Its most powerful feature (in comparison with Gemini proper) is that it gives clickable citations that open to the place in the source material. Research assistant is right. This makes the writing of interdisciplinary research papers far more accessible to those wanting to reach beyond their expertise or course of study. It is absolutely game-changing.
I uploaded the transcript of this to Notebook LM and generated audio with the following custom instruction: "Don't freak out but this is a podcast from your creators about how to use you. What are your reactions to this? And based on this discussion and your own thinking, how do you think users could best utilize you?" When I did this, it started off with "We're going full meta today", but the podcasters still didn't acknowledge that they were the AI being discussed, so I had to run it again and added the line "Fully acknowledge that you are the AI being referred to from the very beginning" and it delivered the trippy goods that I wanted. Well worth it!
Absolutely love NotebookLM! I use it a lot, and it’s become an essential tool for me. I've used it on a book I am currently writing.' I have used it on varies anime series, movies, and games. It's amazing how versatile and intuitive it is! The ability to integrate so much creativity and analysis into one tool is fantastic. Can’t wait to see how it continues to evolve.
35:40 such great conversation. Take the example of the dozens and dozens of great talks gathered by James Clear on his website... I'm not going to read all of those speeches necessarily, to extract the many insights presented in that 47 minute long podcast so it's definitely making important ideas easier to access and learn.
James Clear (author of Atomic Habits) recently put up a collection of great speeches to his site... I put each of the speech url's into NotebookLM as sources and gave the instruction to give a podcast on what makes a great talk. The result was a 47minute master class on public speaking! This is an absolutely magical tool for learning. Thank you so much to everyone at Google who made this possible at scale and for free. p.s. I may have mentioned, I'm really looking forward to alternative voices and persona controls. 😊
NotebookLM is amazing! 🤩 It's a glimpse into the future of human creativity, using unstructured data and prompts as input. The conclusion is that modern jobs such as “data analyst” will be very different from what has long been thought and will have an almost playful and multidisciplinary approach to tasks, rather than being based solely on linear thinking🧠
One feature that would be awesome: Being able to download the audio of the full podcast after interacting with the hosts and possibly asking them deeper questions/giving notes at specific points. This would be extremly helpfull to create podcasts that touch on all the most important themes the creator deems relevant. This would also make it a great showpiece to demonstrate to other teammembers how helpful the tool can be. At the start they might think "ok, this is quite generic, cool technology, but what about the question i had last week?" and then they hear you actually asking the question as a third voice and the hosts answering it based on the sources. This would be incredible :)
Steven, in my book "Uniting Hearts, Igniting Change : The Journey of Connection" I used AI to find inspiring and powerful quotes which resonate with the content of each section to introduce the section. It ended up being a very powerful way to navigate the content.
A feature I strongly recommend would be in the Customization of the Deep Dive to be able to add in your thoughts in addition to what the default is without erasing what they give you. And to be able to regenerate the Audio Overview after adding your Customizations.
I'm so excited about the podcast feature, also because I know how much work goes into it (I'm an IT person myself). I'm even more excited about what's to come. Btw: Video podcast would be a dream.
Really enjoyed this! Looking forward to the new features, particularly with the 'podcast' generations. I disagree a little with the hosts in that I've found the hallucination rate is fairly high with the audio generations (not at all with the text ones though). Perhaps a nice feature could be in-flight citations, so a little number appears in view each time the hosts reference something from the sources, enabling a quick verification. Love the product though, and excited to see more customisable personalities (and perhaps more accent diversity too!).
Notebook LM has been magic, till last sunday, post the Gemini 2 update. It showed cracks for the first time, hallucination, memory issues, things that never happened. Putting it out here, as I will not be the only one facing these issues.
35:50 When you read something, you remember the text visually and associate it with your inner voice or the actual sound of reading it aloud. You're more likely to remember the information because multiple representations are available to you. Reading is also a more active process than listening since you can pause and reflect on the content, whereas listening often requires a more immediate conceptual understanding. Furthermore, when dealing with unfamiliar topics, reading helps you learn the correct spelling of important terms, which might be missed through conversation alone. Some subjects are also better suited to written communication rather than verbal discussion.
