I love reading in French and am always looking for book recommendations. So I thought, why not use AI? Maybe I’ll get some good ideas. And I did, but it wasn’t as straightforward as I had hoped.
Getting Started
I’m not an AI user, and I wasn’t sure how to pose my question, so I googled it and got an answer from Google’s AI. In other words, AI was helping me use AI.
This is the prompt it suggested:
“I’m looking for book recommendations. Here are books I loved: [list of books.] Here are books I disliked: [list of books.] Please act as my personal book curator. Analyze the themes, writing style, and elements I enjoy (e.g., character-driven, fast-paced, specific tropes) and suggest 5 books in French for me and tell me why you suggest them.”
Well, that seemed easy! So off I went.
ChatGPT
I began with the tool that launched the whole AI craze, ChatGPT. I loaded in my prompt and waited…about 5 seconds. Gosh, these things are fast!
As expected, ChatGPT began with flattery, telling me that I had given it a “wonderful list— it’s rich and very coherent.” Yay, me! Then it got down to business.
It offered a thorough and generally accurate description of my likes and dislikes.
Impressive! Then came five books, and why it recommended them.
In reviewing its list, I found that ChatGPT had made two mistakes, one a mix-up and the other more subtle.
First, it recommended a book by an author I like, but it turns out that the book had been written by someone else, and the author I like had only written the foreword. Then, it recommended a book that is the second in a series of three, where you really need to read the first one to follow the story.
On the other hand, one book it recommended looked intriguing, so I ordered it.
Perplexity
My prompt was in English, but to my surprise, Perplexity responded in French! And it was culturally French as well—it didn’t gush about my “wonderful list” or blather on about my literary tastes, it just went straight to its recommendations.
Unfortunately, the books it suggested were all dated, and one I had already read and didn't like.
So, I asked for books written in the last 50 years.
When I got the new list, I found that I had read four of the five it recommended. I had only liked one of the four, which was not encouraging, but the book I hadn’t read looked interesting, so I ordered it.
Claude
Claude started as a mix of the first two AIs. It didn’t complement me on my book list, but it did summarize my tastes and did so in a rather elegant way—a good start.
It then recommended five books plus a “bonus” book, but when I looked into them, I found that four don’t actually exist! Claude was hallucinating!
When I told it this, I got a quick apology: “You’re absolutely right, and I apologize! I made up those titles instead of recommending real books. That’s a serious error on my part.” Indeed.
Then Claude gave me five new titles. I noticed that one had received the 2021 Prix Renaudot, one of France’s top literary awards. That didn’t sound right, and when I looked into it, I found that Claude was hallucinating again. This led to another apology—it seems that the book had been a finalist for another award that year, and Claude had mixed them up.
Despite all these errors, I did find one book that was both real and enticing, and I ordered it.
What Did I Learn?
The first thing I learned—like many before me—was to not trust everything an AI tells you. Sometimes it makes things up, sometimes it mixes things up, sometimes it misses the subtleties.
I was also surprised that none of the AIs picked up on the fact that many of the books I like have won French literary awards. I expected they would have noticed this common theme and used that knowledge in their recommendations. Perhaps they would have, had I given them a different prompt.
The real test will be whether I enjoy the books I’ve ordered, one recommended by each AI. Stay tuned!
Have you had similar experiences?
Image Credits:
All images courtesy of Pixabay royalty-free images
Introducing Contributor, Keith Van Sickle
Immerse yourself in all of Keith’s articles on his Contributor page.