Content creation with artificial intelligence is already old hat as it’s been going on for a few years and, unfortunately, slop is now populating the Internet at an increasing pace. Yet, when I received this message from a good friend of mine last week regarding Google’s new app entitled NotebookLM, I was shellshocked. I tried it and tested it and felt immediately overwhelmed. Having slept over it for a few days, I’m just recovering, so here are my impressions.
NotebookLM by Google: Artificial Voices, Real Concerns
The other day, a friend of mine sent me a message about Google’s new NotebookLM AI application. As I always do that kind of thing, I tried and tested it immediately. The catchphrase for NotebookLM is “Think Smarter, Not Harder” and Google presents it as “The ultimate tool for understanding the information that matters most to you, built with Gemini 1.5”.
NotebookLM: made to ‘understand’ information?
I wonder about that. Is it really a tool made to “understand information”? This sounds a bit dubious.
The aim of NotebookLM is to turn a piece of text, a video, a web link into a conversational podcast. It is semi customisable and not quite finished. But it gives you an idea of what the future has in store for us, content creators.
On the one hand, the technology is great and works fine, barring a few glitches, on the other hand, it’s a window on a very weird and dark future (once again, it’s not the tool that is the problem but the people using the tool, as Bradbury remarked).
For my initial test, I selected one of my English pieces about artificial intelligence (AGI). I copied and pasted my text into the window and hey presto! a few seconds later, a proper conversational podcast between an American man and woman was available. Here it is
I must admit I couldn’t believe my ears when I heard this podcast produced from a mere piece of text. It was both brilliant and daunting. I immediately thought that anybody could produce an audio conversation out of anybody’s blog piece and I suppose that some of the laziest of content creators will do just that.
When I looked into the podcast in greater detail, I spotted that there were a few glitches here and there and especially the quote by Ray Bradbury which is definitely not taken from Fahrenheit 451. It’s clearly mentioned in my text.
Man | 01:24.308
It’s like that line from Fahrenheit 451. I’m not afraid of robots. I’m afraid of people, people, people.Woman | 01:28.691
Yeah.
Well… nope, sorry (and by the way, I hate your “yeahing” at me). This was taken from Bradbury’s 1974 letter to Brian Sibley. It’s clearly stated and the link to the source file is explicit.
One more test
This very morning, I went back to the application and tried it once more. I inserted a YouTube video and it failed a couple of times. So I gave up and copied a web URL and it worked wonders. This time I used my fraud and AI piece with Fujitsu.
I tried to customise the podcast but I couldn’t change the American voices nor the tone of voice which isn’t consistent with mine. I tried to turn it into a more professional, less casual, tone of voice. This didn’t work as planned. But my instructions aimed at making the podcast more factual and to focus on the numbers were executed correctly by the AI.
That said, when you listen to the entire podcast, especially towards the end you will realise that the AI is adding quite a lot of content to it and making its own commentary.
A few worrying signs
This experiment raises quite a few questions.
- To start with, a lazy content writer could start publishing its own podcast channel from scratch using other content creators’ content without even mentioning their names;
- Second, hallucinations are still part of this equation and they are quite wicked and hidden and hard to track. Once again, if you are a lazy content writer, then it doesn’t matter at all. On the contrary, if you are a conscientious content creator then it all makes the difference.
- The fact that this AI app is adding content is another kind of hallucination even though the text is perfectly sensible compared to the original piece. But I didn’t write that, I didn’t think that and I’m not even sure I want to add it. This is not only annoying, but downright worrying.
- Last but not least, the casual tone of voice used by the application and the voices and accents that aren’t customisable yet are a showstopper as far as I’m concerned. But I suspect that this can be easily corrected.
In conclusion
As I mentioned already, I find this kind of application a little worrying. On the one hand, it is great to be able to produce a conversational podcast which is very engaging, very quickly. In hindsight, this is probably conducive to producing even more slop on the Internet, which will probably end up collapsing in on itself. It’s a matter of years if no one stops this nonsense.
I’m not sure that Google will maintain this application nor that it will even let you use it for very long. They have a track record of crucifying innovations.
It could be quite tempting to use NoteboookLM to produce conversational podcasts without any efforts. But I will never use because we have this 100% human content commitment on visionary marketing. I may end up being the last of the Mohicans in that concern but I intend to keep the upper hand in this content creation process. I’m the one doing the thinking here, not the tool I’m using to write it.
Transcripts of both NotebookLM podcasts
The post NotebookLM by Google: Artificial Voices, Real Concerns appeared first on Marketing and Innovation.