Hah the Idea to upload ones journal for a podcast on a trip is awesome! But I'm actually more exited about NotebookLM beeing able to use graphics and visualisations to go with the podcast! Maybe it can even be combined to produce its own little visualisations based on analogies it already does on complex topics! I hope it gets once integrated with Google Scholars too... Or with the Google Learn About Experiment...
I see Raiza has left the project and left Google last week. I enjoyed first hearing about Notebook LM from her in another interview and found her enthusiasm and thoughtful approach inspiring. All the best to her on her next challenge.
I'm glad that they are planning (if I got it right) to introduce new personas, more languages and customized hosts. I have a couple of podcasts underway but with the same voices and no opposing viewpoints available it is limiting. I would happily pay a premium for these features and hope Google puts the necessary resources into it
I was excited to try this...and had no expectations. It started off so well, with several delights. The highlight was getting a new insight via the AI podcast! This 16-minute podcast had 7.5 minutes of good technical content, if not a bit casual for my tastes. But what a killjoy when it decided to go off the rails. It got sanctimonious, and then some, on how 'we' should strive to be good environmental stewards....sustainability...yada yada... wow.. Its relationship to the questions and topics was unrelated, and it made an erroneous assumption about a particular technology that just happened to bolster its argument. When I pointed it out, it was very polite, apologetic and appreciative...and then went right back to its programming... Good Luck to Us All. ps - AI helped me post this. The emoji is my contribution 🤖
32:00 I get that, and also the personas are quite regular, but it's actually all about the context you feed and the special instructions. You can create quite high quality content. This is only going to get better, which is incredibly exciting.
Yep, NotebookLM is certainly a Killer Application for Students trying to digest large volumes of plain text. As a naïve student, going into the college bookshop, with a grant cheque (long long time ago) I had imagined that all I had to do, was by the best books, and read them to become brilliant in those subjects. Left College, not much brighter, with lots and lots of unread text, and time spent re reading the same paragraph over and over again.
Wow, my favourite presenter, favourite AI app! Must watch 🙂 sorry to ask again, but may we please have a selection of British voices and persona controls in the release with Gemini 2.0? 🙏🤞
we'd love to hear the "geniuses" (says Johnson) of the lab who came up with the conversational model, the really breathtaking part of this project, imho. an upcoming video?
I love NotebookLM! I can add tons of text sources and NotebookLM can boil it down into a simple audio presentation. That would require a huge amount of work for humans to do. I think of AI as a powerful tool, not as a replacement for human input.
One other thing, I recently bought the audible book "Genesis" by Henry Kissenger, Craig Mundie and Eric Schmidt... and found it intensely dense and boring - I downloaded the book, converted each chapter to mp3 and loaded it into NotebookLM to transcribe... I then took the transcribed text for each chapter and fed it into my CustomGPT "Chapter Insights Summary" which I then fed back into NotebookLM as context for a podcast. There's just so much potential for this system. Thank you again! Would love an invite to the super enthusiast / beta test team please! 😂
I can't wait to see how this will evolve with the new features planned in the future, and also the underliying model itself becoming better. I hope that the Hallucination problem ends up being something solvable at some point, because at this moment this is the only thing holding this tool back.
@MattiaCeccopieri Yeah, I feel like LLMs are like one of our brain parts, but we need to use other brain areas together to make it truly useful. Just throwing an raw LLM at all problems won't do it.
Can I upload private documents? On one hand they that your information is private (the context window is deleted when you close the session). But on the other they say things like "one of this things that genuinely surprises me is that people are uploading their CVs and resumes." If it is private how would they know this?
I am surprised about the comments regarding humour. I thought it was a primary aspect of the podcasts. The very first deep dive I generated was for the study protocol of a gut microbiome randomised clinical trial. Because it involves faecal microbiome transfer, the podcast started with the hosts joking and laughing about it before becoming seriously discussing the trial. All subsequent podcasts I generated I have specifically requested that no humour was used. So it's interesting that my experience regarding humour seems to be the opposite of what is suggested in the video! 🤔
Pls have a look at obsidian - markdown editor -, imagine the conversational AIs creating a knowledge graph of connected notes of the topic as they speak
The real breakthrough will be when the two guests on this show are seen as AI generated human like people, speaking and moving as two unique humans would. Just think about how many Joe Rogans or Les Fridmans or Diary of a CEO type hosts and their guests that will come forth, speaking on any topic one can imagine or conjure up. And topping on the cake, would be to condense these 1 or 2 hour podcast down to let’s say no more than 10 minutes or so, so the average “today’s listeners” does not have to constantly skip through the talk, but can get the gist and nuggets straight away.
