This is fascinating and I do hope it continues to develop as a support for writers and that writers use it within these guidelines. I think we bump up against larger cultural conflicts that certainly aren’t the fault of the tool, but it highlights these issues. For example, a college professor friend has seen an alarming increase in her composition course of students doing what you advised against: copying and pasting without applying any thinking or analysis. I am sympathetic to students who have been taught that a good grade is the desired result, rather than learning to think. Mistakes are so much more instructive, but we never seem to allow anyone to make them. Here’s hoping that the increase in AI tools makes human elements in writing more valuable overall. Ok,
I feel like an old granny, even though the prospect of quoting sources at lightning speed is genuinely delicious. I will give this a test drive for sure.
Steven... I have just found this product and have used it for the last 3-4 days and find it really useful. I have a notebook with 4 long pdf documents related to Biblical prophecy and it is really surprising to see a concise summary of each document just by clicking on that document. Two great features I have already found; each question shows the number of citations and you can clickeach and review the actual citation, and the second is the suggested questions from the documents. A really beneficial way to study the material.
I'm not totally sure this is correct but it appears that when asking questions the results only come from the sources in the notebook which limits a bit continual investigation outside the notebook sources. For my needs it would be beneficial to have a ChatGPT like interface to the broader band of knowledge of the large lanugage models of the recent AI vintage.
I really think Google is onto something here and combined with broad LLM results could be spectacular.
As a relatively recent convert to writing both fiction and essays in Scrivener, I’m curious how you find using Notebook LM in comparison to a deep writing tool (which is more devoted to creating and structuring pieces rather than finding research), and/or how you see them interacting in your workflow if you’re using both. I’m in the UK so unable to try for myself for now, but I must confess I’m interested.....
It's a great question. Right now I think it's kind of a two window experience, because Scrivener has so many tools for actually creating a long-form document, as you say. (For instance, NotebookLM doesn't have the ability to create footnotes/endnotes in things that you write.) So what I would do is compile your existing Scrivener file as a PDF and bring that into NotebookLK and use that to ask questions, get overviews, quotes, etc. And then paste into Scrivener to actually do your writing. Not ideal but I think would be an improvement on just using Scrivener. Over time, it may be possible to write a whole book in NotebookLM or NotebookLM and Docs in a way that would be equivalent to Scrivener. We'll see. (And I think you can use NotebookLM in the UK with a VPN.)
Thank you - some food for thought and new tools for thought as well! I will be interested to hear anything about how the adoption of a tool like this affects your future output if you have any pieces that happen to be a controlled experiment in your writing and thinking before and after adopting it.
Hey Stephen! I'm a huge fan of the concept of the adjacent possible as you've outlined in (most recently for me, anyway) "Where Good Ideas Come From." I agree completely with this way of thinking about how ideas form. I'm a huge fan of Feynman, and write about him way too much, largely because I think this is just how he saw the world.
Just wanted to throw that out there first! I'm playing with Notebook LM now, but it doesn't seem quite ready for prime time. it loses the conversation thread quickly and isn't keeping up with my conversation (for reference, I use GPT4 and Bard every day). I'm confident this will improve rapidly, and i am excited!
Thanks, Andrew! We're working on the conversation history issues. It should perform reasonably well with some followup questions, but its "memory" is currently only one turn in the conversation. It's tricky with the source grounding because each time you ask a question, we are retrieving a bunch of passages from your sources to put in the context window -- and those passages are based on the current query. It's going to get much better as we fully transition to Gemini over the next week, but in the meantime, I think you will get very solid results if you don't reference the conversation history in your questions. In other words, say: "tell me more about pirates" and not "tell me more about them" etc.
We are definitely going to be rolling out more input types -- integrating with the Web would make particular sense, given that it's Google! Besides the limited source categories, what's holding you back from finding it helpful? What would you be trying to use it for?
I would love to have a powerful AI-based notebook. I would use it for everything. However, like most other potential users I already have a "notebook system" so Notebook LM would have to at least come close to my current system for me to consider switching.
Right now it is a nice plaything for exploring and asking questions about a small set of notes. However, even here it seems to currently have many issues. Bug every minute or two.
Just curious - is the multiple notebook model gonna stay? Because that is a deal breaker. I think everyone would want to be able to search, find common information, correlate, etc... across multiple "chapters" of their notes.
Thanks, Stephen! I appreciate the product a great deal. I run my own businesses, and being able to analyze some things beyond what I currently can see, and having it live in the Google ecosystem is appealing.
The source materials are stored on Google's servers while you are an active user of NotebookLM. But we do not train the model on any of the information in those sources -- we effectively just put it in the "short term memory" of the AI and then wipe it as soon as your conversation is over. So you can feel confident using private or corporate information, or material under copyright that you have permission to use.
