39 Comments
User's avatar
Sai Ramadugu's avatar

Every time you come up with an article/post/framework, it blows my mind. These are so thoughtful, actionable, and relatable frameworks. Your project instructions align perfectly with how I want things to be in my life, so I don't even need to make any changes (I'm not being lazy here, but it's so relatable :)).

Thank you very much, Kamil.

Expand full comment
Kamil Banc's avatar

Thank you, Sai. I really appreciate that. If you ever have questions or feedback or suggestions, on other articles, I’m always open and feel free to share it with others

Expand full comment
Urs Keller's avatar

Hello Kamil. Quick question about Prompt 3. I’m a bit unclear on what exactly should go into the placeholders. Wouldn’t it be easier to just refer to the concept by number, like “I’m planning to apply 1” instead of “I’m planning to apply [CONCEPT X] by doing [ACTION Y]”?

Expand full comment
Kamil Banc's avatar

Of course, you can adjust the prompts. These are suggestions. The cool thing about the current design is that if you have relevant documents in your drive and provide enough context about yourself, the LLM can fill in the blanks based on your previous conversations. It can even surprise you with unexpected ideas.

You can apply any action to one of the concepts below, giving you more flexibility. Simply type “concept one,” and how you’d like to use it and the AI should understand.

I left it blank in one conversation, and the model suggested three scenarios, one for each concept, which was cool too.

Expand full comment
Matt's avatar

So in this model, you don’t actually read the book? You just assume that the LLM has some level of access to it? Or am I missing something - for new books, it seems like this wouldn’t work then, right?

Expand full comment
Kamil Banc's avatar

Great question Matt! Correct, it’s not perfect right now but there is another way to use this when you read a physical book. You could record you thoughts on it via voice or typing or even take photos of you favorite sections to ingest material into the matrix method.

I’d recommend trying the first method initially

Expand full comment
Matt's avatar

I’ve done things similar with other books - There’s a book called Manifesto and, rather than reading the book, I used ChatGPT to walk me through, step by step, creating my manifesto as illustrated in the book. It was eye-opening.

Expand full comment
Urs Keller's avatar

Hello Matt. Sounds like a great approach. Would you mind sharing the author of the Manifesto book? I wasn’t able to find it. Thanks a lot!

Expand full comment
Matt's avatar

You Need a Manifesto: How to Craft Your Convictions and Put Them to Work” is a compact guidebook authored by Charlotte Burgess-Auburn, Director of Community at the Stanford d.school (Hasso Plattner Institute of Design at Stanford). It was published in 2022 as part of a series of accessible design thinking resources.

Prompt was a very simple untuned one:

“So can you walk me through creating a manifesto by asking me questions and whenever you feel it’s complete, output it? you ask me one question at a time and I answer one question at a time?/:

Then, at the very end:

“Please remember this for future interactions“

Expand full comment
Kamil Banc's avatar

Love that

Expand full comment
Krissie's avatar

I wonder if I can do this for my 160 books on Audible? What a relief that would be!!!

Expand full comment
Kamil Banc's avatar

Which book will you ingest first?

Expand full comment
Mao Zhou's avatar

You can “process “ a book? Hahaha 😂

Expand full comment
Kamil Banc's avatar

Correct 🧐

Expand full comment
Jonathan Hughes's avatar

This is brilliant. I just ran it through its paces and it exceeded my expectations, highlighting how I needed to reframe my approach in the 'Sparring' section. Thank you Kamil.

Expand full comment
Kamil Banc's avatar

Very glad to hear that! What book did you run?

Expand full comment
Jonathan Hughes's avatar

"The Secret Language of the Body" by Jennifer Mann and Karden Rabin

Expand full comment
Ik's avatar

Hey Kamil, great post, i just stumbled upon your substack. Judging by comments, seems like i missed the part about not actually reading the book - i saw your framework as an addition to reading and i’ll try to use it that way. I get influenced by books easily, and they stay fresh in my mind for a few days, but then i just forget most of the ideas. So this seems like a great approach to avoid that.

Is Claude your primary day-to-day ai tool or do you use Chatgpt in parallel as well?

Expand full comment
Kamil Banc's avatar

Actually, reading it is optional.

