I agree with you on why Google didn't (bother to?) market it. I doubt it was out of fear of misuse - had this been a real concern they wouldn't have released all of it to everyone.
I see your goosebumps and I raise my own... not sure if it's from a state of awe, or fear. Probably the latter.
It is incredible, but after trying it a few times, I am beginning to get tired of the stylized nature of these podcasts. It is still too stereotyped or templated. I have caught it in minor mistakes, but nothing significant yet. The way the summation of each podcast goes feels like it is trying to lock the listener into a cautious but optimistic mindset.
100% agree. Incredible the first time. Repetitive after a few more. But it's a experimental product. I imagine those things can be easily tweaked. To me the biggest downside is that they can't really advance the conversation in any meaningful manner. It's pure summary/overview. I don't expect more than that yet but perhaps when the new kind of reasoning AI gets implemented...
I publish regularly on Substack so I turned one of my pieces into a podcast using NotebookLM. Astonishing. With that in mind, Alberto, what will you do with an AI podcast? Create another platform or put these podcasts up on Substack? Personal preference, for sure but what are you planning to do? Thanks for flagging this. Great take.
The best use case for me in particular (and it's very personal) is using it to restore to my short term memory a draft that has been sitting in my draft shelf for, say, a month. I have quite a lot of those. Drafts I wrote and then never got to edit them appropriately. In 5 min I get a clear overview of my ideas (and because they're mine I only need to refresh my memory) and also get into the mood of the topic specifically. Those are two critical frictions I have: remembering old ideas and getting in the mood of diving deep into them once again. Truly a productive enhancement for me.
(Maybe I should do a post about my process on this (I've already done twice and it works well for me). Writers who are hesitant to try AI as a tool to help them but not replace them may consider it.)
Very interesting and I can sort of see the use case that Alberto has laid out. Not for making podcasts but testing your own material, getting a new perspective, etc. I'd be cautious though as there is an inherent excitement and flattery to experiencing your content this way which could blind you to its real value. As an audio editor who has edited thousands of podcasts I can also say that the example given (and others I've listened to) does not convincingly portray how people really talk. It sounds at best like a paid promotion and people will get really tired really fast by the obviously forced back-and-forth and the goofy and gushing reactions to commonly made points (such as the ambivalent good of using the atomic bomb to end WWII when it led to a more unstable world after). Anyway, fun tool. Maybe it will even become part of some people's workflow. I'm going to test this with some things I've written on a very obscure topic (Taiwanese temple culture) and see how it performs.
100%. Agreed with both caveats, that it's easy to feel flattered and that it feels like paid promotion. Google would have to change that and offer the possibility to steer "criticism" and other variables.
I couldn't figure out how to use NotebookLM to make a podcast, but I asked it to summarize a very long and complicated doc and it did well. Then I asked it whether it could write a series of conversational blog posts based on the doc. That was an even better summary.
This is a summary of the essay, not a podcast of a conversation between 2 hosts. The male voice seems to get more words, they do not analyze or disagree, they simply paraphrase, they never mention anything not in the essay, they never explain what “easy” means or doesn’t, the sound like individuals speaking, but they don’t make jokes, have names, or refer to anything they know about the author that they might add, “the writer’s a philosophy professor, you know, and he is making a variation on the Trolley Problem.”
No Digital AI — text or verbal —should not be identified as such. If it’s a improv by AI, the text on which it’s based should be available as well. Podcast host must have an identity that can be vetted. Gender imbalances must be edited out (just as a person using fake ethnic accents would be no more acceptable in a podcast.)
Summaries — even in voices that sound like real individuals — should never be called Podcasts.
The timing of your post is uncanny. I'm just working on my Sunday Bonus section which is about a way I found to "nudge" the Audio Overview into using the style and structure you want. (There are no native controls for it besides the "Generate" button.) Here's what I wrote in the intro:
"But that [doesn't do] justice to just how impressive these AI podcasts are.
We’re not talking about a monotone robotic voice giving you a dry summary. These overviews genuinely feel like natural conversations. The two AI speakers crack jokes, laugh, interrupt and riff on each other’s statements, stop to catch a breath, occasionally stumble over their words, and so on.
As a taster, here’s a snippet of an audio overview made from the DALL-E 3 research paper."
I then share a short clip full of jokes, cultural references, and natural interplay between the speakers.
