The Most Human AI Yet: Google NotebookLM

Every now and then, an AI tool comes along that blows your idea of what’s possible out of the water. This one is free, easy to use and it’s made by Google. Tick, tick, tick.

Just quickky, before we get into it - a bit of news from my side. I’m launching a 3 day intensive, super practical bootcamp called Build an AI-First Business next week. There are 100 spaces available, and the waiting list is already at 80 people - so get yourself on the list for first dibs.

Right, so what’s this tool then, Heather?

OK, so it’s called NotebookLM by Google. (Not LLM, just LM.) It’s an AI-first notebook, which sounds pretty boring, but I assure you, it’s definitely not. You have to see it in action to understand why it’s special.

It acts as a big AI-enhanced brain, taking knowledge you upload, analysing it and transforming it into lots of other different formats. This means we’re able to dig into information into a way we’ve never been able to before.

Humans learn in different ways. In fact, there are 7 different learning styles that are spoken about in education: visual, auditory, physical, verbal, logical, social and solitary. Despite this, most education is only presented in one format - a video, a podcast, a document to read. Well, now you can pick how you learn.

It couldn’t be easier to use, just two simple steps:

Step 1: Upload documents

It’ll take PDFs, .txt, audio and markdown files. You can also grab stuff from from your G Drive, add URLs, YouTube vids or simply paste text. Each source can contain up to 500,000 words or 200mb for uploaded files.

The idea is that you’re building a knowledge bank that you can draw upon. For example, imagine loading it with your marketing strategy documents, tone of voice, personas, example content, etc. Or perhaps loading it with a huge amount of information on a topic you need to wrap your head around properly.

Step 2: Create New Stuff

After a few minutes of upload, you can now interact with and create lots of useful new stuff.

• Q&A documents
• Study guides
• Table of contents
• Timelines
• Briefing documents

The info needed is intelligently gathered from the docs you’ve uploaded, analysed and then presented back to you in these new ways. I’m really impressed with how well it reads and seems to ‘understand’ the information.

It’s not just pre-defined documents either, you can interact and create whatever you need: scripts, idea generation, investor questions, whatever you want.

The most exciting bit (perhaps not the most useful, definitely the most exciting) bit is what comes next. Prepare to have your mind blown.

??? Notetaker Podcasts: the most human AI I’ve encountered yet

The most impressive way is the podcast conversation feature. This is the bit that’s got everyone talking (pun intended). I actually couldn’t believe my ears when I first heard one of these.

It takes your knowledge and creates a hyper realistic sounding podcast of two American people talking about the topic. They riff, they draw conclusions, they make comments and they even make jokes. It’s indistinguishable from a real life podcast. You’d never know it was AI.

Here’s a short sample of one I made this morning:

Can you believe that?!

• They’re not real people
• I didn’t give them any form of script
• In fact, the only info I uploaded to this Notebook was ONE short carousel. This one, in fact.

It created an entire 5 minute podcast show talking about AI safety. It fleshed out the key points from my doc, added its own research and made a perfectly plausible (and very educational) podcast.

?? Why’s it so impressive?

One of the main issues when using things like ChatGPT is hallucinations, i.e. the fact that generative AI sometimes makes stuff up. So you can’t 100% trust the results. Whereas NotebookLM uses what’s called RAG - meaning it prioritises certain pieces of information (what you upload), and cites everything so you can easily double check to make sure it’s all accurate.

If you’re interested in RAG, here’s a brilliant video by IBM explaining it:

?? Real Use Case: Let’s test it out

I decided to do a test with my 12 week Ultimate AI for Beginners course, which is broken into 15 video lessons with further reading, masterclasses, tasks, and quizzes.

To start, I chucked every file into Notebook to see what it would make of it - all the slides, tasks, URLs to articles. I was surprised to see this only took up a small chunk of its memory, less than 20%. So you can upload a LOT.

After about 4 minutes, it was all done, ready to go. It looked like this:

I created a study guide. Weirdly it appears in a small box, so here’s a chunk of it:

The rest of the file contains short-answer answers, really good quality essay questions, and a terminology guide. It’s pretty formally written and not in my non-techie, more conversational style but they’re definitely questions that mirror the final exam in the course.

I also created a timeline, which successfully pulled together a solid history of AI and tool release dates, though it did factor in random bits like when YouTube videos were published, they weren’t really relevant.

But then it came to the pièce de résistance…the podcast. Here it is, in full - for you to enjoy, and probably, learn from.

There’s a guide Google wrote themselves here (although I will say, ironically it’s quite poorly structured!).

Right, that’s quite enough of my yabbering for this week. I hope you found that helpful.

Ooh, before I go - most paid users of ChatGPT now have access to Advanced Voice Mode via the ChatGPT app. Go and have a play, it’s addictive!

Take care and until next week,

Heather