Over the 4th of July, I was sitting outside with my mom and sister. At one point, the conversation turned to books my mom was reading. My mom loves to read, so this is something we talk about from time to time. She is a big mystery fan, which doesn’t really appeal to me, but occasionally our reading interests align, e.g. Michael Crichton (Jurassic Park, Sphere, Airframe, Prey, etc.).

In this instance, however, my mom told us that she hadn’t been able to find any good books to read lately. She seemed bummed. After all, she loved to read.

“I think I can help you solve that problem, Mom!” I told her.

I told her that I thought she should try to leverage AI. My sister agreed!

I explained that AI was really good at taking a collection of data points (in this case authors, genres, etc.) and finding patterns that the AI could leverage to create new data points (e.g. suggested authors, books, etc.)

She seemed intrigued, so I said, “Here, let’s try it.”

I didn’t have my phone with me at the time, but my sister did, so we used her phone for this exercise. Her preferred large language model (LLM) AI at this point was ChatGPT, so we used that.

“Give us a list of your favorite authors, Mom.”

My mom gave us a list of about 5 authors and my sister typed them into ChatGPT.

Above this, my sister had already typed the action for the AI to take, something like “Give me a list of book recommendations based on this list of authors that I like.”

Before we sent the prompt, I asked my sister to let me enhance our prompt a little bit with additional context and a role for the AI to play.

For anyone newer to AI, adding context is a very important part of prompting. Today’s LLMs (Claude, ChatGPT, Gemini, etc.) are trained on vast amounts of data. And when prompting the model with a generic “do this” prompt, the result (output) does not always do what you want. However, by adding context, this helps the model by giving it clearer insight into what you are really asking it to do. In turn the model now has the ability to narrow its focus and utilize this additional information to determine how to create a better output, more aligned with the user’s initially desired goal.

So, I added the context:

We were providing a list of authors that we enjoyed reading, that we were having trouble finding new books to read, and that we were searching for books in the same genre and with similar themes.

I also gave the model a role to help it narrow its focus, something like:

You are a librarian with 25 years of experience. You are a literary expert. You love to read. You are great at making book recommendations based on authors and books that a person has enjoyed reading in the past.

We then sent the prompt.

The model began to research our request. And then several minutes later: a list of 5 curated book recommendations!

I read the list aloud to my mom and sister.

The first one my mom had already read, and greatly enjoyed!

The second one, she had also read, but didn’t like it as much, it was too violent.

The third one, she had never heard of, but as I read the AI provided summary, she said it sounded interesting.

I went back to the AI, and shared feedback which provided additional contextual data for it to use: “… have read this book, and really liked it,” “… have read this book, but didn’t like it — too violent,” and so on.

I sent the follow-up prompt and with this new context, the model used it to repeat the exercise and offer 5 new selections that were “similar to the first,” “not violent like the second,” etc.

The model returned 5 new selections, which I then shared with my mom.

She seemed excited! She made notes on her own phone and told us she was excited to try out a few of these new authors & books!

The next morning, at breakfast, I asked my mom her thoughts again. She reiterated her sentiment of excitement and thanked me for my help.

“Mom, do you want to learn how to do what I did last night so that you can do it yourself in the future?” I asked.

“Yes!” she exclaimed.

“Okay, let me show you on your phone!”

As she got out her phone, I explained that there were multiple options for AI (LLM) chat models.

“But, my favorite is Claude,” I told her.

“Let’s use that then,” she said.

So I helped her download Claude.

She then logged in and I gave her a quick rundown of how the app worked.

We then opened a chat and I repeated the exercise for her, as she watched, I typed:

Context / Goal: I am trying to find new books to read that match my interests and mirror other authors and books that I have enjoyed.

Role: You are a librarian with 25 years of experience. You are a literary expert. You are great at making book and author recommendations based on authors that your user enjoys.

Action: Give me a list of 5 books that you think I will enjoy based on the list of preferred authors I provide.

She then gave me a list of authors and we hit send.

Claude thought for a moment and began to research our request. And a few minutes later, she had a fresh list of books to read!

My mom was excited to get started on the list! And she now had Claude in her pocket, ready to help whenever she needed new book recommendations!

—

Fast forward a few days, my mom had started one of the books, and told me she was really enjoying it!

And then a few days later, while at my sister’s house, I watched her open a package with 2 books in it. She smiled at me and said: “I used AI to find new books for myself to read!”

This brought me a lot of joy. I had successfully shared (& taught) a new way for my mom and sister to leverage AI!