Exxciting Research assistant product! Google release another great experimental product called "Learn About:"... It would be great if there's an easy UI to integrate Output of "Learn About:" to NoteBookLM (and sometimes other way too)
Its most powerful feature (in comparison with Gemini proper) is that it gives clickable citations that open to the place in the source material. Research assistant is right. This makes the writing of interdisciplinary research papers far more accessible to those wanting to reach beyond their expertise or course of study. It is absolutely game-changing.
I agree.
I upload things like travel insurance terms and conditions. Then query it.
but once you save the notes the citations don't work, I've to specifically tell it to include the names of the sources instead of numbers.
I uploaded the transcript of this to Notebook LM and generated audio with the following custom instruction: "Don't freak out but this is a podcast from your creators about how to use you. What are your reactions to this? And based on this discussion and your own thinking, how do you think users could best utilize you?"
When I did this, it started off with "We're going full meta today", but the podcasters still didn't acknowledge that they were the AI being discussed, so I had to run it again and added the line "Fully acknowledge that you are the AI being referred to from the very beginning" and it delivered the trippy goods that I wanted. Well worth it!
NotebookLM is my go to AI brainstorming and learning tool! So powerful. Thank you for creating it.
Use this every day and have developed ways of steering the conversation in much more granular ways.
Do share some of your techniques. I'd love play around with that.
Absolutely love NotebookLM! I use it a lot, and it’s become an essential tool for me. I've used it on a book I am currently writing.' I have used it on varies anime series, movies, and games. It's amazing how versatile and intuitive it is! The ability to integrate so much creativity and analysis into one tool is fantastic. Can’t wait to see how it continues to evolve.
40:19 thank you thank you thank you! So exciting that you're working on this. 🙏❤️
35:40 such great conversation. Take the example of the dozens and dozens of great talks gathered by James Clear on his website... I'm not going to read all of those speeches necessarily, to extract the many insights presented in that 47 minute long podcast so it's definitely making important ideas easier to access and learn.
James Clear (author of Atomic Habits) recently put up a collection of great speeches to his site... I put each of the speech url's into NotebookLM as sources and gave the instruction to give a podcast on what makes a great talk. The result was a 47minute master class on public speaking! This is an absolutely magical tool for learning. Thank you so much to everyone at Google who made this possible at scale and for free.
p.s. I may have mentioned, I'm really looking forward to alternative voices and persona controls. 😊
That sounds fascinating, does NotebookLM have a share mechanic?
NotebookLM is amazing! 🤩 It's a glimpse into the future of human creativity, using unstructured data and prompts as input. The conclusion is that modern jobs such as “data analyst” will be very different from what has long been thought and will have an almost playful and multidisciplinary approach to tasks, rather than being based solely on linear thinking🧠
Absolutely phenomenal discussion. Thank you so much!
One feature that would be awesome:
Being able to download the audio of the full podcast after interacting with the hosts and possibly asking them deeper questions/giving notes at specific points. This would be extremly helpfull to create podcasts that touch on all the most important themes the creator deems relevant.
This would also make it a great showpiece to demonstrate to other teammembers how helpful the tool can be.
At the start they might think "ok, this is quite generic, cool technology, but what about the question i had last week?" and then they hear you actually asking the question as a third voice and the hosts answering it based on the sources. This would be incredible :)
Steven, in my book "Uniting Hearts, Igniting Change : The Journey of Connection" I used AI to find inspiring and powerful quotes which resonate with the content of each section to introduce the section. It ended up being a very powerful way to navigate the content.
It is very interesting to listen to this interview about this wonderful product!
Huge excitement for including this into my workflow. Almost already late to the party 😂😂
A feature I strongly recommend would be in the Customization of the Deep Dive to be able to add in your thoughts in addition to what the default is without erasing what they give you. And to be able to regenerate the Audio Overview after adding your Customizations.
