I have made notebooks with links from all sorts of sites which are then populated as the "data" for the chatbot within the notebook (using Gemini). I've made some with my favorite substack posts (yours included) so I can compare and contrast and interrogate what's being discussed. For example, you could load up to 50 of your posts simply by pasting in the urls (300 if you get the upgrade - otherwise, all free), and then have them all in one place and explore what you've written, connections with other writers, common themes, etc... Very, very useful and not the kind of use cases that tend to get discussed in this space because we are all so worried about students "cheating" through writing. Tools like this, combined with the Deep Research models as well as agentic tools like Manus are what's going to overwhelm the AI conversation in the next year or so I believe. NotebookLM is also the one where you can make the AI podcasts which freaked everybody out last fall.
I see. I remember when people were wild about the podcasts from LM. I have experience with what you are referring to. Claude has a “project” feature that allows uploads of 10 pdf files which I use quite a bit. My focuses have been largely on exploring implications and interlocking themes among research articles. My procedure has been to identify an issue or question of interest to me, to use the Sac State library database for uncovering important and/or recent empirical studies, down load them based on what I see in the abstracts, winnow them out if my preliminary questions to the bot indicate particular studies are not especially relevant in the collection, familiarize myself with the details of each remaining study (i.e., research question(s), methodology, theoretical framework, and core findings), and then stage exploratory questions to help me grasp implications and themes either in conflict or convergent with one another. My old approach to building a bibliography for potential application in a research article (publish or perish) was very similar except it could take a months, and the takeaways weren’t quite as rich and illuminating. Claude in combination with the university library, ResearchGate, and Perplexity makes this process happen much more quickly and often better results. I still have to double check any quoted matter the bot. It’s not hallucinatory, but it’s often not verbatim. either. Sometimes it’s not fully contextualized when it’s excerpted from the body of the study. Nonetheless, it’s an amazing tool. I’ll take a look at NortebookLM. I have to say I don’t use ChatGPT because I don’t trust its filters on hallucination. Moreover, it can be very stubborn when challenged and render the whole conversation useless. I do not like bots wasting my time:) Thank you so much for reading and commenting, Steve. I know I can be cantankerous, but you’re a lawyer. You’ve dealt with cantankerous people before.
My preference is Claude as well. I just it find it overall more user-friendly and 3.7 Sonnet I think is top of the line. NotebookLM might be very useful for what you're doing I think. You should definitely check it out. All your discussion about research is relevant to the post I have coming out later today. Please take a look. And cantankerous is my middle name - I am the debate coach as well.
I’ll give Notebook LM a look. I have done some work with Gemini, and it looks pretty powerful to me. As you probably know, Claude has a toggle switch in Projects to let you connect to the internet or not. I like having the option. Does LM have that option?
Very familiar with Claude Projects - also with Custom GPT's which OpenAI seems to have lost interest in but are still very useful. NotebookLM is essentially a project. There are tons of videos and tutorials, but here is a short one that should give you at least a sense of how it works - best of all, it's free.
Terry - Have you used NotebookLM?
I have not, Steve. Nick P has, I think, and I’ve read a little about it. Why do you ask? I’ll check it out.
I have made notebooks with links from all sorts of sites which are then populated as the "data" for the chatbot within the notebook (using Gemini). I've made some with my favorite substack posts (yours included) so I can compare and contrast and interrogate what's being discussed. For example, you could load up to 50 of your posts simply by pasting in the urls (300 if you get the upgrade - otherwise, all free), and then have them all in one place and explore what you've written, connections with other writers, common themes, etc... Very, very useful and not the kind of use cases that tend to get discussed in this space because we are all so worried about students "cheating" through writing. Tools like this, combined with the Deep Research models as well as agentic tools like Manus are what's going to overwhelm the AI conversation in the next year or so I believe. NotebookLM is also the one where you can make the AI podcasts which freaked everybody out last fall.
I see. I remember when people were wild about the podcasts from LM. I have experience with what you are referring to. Claude has a “project” feature that allows uploads of 10 pdf files which I use quite a bit. My focuses have been largely on exploring implications and interlocking themes among research articles. My procedure has been to identify an issue or question of interest to me, to use the Sac State library database for uncovering important and/or recent empirical studies, down load them based on what I see in the abstracts, winnow them out if my preliminary questions to the bot indicate particular studies are not especially relevant in the collection, familiarize myself with the details of each remaining study (i.e., research question(s), methodology, theoretical framework, and core findings), and then stage exploratory questions to help me grasp implications and themes either in conflict or convergent with one another. My old approach to building a bibliography for potential application in a research article (publish or perish) was very similar except it could take a months, and the takeaways weren’t quite as rich and illuminating. Claude in combination with the university library, ResearchGate, and Perplexity makes this process happen much more quickly and often better results. I still have to double check any quoted matter the bot. It’s not hallucinatory, but it’s often not verbatim. either. Sometimes it’s not fully contextualized when it’s excerpted from the body of the study. Nonetheless, it’s an amazing tool. I’ll take a look at NortebookLM. I have to say I don’t use ChatGPT because I don’t trust its filters on hallucination. Moreover, it can be very stubborn when challenged and render the whole conversation useless. I do not like bots wasting my time:) Thank you so much for reading and commenting, Steve. I know I can be cantankerous, but you’re a lawyer. You’ve dealt with cantankerous people before.
My preference is Claude as well. I just it find it overall more user-friendly and 3.7 Sonnet I think is top of the line. NotebookLM might be very useful for what you're doing I think. You should definitely check it out. All your discussion about research is relevant to the post I have coming out later today. Please take a look. And cantankerous is my middle name - I am the debate coach as well.
I’ll give Notebook LM a look. I have done some work with Gemini, and it looks pretty powerful to me. As you probably know, Claude has a toggle switch in Projects to let you connect to the internet or not. I like having the option. Does LM have that option?
Very familiar with Claude Projects - also with Custom GPT's which OpenAI seems to have lost interest in but are still very useful. NotebookLM is essentially a project. There are tons of videos and tutorials, but here is a short one that should give you at least a sense of how it works - best of all, it's free.
https://youtu.be/YKNWkutNK6Y?si=abpAI5mmbKbHghsk