is back at the office, but am I noticing any difference? On the API and developer side sure, on the consumer side, not so much.

But will a NotebookLM type product take off? Notebook Language Model is a fun tool, though I wouldn’t use it for any serious work or study yet. It’s certainly not something like ChatGPT that power users might be using many times a day. Nor do I see this easily evolving into a personalized EdTech platform that Google could spin out (as I had originally hoped).

I get why Google is trying to celebrate the hype, but it’s not an epiphany. Google’s reputation is already stained by many missteps, faked demos, Bard hallucinations and being out of touch with consumers and not executing very well in accuracy in the realm of hallucinations. I have managed to make Audio overviews make some pretty jaring mistakes in this regard. Jailbreaking NotebookLM is not the point. Google just got actual credit on social media for an unfinished product which does sound like ChatGPT in November, 2022.

Google is a Search Ads company with some elements of AI. It’s not a Generative AI company or innovator per se. There is a reason why they invested in Anthropic (announced about one year ago). So none of the facts have really changed here. For the product to still feel like a very experimental preview demonstrates just how hard these things are to turn into real products. As a certain point you just have to release it, re-launch it, change the product name, and hope it sticks. If it goes viral on social media, is that good?

It’s not as if they haven’t had time to iron out the kinks in NotebookLM. They have been working on this for a long time (go to 1:30 of this video).

AI Product Innovation in 2024

You can see the LinkedIn version of this poll here .

Is Google Keeping up in Generative AI Products?

AI Products will be a Marathon, not a Sprint 😬

As Anthropic becomes more product centric it might begin to really overtake Google and OpenAI in AI products in 2025 according to how momentum currently stands in September, 2024. Take what you can from OpenAI’s recent dev day: DevDay in 5 Minutes: 4 Major API Updates. OpenAI’s advanced voice mode API is indeed expensive. For a slightly longer version of DevDay go here.

While Meta Platforms is making smart glasses, Google only makes demos (Project Astra) with smart glasses obscurely present in the demo. Meta is likely going to train on all the images we take on their ‘Orion’ smart (AR) glasses hardware product whether you like it or not. Fun times.

Meanwhile, Microsoft is turning Copilot into an AI presenter to read you the news is pretty cringe (like they do already in China), or how Microsoft trains on our data. Its Recall product is downright dystopian. All of these companies have their internal efforts to fake it with Generative AI. But how many are innovating and how many are just reacting and building products that will never be fully realized?

Some Tools and Topics we have Covered:

• Read our guide to Perplexity

• Read our guide to Google’s NotebookLM and Audio Overviews.

• Read some of our AI Tool rundowns here, here and here, along with Prompt Engineering here.

• Read our list of who to follow in AI.

• Read our list of AI apps to try and what comes next.

You can just try to create an Audio Overview for yourself and see what you think.

• You can learn more about Google Labs here. Thanks for reading!

• NotebookLM is better than ever, but the verdict is it still has a long ways to go!