I have a few personal AI projects in addition to my business ones. I first wrote about these projects around four months ago, but they are further along now. The simplest is a storage list of all of the spices and sauces that I keep in my kitchen so that I can get recipe ideas easily. Then if I like the recipe, I save it or tweak it and slowly I’m building my own little personal cookbook. The feedback loop for this is hardly perfect as I don’t always provide the follow up information, but it still serves the basic purpose well.

This post is about a more ambitious project though, one that incorporates my feedback and preferences across different forms of media with the goal of picking up on common threads and uncovering new things that I’d like. We’re talking movies, TV, music, podcasts, and different types of books. The initial seeding is the heaviest lift, one that is still in process. For some categories, it was a matter of downloading history. Those were simple. For others, it was scanning bookshelves and combing through my memory. I continue to fill in some of that back catalog in fits and starts. I also have to provide commentary and feedback on what I liked, what I didn’t like, and why. That’s been a use case for the dictation tool we use as I just ramble for a few minutes and then submit it to be synthesized.

I’ve used the media project for a couple of books already, a few that I read before my Slovenia trip and my current nighttime read. Music has been hit and miss thus far, but I have found a few new subgenres that I’ve enjoyed (Cabo Verde ballads anyone?). This week, I used the accumulated information to help pick a podcast series for my drives, a new television show to watch, and a few nonfiction books to read. The immediate impetus was that I finished a show—season two of The Chestnut Man—and needed a new one. It was almost as gripping as the first season, though definitely not bedtime viewing for many people as it is a fictionalized serial killer drama based on a Danish novel. I also haven’t been reading as many idea books in the last year or so and want to get back into that as an antidote for brainrot.

For the TV quest I gave some initial guidance on what I wanted my next show to be, namely another Nordic noir. The chat then gave a few recommendations of ones I hadn’t watched and gave short synopses of why it thought I might enjoy them based on what I’ve already reviewed. I’m two episodes into the one I selected, but it’s enough to keep going even if I have to access it through an app available in conjunction with my library card. This round is also the first time that the cross-pollination has been useful, even if still nascent. This was particularly true in some crossover between books and podcasts, though I also added some layers of refinement to the instructions when it recommended books based on some of my movie preferences but that don’t fit with my reading goals. With each round of refinement and each additional bit of context the results should continue to improve, and I expect that after a few more rounds it will be a really useful toy.