Our Director of Marketing & Communications uploaded our athletic director’s recent sports newsletter to the NotebookLM AI website and had it produce a podcast from the newsletter material. This link is that podcast.
There are no humans in this podcast and no editing was done.
I’m astounded by how realistic this sounds. Bad actors can use AI technology to mislead whole populations of people. As this technology becomes more capable, its potential for good and evil will increase exponentially.
I think there will be entire websites and channels for AI in the future. People who use social media are already experiencing most of the content being AI generated.
I’ve stopped using the internet as a source of information since the advent of GenAI, as I don’t have any faith in the information on it anymore. I’ve returned to newspapers, books and magazines that I trust exclusively, and as more AI generated content comes along I think people will cherish human-produced content.
I would have been fooled by this podcast. The AI makes it sound astonishingly real with imperfections in speech added in, such as stuttering and “umms” and pauses. Crazy.
I agree. “William Randolph Hearst, along with Joseph Pulitzer, is often credited with helping to lead the United States into the Spanish-American War through their use of yellow journalism.”
AI will make doing the same thing much easier. And people will continue to believe what they want to believe.
I heard them showcasing the service on the last ATP, it was a little scary and also a little exciting, but like all AI, it has the usual flaws with accuracy.
I threw in my State EMS Protocols and going to see what it spits out. If it makes going over the protocols entertaining, and accurate, it could be a good way to get people within the department to actually go over them.
I’ve been experimenting with NotebookML a week ago before it suddenly “blew up” on the Internet.
Here’s a subtle take:
Upload text of blogs, articles, or have it scan a YouTube video (it will generate the transcription to scan automatically).
Then use the created podcast as a “personal audio summary” when commuting, jogging, etc. to get the information delivered in a way that doesn’t put you to sleep and you can listen while you are doing something else.
Feed your own writing into it. Listen to the generated podcast and see which points got highlighted and which got ignored or only briefly mentioned.
Use it as an editor to tweak/improve your writing. The audio format makes it just different enough than proof-reading your own writing that you can spot things normal review isn’t catching?
I do something similar. I highlight text and then use text to speech to proof. I discover typos and other errors with this process that I miss doing normal editing. I’m not sure how the podcast approach would work, but it may be worth a try.
I’ve been using it to create personal podcasts based on sources I’m reviewing. It is impressive, and often draws out common and contrasting ideas from the sources in an entertaining manner. Other NotebookLM tools are useful to have… particularly asking questions of the sources which improves quality enormously vs training data.
Sadly the Reddit channel for NotebookLM seems to have several people thinking they’re doing the world a service by creating YouTube channels stuffed with these “podcasts.”
The improvements in text to speech technology have been staggering. Omnivore would read out articles to me using some of the most natural OpenAI voices. They recently stopped offering that, perhaps due to cost, but it was genuinely impressive and pleasant to listen to. There’s huge potential to create audiobooks from written books for niche interests that would never justify paying a human reader. I dare say radio stations may use the technology to dynamically insert hyper-local content (such as weather) into online streams.