This is fascinating and I do hope it continues to develop as a support for writers and that writers use it within these guidelines. I think we bump up against larger cultural conflicts that certainly aren’t the fault of the tool, but it highlights these issues. For example, a college professor friend has seen an alarming increase in her composition course of students doing what you advised against: copying and pasting without applying any thinking or analysis. I am sympathetic to students who have been taught that a good grade is the desired result, rather than learning to think. Mistakes are so much more instructive, but we never seem to allow anyone to make them. Here’s hoping that the increase in AI tools makes human elements in writing more valuable overall. Ok,
I feel like an old granny, even though the prospect of quoting sources at lightning speed is genuinely delicious. I will give this a test drive for sure.
Steven... I have just found this product and have used it for the last 3-4 days and find it really useful. I have a notebook with 4 long pdf documents related to Biblical prophecy and it is really surprising to see a concise summary of each document just by clicking on that document. Two great features I have already found; each question shows the number of citations and you can clickeach and review the actual citation, and the second is the suggested questions from the documents. A really beneficial way to study the material.
I'm not totally sure this is correct but it appears that when asking questions the results only come from the sources in the notebook which limits a bit continual investigation outside the notebook sources. For my needs it would be beneficial to have a ChatGPT like interface to the broader band of knowledge of the large lanugage models of the recent AI vintage.
I really think Google is onto something here and combined with broad LLM results could be spectacular.
As a relatively recent convert to writing both fiction and essays in Scrivener, I’m curious how you find using Notebook LM in comparison to a deep writing tool (which is more devoted to creating and structuring pieces rather than finding research), and/or how you see them interacting in your workflow if you’re using both. I’m in the UK so unable to try for myself for now, but I must confess I’m interested.....
It's a great question. Right now I think it's kind of a two window experience, because Scrivener has so many tools for actually creating a long-form document, as you say. (For instance, NotebookLM doesn't have the ability to create footnotes/endnotes in things that you write.) So what I would do is compile your existing Scrivener file as a PDF and bring that into NotebookLK and use that to ask questions, get overviews, quotes, etc. And then paste into Scrivener to actually do your writing. Not ideal but I think would be an improvement on just using Scrivener. Over time, it may be possible to write a whole book in NotebookLM or NotebookLM and Docs in a way that would be equivalent to Scrivener. We'll see. (And I think you can use NotebookLM in the UK with a VPN.)
Thank you - some food for thought and new tools for thought as well! I will be interested to hear anything about how the adoption of a tool like this affects your future output if you have any pieces that happen to be a controlled experiment in your writing and thinking before and after adopting it.
Curious Steven - do you guys have "beta" users providing input/feedback on this product?
Thanks!
Hey Stephen! I'm a huge fan of the concept of the adjacent possible as you've outlined in (most recently for me, anyway) "Where Good Ideas Come From." I agree completely with this way of thinking about how ideas form. I'm a huge fan of Feynman, and write about him way too much, largely because I think this is just how he saw the world.
Just wanted to throw that out there first! I'm playing with Notebook LM now, but it doesn't seem quite ready for prime time. it loses the conversation thread quickly and isn't keeping up with my conversation (for reference, I use GPT4 and Bard every day). I'm confident this will improve rapidly, and i am excited!
Thanks, Andrew! We're working on the conversation history issues. It should perform reasonably well with some followup questions, but its "memory" is currently only one turn in the conversation. It's tricky with the source grounding because each time you ask a question, we are retrieving a bunch of passages from your sources to put in the context window -- and those passages are based on the current query. It's going to get much better as we fully transition to Gemini over the next week, but in the meantime, I think you will get very solid results if you don't reference the conversation history in your questions. In other words, say: "tell me more about pirates" and not "tell me more about them" etc.
Waiting for that full transition over to Gemini... right now... it's just not that useful.
Also, is there a reason why sources can't be article links? Google Docs and PDFs are very limiting. Are there plans in the future for more sources?
This seems like a great idea, but the current implementation is really weak.
We are definitely going to be rolling out more input types -- integrating with the Web would make particular sense, given that it's Google! Besides the limited source categories, what's holding you back from finding it helpful? What would you be trying to use it for?
I would love to have a powerful AI-based notebook. I would use it for everything. However, like most other potential users I already have a "notebook system" so Notebook LM would have to at least come close to my current system for me to consider switching.
Right now it is a nice plaything for exploring and asking questions about a small set of notes. However, even here it seems to currently have many issues. Bug every minute or two.
Just curious - is the multiple notebook model gonna stay? Because that is a deal breaker. I think everyone would want to be able to search, find common information, correlate, etc... across multiple "chapters" of their notes.
Thanks, Stephen! I appreciate the product a great deal. I run my own businesses, and being able to analyze some things beyond what I currently can see, and having it live in the Google ecosystem is appealing.
Does Google retain access to either the source materials (as in your Commonplace Book) or your questions?
The source materials are stored on Google's servers while you are an active user of NotebookLM. But we do not train the model on any of the information in those sources -- we effectively just put it in the "short term memory" of the AI and then wipe it as soon as your conversation is over. So you can feel confident using private or corporate information, or material under copyright that you have permission to use.
THX, will surely have a close look.