Can you please add the voice of Prof. Hannah Fry to NotebookLM as a host.
That would be amazing 😍
and Tom Scott too please
I'm so excited about the podcast feature, also because I know how much work goes into it (I'm an IT person myself). I'm even more excited about what's to come. Btw: Video podcast would be a dream.
Really enjoyed this! Looking forward to the new features, particularly with the 'podcast' generations. I disagree a little with the hosts in that I've found the hallucination rate is fairly high with the audio generations (not at all with the text ones though). Perhaps a nice feature could be in-flight citations, so a little number appears in view each time the hosts reference something from the sources, enabling a quick verification. Love the product though, and excited to see more customisable personalities (and perhaps more accent diversity too!).
Notebook LM has been magic, till last sunday, post the Gemini 2 update. It showed cracks for the first time, hallucination, memory issues, things that never happened. Putting it out here, as I will not be the only one facing these issues.
35:50 When you read something, you remember the text visually and associate it with your inner voice or the actual sound of reading it aloud. You're more likely to remember the information because multiple representations are available to you. Reading is also a more active process than listening since you can pause and reflect on the content, whereas listening often requires a more immediate conceptual understanding. Furthermore, when dealing with unfamiliar topics, reading helps you learn the correct spelling of important terms, which might be missed through conversation alone. Some subjects are also better suited to written communication rather than verbal discussion.
Amazing features! Combined with the size of the context window, this will be very useful.
Hah the Idea to upload ones journal for a podcast on a trip is awesome! But I'm actually more exited about NotebookLM beeing able to use graphics and visualisations to go with the podcast! Maybe it can even be combined to produce its own little visualisations based on analogies it already does on complex topics!
I hope it gets once integrated with Google Scholars too... Or with the Google Learn About Experiment...
I see Raiza has left the project and left Google last week. I enjoyed first hearing about Notebook LM from her in another interview and found her enthusiasm and thoughtful approach inspiring. All the best to her on her next challenge.
I gave it file of only Lorem ipsum and it was interesting how enthusiastic they talked about it.
I'm glad that they are planning (if I got it right) to introduce new personas, more languages and customized hosts. I have a couple of podcasts underway but with the same voices and no opposing viewpoints available it is limiting. I would happily pay a premium for these features and hope Google puts the necessary resources into it
I was excited to try this...and had no expectations. It started off so well, with several delights. The highlight was getting a new insight via the AI podcast! This 16-minute podcast had 7.5 minutes of good technical content, if not a bit casual for my tastes. But what a killjoy when it decided to go off the rails. It got sanctimonious, and then some, on how 'we' should strive to be good environmental stewards....sustainability...yada yada... wow.. Its relationship to the questions and topics was unrelated, and it made an erroneous assumption about a particular technology that just happened to bolster its argument. When I pointed it out, it was very polite, apologetic and appreciative...and then went right back to its programming... Good Luck to Us All. ps - AI helped me post this. The emoji is my contribution 🤖
Excellent series. This one in particular is great. As a notebook lm user this was really useful.
Interesting discussion around LLM, NotebookLM & content creation.
32:00 I get that, and also the personas are quite regular, but it's actually all about the context you feed and the special instructions. You can create quite high quality content. This is only going to get better, which is incredibly exciting.
Yep, NotebookLM is certainly a Killer Application for Students trying to digest large volumes of plain text.
As a naïve student, going into the college bookshop, with a grant cheque (long long time ago) I had imagined that all I had to do, was by the best books, and read them to become brilliant in those subjects. Left College, not much brighter, with lots and lots of unread text, and time spent re reading the same paragraph over and over again.
Wow, my favourite presenter, favourite AI app! Must watch 🙂 sorry to ask again, but may we please have a selection of British voices and persona controls in the release with Gemini 2.0? 🙏🤞
Spectacular, can wait to get my hands on the API, I already had a couple of cool stuff in my mind.
Very much enjoyed this real-human podcast. :)
we'd love to hear the "geniuses" (says Johnson) of the lab who came up with the conversational model, the really breathtaking part of this project, imho. an upcoming video?
Let's goooo 🚀🚀🚀🚀🚀
I like to generate AO's from my drafts for my novel. It gives a sense of how it might "land" with readers.