Yesterday, I was telling my wife how insanely impressive it is and we tested it on a 300+ page public annual report for Nordea (the bank where she works). We got back a thorough 15-minute podcast that made the matter engaging without losing any key details.
I've already felt that NotebookLM was extremely underrated and undermarketed. With the release of Audio Overviews, I personally feel it's perhaps the single best way to dive into a new topic and get the gist of complex sources.
It’s fascinating to see how far AI tools have come, especially with something like Google’s NotebookLM. The ability to turn complex written content into a podcast with just a click is pretty mind-blowing! It’s a game-changer for creators who want to reach audiences in new formats without the extra work.
I’m curious, though—do you think tools like this will eventually replace human-led podcasts or will they just become another tool in a creator’s toolkit? There’s something special about the human touch, but AI is making that line blurrier every day!
I tried it and it is pretty cool but I am not sure I see an application. If they find a way that the tool can learn voices so the audio is using my own voice, I could see it as a great tool to add to my podcast episodes and increase efficiency by using my Substack posts as content for podcasts, but this way, as you show, I can only see it as an addition to a post that people can read on Substack. Do you know if the system will be expanded to learn voices?
Been using it for digging old letters of some writers (in multiple languages). Podcasts aside, it handles it incredibly well, with citations, etc. and stays really focused on the uploaded sources, which does make it less creative, but I reckon that's not the point of this tool.
I tried it and was blown away- in the first sight/ sound as the AI simplified the contents and a complex idea I was sharing into an easy to understand voice conversation. Here is the link to audio: https://youtu.be/4AzdHsVTxm4?si=xBmaFzm6n0ouFJv9
I agree with you on why Google didn't (bother to?) market it. I doubt it was out of fear of misuse - had this been a real concern they wouldn't have released all of it to everyone.
I see your goosebumps and I raise my own... not sure if it's from a state of awe, or fear. Probably the latter.
I mostly find this replacement of humanity to be ever more terrible
right?! the goosebumps!
It is incredible, but after trying it a few times, I am beginning to get tired of the stylized nature of these podcasts. It is still too stereotyped or templated. I have caught it in minor mistakes, but nothing significant yet. The way the summation of each podcast goes feels like it is trying to lock the listener into a cautious but optimistic mindset.
100% agree. Incredible the first time. Repetitive after a few more. But it's a experimental product. I imagine those things can be easily tweaked. To me the biggest downside is that they can't really advance the conversation in any meaningful manner. It's pure summary/overview. I don't expect more than that yet but perhaps when the new kind of reasoning AI gets implemented...
I am worried about this sort of thing becoming a crutch for students.
I publish regularly on Substack so I turned one of my pieces into a podcast using NotebookLM. Astonishing. With that in mind, Alberto, what will you do with an AI podcast? Create another platform or put these podcasts up on Substack? Personal preference, for sure but what are you planning to do? Thanks for flagging this. Great take.
The best use case for me in particular (and it's very personal) is using it to restore to my short term memory a draft that has been sitting in my draft shelf for, say, a month. I have quite a lot of those. Drafts I wrote and then never got to edit them appropriately. In 5 min I get a clear overview of my ideas (and because they're mine I only need to refresh my memory) and also get into the mood of the topic specifically. Those are two critical frictions I have: remembering old ideas and getting in the mood of diving deep into them once again. Truly a productive enhancement for me.
(Maybe I should do a post about my process on this (I've already done twice and it works well for me). Writers who are hesitant to try AI as a tool to help them but not replace them may consider it.)
Very interesting and I can sort of see the use case that Alberto has laid out. Not for making podcasts but testing your own material, getting a new perspective, etc. I'd be cautious though as there is an inherent excitement and flattery to experiencing your content this way which could blind you to its real value. As an audio editor who has edited thousands of podcasts I can also say that the example given (and others I've listened to) does not convincingly portray how people really talk. It sounds at best like a paid promotion and people will get really tired really fast by the obviously forced back-and-forth and the goofy and gushing reactions to commonly made points (such as the ambivalent good of using the atomic bomb to end WWII when it led to a more unstable world after). Anyway, fun tool. Maybe it will even become part of some people's workflow. I'm going to test this with some things I've written on a very obscure topic (Taiwanese temple culture) and see how it performs.
100%. Agreed with both caveats, that it's easy to feel flattered and that it feels like paid promotion. Google would have to change that and offer the possibility to steer "criticism" and other variables.
