top of page
A preposterous new way to find something to watch
Most film discovery experiences are based on what you like—algorithms look at what you've watched in the past and try to recommend more of the same. Criterion Connection is more about what you're like—fill in a few details about your life and it'll find the cinematic masterpiece that's a perfect match.
This is a novel way to explore a deep film catalog (almost 1,300 titles), and it uses the same kind of data most discovery algorithms do, but it flips the experience upside-down and puts the viewer first and the viewing material second. Give it a try and let me know what you get!
Here's how I made it:
I found a dataset of Criterion Collection films online, with information like the titles, directors and short descriptions of each film. But in order to make the kind of discovery experience I had in mind I needed more data. Specifically, I needed detailed information about each film. So working with Chat-GPT (I just call it "G" at this point) I created categories of 50 film attributes (types of themes, characters, directing styles, etc). I then sent GPT the data I already had (title, director etc) and asked G to rank each film according to the 50 category system we'd devised. This was a very large amount of information going back and forth between me and G. Too much to handle via Chat-GPT.
To send all this data back and forth, I used Python to run a custom script that pinged the GPT api and received the detailed information I needed. I have no idea how to write Python code, but G does. So I briefed G and G gave me some code I could use to automate the ranking system. A few hours later (and lots of back-and-forth code editing), G had analyzed all 1,300 films and added ranking information to my database.
If you've spoken with me about generative AI in the last couple of years, you've likely heard me talk about how important and overlooked data is to the generative creative process. I'm obsessed with the data that powers AI generation, and this project is a great example of how getting the data right leads to more and more interesting and connective experiences.
bottom of page