Testing Google’s NotebookLM with RAJAR

Testing Google’s NotebookLM with RAJAR

Google has a recent-ish AI tool called NotebookLM. The idea is that you throw a lot of information into it, and it can sift through it, answer questions based on what you’ve told it and so on. Recently they added the ability to auto-generate a podcast!

So I decided to test it. I gave it precisely one source for RAJAR Q2 2024 data – my own site, and specifically the report I wrote based on that data. Then it’s just a question of pressing a button, waiting a few minutes and a “podcast” being auto-generated.

It came back with an 11:46 podcast which was generated in the form of an informal chat between a US-accented man and woman. Here are the first couple of minutes to give you a flavour of what it came up with:

Aside from the interesting pronunciation of “RAJAR” (“Ry-jar”) I would say that this isn’t bad. I get acknowledged as the source of the analysis, and the tone is very relaxed. That’s not inappropriate for this subject, but perhaps it might be for something else. Maybe if I was writing about likely outcomes of cancer treatments, the tone could be inappropriate.

And there’s definitely some stre-e-e-e-t-t-t-c-c-c-h-h-h-ing going on. The “woman” repeats what the man says for example. But the AI is smart enough to know that RAJAR is an industry standard (not something I necessarily explained on that page). Q2 2024 is summarised as “summer” and so on.

To be honest, I’m not sure how this would best be used. I think you could quickly read my full article faster than you could listen to the auto-generated podcast. But perhaps if it were a subject with lots of articles, then this form of summary might work.

As I say, the tone is very relaxed and I’d like a few options there. Similarly, British voices might sound more convincing talking about British radio.

And there’s part of me that wonders really what the point of AI generated podcasts is?

If you can’t be bothered to make something yourself, why should I listen to it? Perhaps I should point my AI-listener towards your AI-podcast and it can write me a two-line summary. What have we achieved at that point aside from consume a lot of AI processing power?

But that might be overly negative. If I’m cramming for a history exam and can throw some on-line articles about The Zinoviev Letter then this might be a quick way to summarise that information for me (That was a subject we studied in history at school as a friend reminded me recently).

Definitely worth a test…

Featured image above auto-generated via Adobe AI with the prompt: “A mixture of soundwaves and documents floating in an empty chamber”


Posted

in

,

Tags:

Comments

2 responses to “Testing Google’s NotebookLM with RAJAR”

  1. Steve Martin avatar
    Steve Martin

    Spookily real. Definitely a use case here for people who have low levels of literacy.

  2. Di Hope avatar

    I feel its not about speed – of content assimilation, or production. But human interactions – including all those that take place during production – especially with guests. Surely that’s why we listen to radio features & podcasts. If we just want to hear audio then why make a pretend conversation. Why not have a narrator – where possible with a voice actor to impart feeling? And how many users of this will be as scrupulous as you Adam? It opens a Pandora’s box of fake content creation – many listeners won’t even realize this is ‘bots’ talking – such content MUST legally required to be flagged going forward, so at least people know what they’re listening . I’ve posted a piece on this and another example of it on my LinkedIn page: https://www.linkedin.com/feed/update/urn:li:activity:7242979307386499072/