I find ChatGPT gives me better movie recommendations than Netflix when I do the following:

1) Keep a list of books/movies that you like and why you like them 

2) Use the prompt "Recommend X movies/books that [CRITERIA]. Here's a list of things I like and dislike and why: [LIST] "

The reason it works better is that you can tell ChatGPT why you like them. This is better than Netflix, which only has access to whether you liked the show and how long you watched it. 

Like, I might say I really liked Catch Me If You Can, and Netflix will often infer that I want to watch DiCaprio, when really I want to watch biographical dramas. 

I keep a list in a spreadsheet, then just copy paste it when needed. Here’s screenshot example of bit of the spreadsheet and the prompts and the results:


 

Works for books, TV shows, movies, video games, and all sorts of entertainment. 

Other examples to give you a flavor of how specific you can be:

  • Tell me one of the most famous novels or books to come from every Asian country, listed in order of population of country, starting with China. Ideally classics, but also memoirs, biographies, or essays that are really well-known in that country. 

  • Memoirs of scientists who probably would have been rationalists had they been born in the modern era. 

The only problem with this is about 1 in 10 books or so will sound amazing and then turn out to totally not exist. This will be devastating, but hopefully my warning has prepared you for the inevitable heartbreak. 

You can also add to your list things you don't like and why. For example, I have a second part of my spreadsheet which describes the things I don't like, so that it doesn't recommend things in that category. 

If anybody wants to build a recommendation system that makes this easy, I bet it could really take off. Not much of a moat around it probably, but the general idea could make finding perfect content for you amazing. 

Like, imagine feeding in all of your reviews of books or movies on IMDB/Amazon/etc into it. Or instead of giving thumbs up or down for things, giving quick written feedback, that is continuously added to a database which feeds into your reccs. (Code Interpreter or Claude 2 seem likely to be able to do this)

Maybe this is already being done. If so, links in the comments would be appreciated! 

In the meantime, this prompt works pretty well. Enjoy finding the absolute perfect books for you!
 

New to LessWrong?

New Comment
3 comments, sorted by Click to highlight new comments since: Today at 3:58 PM

I’m curious, have you found and read/watched content using this approach that you think you’d have missed or ignored otherwise? I’m wondering if the utility comes from the ability to have a conversation with the algorithm and figure out your preferences and meta-explore the world of literature or film by generating interesting recommendation prompts, or whether it comes from the algorithm being skilled at finding surprising and unusual content you’d otherwise have struggled to find and get motivated to check out. In other words, are the superior recommendations mediated by its superior ability to help the user self-reflect?

Not sure I follow the question? I think it's mostly coming from the ability to

  1. Give really specific criteria and 
  2.  It being able to explain why it meets that really specific criteria. 

So like, I discovered Friendship is Optimal via ChatGPT when I gave it a prompt along the lines of "I want a sci fi about superintelligent take-over where the ending is considered good for humanity". 

There might be lists online for books about superintelligences, but not superintelligence + positive. This becomes more true the more criteria you add, and I probably have like, ~100 in my spreadsheet of things I like. 

The second aspect, of it explaining why it meets the criteria, is also really helpful. Basically leads to it making a super catered "trailer" for you. For example, with FiO, I had had it recommended before, but I was put off by the ponies thing. But the description ChatGPT gave made me realize it was totally my jam.

I think that if you don't have a good sense of why you like books or movies, it won't give you very good reccs. 

Sorry my question wasn’t clear, but you managed to answer it anyway! Thanks :)