I couldn't figure out how to use NotebookLM to make a podcast, but I asked it to summarize a very long and complicated doc and it did well. Then I asked it whether it could write a series of conversational blog posts based on the doc. That was an even better summary.
IS the quality good in spanish?
Unfortunately it's only available in English for now
Thanks for the information…
Amazing! I made this issue of the newsletter into a 7+ minutes podcast.
This is a summary of the essay, not a podcast of a conversation between 2 hosts. The male voice seems to get more words, they do not analyze or disagree, they simply paraphrase, they never mention anything not in the essay, they never explain what “easy” means or doesn’t, the sound like individuals speaking, but they don’t make jokes, have names, or refer to anything they know about the author that they might add, “the writer’s a philosophy professor, you know, and he is making a variation on the Trolley Problem.”
No Digital AI — text or verbal —should not be identified as such. If it’s a improv by AI, the text on which it’s based should be available as well. Podcast host must have an identity that can be vetted. Gender imbalances must be edited out (just as a person using fake ethnic accents would be no more acceptable in a podcast.)
Summaries — even in voices that sound like real individuals — should never be called Podcasts.
It's almost like... it's actually an AI experimental project!
The timing of your post is uncanny. I'm just working on my Sunday Bonus section which is about a way I found to "nudge" the Audio Overview into using the style and structure you want. (There are no native controls for it besides the "Generate" button.) Here's what I wrote in the intro:
"But that [doesn't do] justice to just how impressive these AI podcasts are.
We’re not talking about a monotone robotic voice giving you a dry summary. These overviews genuinely feel like natural conversations. The two AI speakers crack jokes, laugh, interrupt and riff on each other’s statements, stop to catch a breath, occasionally stumble over their words, and so on.
As a taster, here’s a snippet of an audio overview made from the DALL-E 3 research paper."
I then share a short clip full of jokes, cultural references, and natural interplay between the speakers.
Yesterday, I was telling my wife how insanely impressive it is and we tested it on a 300+ page public annual report for Nordea (the bank where she works). We got back a thorough 15-minute podcast that made the matter engaging without losing any key details.
I've already felt that NotebookLM was extremely underrated and undermarketed. With the release of Audio Overviews, I personally feel it's perhaps the single best way to dive into a new topic and get the gist of complex sources.
Google absolutely nailed this one!
It’s fascinating to see how far AI tools have come, especially with something like Google’s NotebookLM. The ability to turn complex written content into a podcast with just a click is pretty mind-blowing! It’s a game-changer for creators who want to reach audiences in new formats without the extra work.
I’m curious, though—do you think tools like this will eventually replace human-led podcasts or will they just become another tool in a creator’s toolkit? There’s something special about the human touch, but AI is making that line blurrier every day!
Most like just another tool IMO
I tried it and it is pretty cool but I am not sure I see an application. If they find a way that the tool can learn voices so the audio is using my own voice, I could see it as a great tool to add to my podcast episodes and increase efficiency by using my Substack posts as content for podcasts, but this way, as you show, I can only see it as an addition to a post that people can read on Substack. Do you know if the system will be expanded to learn voices?
I'm not sure but I'm not in favor of creating content using AI. I use it in my internal workflow, not to add to the pollution of the web.
Got it
In this case it is content created based on my original content but I hear you…
Been using it for digging old letters of some writers (in multiple languages). Podcasts aside, it handles it incredibly well, with citations, etc. and stays really focused on the uploaded sources, which does make it less creative, but I reckon that's not the point of this tool.
what the absolute what?! I am scared and impressed. This is full uncanny-valley territory~
Of course I had to try it on my latest post and here's AI Woman with the money quote: 'AI will be the ultimate Phreak'
should I laugh, should I cry? I laugh but it's a nervous laugh
This is insane. Definitely one of a handful of truly mind blowing AI moments so far.
I tried it and was blown away- in the first sight/ sound as the AI simplified the contents and a complex idea I was sharing into an easy to understand voice conversation. Here is the link to audio: https://youtu.be/4AzdHsVTxm4?si=xBmaFzm6n0ouFJv9
And this is to the post:https://open.substack.com/pub/rajeevlunkad/p/exploring-the-infinite-potential?r=76o9k&utm_campaign=post&utm_medium=web
What do you think?