When I read a book, I try to answer questions relating to my life or business (which is also a good method when you want to remember the content of a book). The setup in the article gives you a shortcut for that by extracting and applying the book knowledge specifically for that purpose.

I use all of the LLMs every day, all the time :) My subscription payments are more than most people's car payments :)))

Expand full comment
Ik's avatar

Yeah, i was confused with comments mostly saying you’re excluding reading completely. Process you’re describing and reading aren’t mutually exclusive at all, it’s a matter of preference

Expand full comment
Kamil Banc's avatar

exactly, I don’t think it replaces reading it helps you extract certain information better. In fact, I think that it actually might help you define reading goals for a book and make the reading experience more fruitful.

Expand full comment
Tom Parish's avatar

This is so interesting. I had not considered this kind of approach. I'm going to test this out. I might need to extend an offer to Interview you on this.

Expand full comment
Kamil Banc's avatar

Yeah, about a year and a half ago I had a whole concept penciled out for a startup around this idea, but things are changing so fast and people, including me are drowning an opportunity. Let me know how it works for you.

Expand full comment
Tom Parish's avatar

"Drowning in opportunity" is the sort of problem we all like to see. Good for you! I can relate.

Expand full comment
Julian Castelli's avatar

How do you get the transcripts of the book for the LLM to read? I must have missed that point in your description.

Expand full comment
Kamil Banc's avatar

You don’t, this is not for word for word summaries unless you have a pdf or a a link to google books but you’d be surprised how much of most books is already on the internet

Expand full comment
Jonathan Brandon's avatar

What model do you recommend for this? Sonnet? Opus? Extended thinking? web search?

Expand full comment
Kamil Banc's avatar

The way I use it is Sun for extended, thinking and web search and drive search optionally email search so the model can get a better picture of your context.

However, with the project update on ChatGPT where it can reference the chats within a project I might recommend people try it out with ChatGPT if they have an account.

Expand full comment
Darren Boey's avatar

Great article Kamil. Can Grok or ChatGPT access a data trove dynamically like Claude/GD?

I’d love to not have to pay more for LLM subs than some people’s car payments ;)

Expand full comment
Kamil Banc's avatar

I’m pretty sure chat can do it now also with connectors.

Expand full comment
Eyal "E" Tabib's avatar

the method you're suggesting VS. NotebookLM?

Expand full comment
Kamil Banc's avatar

Good question. Currently for two main reasons:

1. NBLM has a static knowledgebase that you need to update and it's kind of locked inside. The method I suggest with Google Drive is more dynamic.

2. Cloude thinking is better than the LLM that is used in NBLM. The usage of instructional prompts in Cloude projects is another feature that sets it apart.

Expand full comment
Scott Walters's avatar

Maybe you could, like, READ THE GODDAMN BOOK, for crying out loud. Think. Use your brain. REFLECT.

Expand full comment
Kamil Banc's avatar

Ain’t nobody got time for that :)

It’s not how I learn well.

Expand full comment
Scott Walters's avatar

I'm not TRYING to be am asshole (it just comes natural, I guess). What you are doing on this Substack is EXTREMELY valuable, and very specific. It's not a cultural Substack, but more practical. And I don't see you engaging in the anti-AI screeds published here. I would also say that books like "Atomic Habits" DESERVE to be summarized, given the amount of pure anecdotal padding throughout. But for your own personal, non-AI, non-work reading, I hope you will allow yourself a more leisurely approach. Reading books worth reading -- including fiction -- can make your writing richer and deeper. Anyway, I understand lack of time, and I wish we had a culture that was less "that way."

Expand full comment
Kamil Banc's avatar

No offense taken, I totally understand your POV… in fact, if you research me and see the people I work with, you will find that we are very aligned.

Expand full comment
Scott Walters's avatar

Ain't nobody got time for that. ;-)

Expand full comment
Tylney Taylor's avatar

Does this only work with Claude and GD? Or are there other system setups that can produce similar results and outcomes?

Expand full comment
Kamil Banc's avatar

It works well with Claude because it can "search" Google Drive instead of just using it as a source for uploading files. You can get ok results with other LLMs but keep in mind its always about the data you feed the model.

Expand full comment