This morning I decided to take Google’s new Notebook LLM experiment for a spin, specifically the ‘Studio’ feature, which will generate a podcast style recording from your inputs.
I fed it one of my own blog posts as the source and hit “generate”. I used the recent article about the EasyList as input. It’s a complicated topic with a lot of points so it felt like a pretty good challenge.
Listen to the full recorded discussion here: Google LM Recording (takes a few minutes to load)
Here is the original blog post: The EasyList, AdBlockers and Affiliate Marketing
NotebookLM read and understood the blog and recorded a 30 minute discussion about all the key points in the article. To my surprise, the output was much better than I expected. It felt like a lively chat between two peers rather than a cold machine reading back my own words. It covered everything I wanted to get across, and even added a few nice twists to help get the message across more effectively.
For the most part the output was impressive. About ninety percent of the discussion landed exactly where I wanted, sometimes even rephrasing my thoughts more clearly than I could. The flow was smooth, the transitions logical, and the recorded audio captured a natural back-and-forth. The ‘two person conversation’ format works very well – I would have liked to try a ‘single person’ option, but I can see how the discussion approach makes this possible.
I didn’t give it any special ‘instructions’, but you can ask the tool to concentrate on specific aspects or adopt a specific approach. Next time I would probably specify who the audience is for – it flip-flopped between affiliate publishers and merchants a little. In fact, I intend to do just that, as soon as my daily recording allowance resets.
For the test recording I just used one source to make it easy to do a direct comparison between the post and the output, but you can also add and combine multiple ‘sources’, which seems super powerful and possibly a real game changer. There’s also a research tool to find new notes, which can also be converted to sources.
The tool isn’t just for making recordings – that isn’t even its main feature. It is a novel approach to gathering research and constructing AI insights based on that research. Unlike ChatGPT it makes citations and references really easy to confirm, and it’s approach means hallucinations are less likely.
- Customization is basic. You can tweak prompts, but there’s no deep “personality” or tone control just yet.
- Usage caps apply. The free tier enforces a strict daily limit, so if you did want to use it for something professional you wold have to pay for the ‘plus’ plan.
- Minor hiccups occur. All the facts were as I had written them, but occasionally small points were given a bit too much priority.
- No British voice. In fact, no voice customisation at all without paying for the Plus plan.
More important, this is a shiny new AI tool that can generate quality content for free. You can guarantee everyone will jump on the bandwagon in a heartbeat. Expect a flood of low-quality, generic podcasts to swamp the internet very soon. Even if yours are not spam, these recordings might not be good for your reputation, regardless of the quality. Let’s wait and see.
So, would I deploy output from Notebook LLM “in anger”? Not just yet (possibly never). The toolset isn’t quite complete enough yet (British voices, please!) and I worry about how it will be perceived when the world and his dog jump onboard. On the positive side, it’s generative AI content, but only trained on my own articles. As long as it was my own work being used as the source I would be less worried about the ethical side of things.
However, as a brainstorming partner or first-draft generator, it already feels remarkably capable. When I wrote my first book I did think of recording a series of podcasts about some of the key topics, and I’ll definitely be using it to make a first draft and see how certain ideas sound. I look forward to future updates.