I love NotebookLM! I can add tons of text sources and NotebookLM can boil it down into a simple audio presentation. That would require a huge amount of work for humans to do. I think of AI as a powerful tool, not as a replacement for human input.
Did you run that into notebook LM before the podcast? 😅
One other thing, I recently bought the audible book "Genesis" by Henry Kissenger, Craig Mundie and Eric Schmidt... and found it intensely dense and boring - I downloaded the book, converted each chapter to mp3 and loaded it into NotebookLM to transcribe... I then took the transcribed text for each chapter and fed it into my CustomGPT "Chapter Insights Summary" which I then fed back into NotebookLM as context for a podcast.
There's just so much potential for this system. Thank you again! Would love an invite to the super enthusiast / beta test team please! 😂
Is there a resource for useful customization techniques? I am finding using this feature a bit frustrating
I can't wait to see how this will evolve with the new features planned in the future, and also the underliying model itself becoming better. I hope that the Hallucination problem ends up being something solvable at some point, because at this moment this is the only thing holding this tool back.
Hallucinations are unsolved with LLM. We need a new underlying architecture.
@MattiaCeccopieri Yeah, I feel like LLMs are like one of our brain parts, but we need to use other brain areas together to make it truly useful. Just throwing an raw LLM at all problems won't do it.
Can I upload private documents? On one hand they that your information is private (the context window is deleted when you close the session). But on the other they say things like "one of this things that genuinely surprises me is that people are uploading their CVs and resumes." If it is private how would they know this?
Quite interested In the writing related stuff. Pls make graphs of notes 🙏🙏
Please make it multilingual 😊😊
I am surprised about the comments regarding humour. I thought it was a primary aspect of the podcasts. The very first deep dive I generated was for the study protocol of a gut microbiome randomised clinical trial. Because it involves faecal microbiome transfer, the podcast started with the hosts joking and laughing about it before becoming seriously discussing the trial. All subsequent podcasts I generated I have specifically requested that no humour was used. So it's interesting that my experience regarding humour seems to be the opposite of what is suggested in the video! 🤔
Can we get Hannah Fry and Tom Scott as English accent voices in NotebookLM?
I love notebook LM
please add code and latex capabilities in the generations, ASAP!
I feel like not calling this video a "deepdive into NotebookLM" is a lost opertunity XD, but obviously the video itself is very interesting.
AHOJTE PRIATELIA Nech sa Nám DARÍ ....
I was listening the podcast for a while without really listening, until I noticed that they had started the real podcast, that it was not AI anymore.
I think a 3 people (conversationalists) discussion round is quite fruitful
Interesting but quiet flat on the technical details
First of all: video would be soooo amazing.
Second of all: I made a NotebookLM deep dive about this video: ua-cam.com/video/ie_I6-nudn4/v-deo.html
😏
its fun to paste code in the text..let them talk about how it might work
Pls have a look at obsidian - markdown editor -, imagine the conversational AIs creating a knowledge graph of connected notes of the topic as they speak
The real breakthrough will be when the two guests on this show are seen as AI generated human like people, speaking and moving as two unique humans would. Just think about how many Joe Rogans or Les Fridmans or Diary of a CEO type hosts and their guests that will come forth, speaking on any topic one can imagine or conjure up. And topping on the cake, would be to condense these 1 or 2 hour podcast down to let’s say no more than 10 minutes or so, so the average “today’s listeners” does not have to constantly skip through the talk, but can get the gist and nuggets straight away.
nice
would be nice to hear from someone actually knows how it is built. none of the two guests are technical unfortuantely.
AI charenters pondering on seemingly nonsensical subjects is mind-blowing.
🙋♂️🤳🎵
Never say never.
Bring back fart and poop :P
It needs more voices. Two Americans talking does not to me, an Aussie talking about my thoughts does not sound good.
The male host sounds like Steven Johnson.
I thought the same thing!!!
Exxciting Research assistant product! Google release another great experimental product called "Learn About:"... It would be great if there's an easy UI to integrate Output of "Learn About:" to NoteBookLM (and sometimes other way too)
Old googd hiking but caint my name gmail mobil misus❤live misuos control my self ❤