Note: I'm writing every day in November, see my blog for disclaimers.

I’m hoping that multi-modal embedding models (which convert videos, images, and text into points in a high-dimensional space such that a video of a dog, a photo of a dog, and text describing a dog all end up close together) will allow this statement to be formalised cheaply in the near future.

If you consider the “range” of ideas that are conveyed in books, and the range of ideas that are conveyed in movies, I certainly have the intuition that books seem to be much broader/weirder/different in the ideas they describe when compared to movies. Movies, by comparison, seem muted and distinctly more “normal”.

I believe this to be due to how these two media come to be: movies are enormously expensive, and require the creative input of dozens to ~100s of people (writers, actors, producers, sound & music, lighting, video, VFX, set design, the list goes on). If a screenwriter has a complicated idea and wants it to appear in a film, each of these people has to understand (and importantly, care) about this idea for it to actually arrive on-screen in its original form. This puts an upper bound on how “weird” an idea can be for it to be on the big screen.

I’m less informed about the production of books, but I’m struggling to imagine more than 30 people who could have an impact on the creative aspects of a book. As an upper bound, maybe you have multiple authors, multiple editors, and several people from the publisher. But as a lower bound, I imagine you could write a book with creative input from just the author and the editor. If an author has a unusual idea they want to convey, they just have to make sure their one editor doesn’t dilute the core paragraphs too much, and this idea can then go straight into the reader’s noggin.

You might quibble with my “100s of people” characterisation for movies, but that’s okay because this exact number doesn’t matter, it just matters that most movies involve more people than most books. Because the distinction isn’t really movies-vs-books, it’s about the number of people involved: The more people involved, the trickier it is to get slippery ideas through them all, and so the more “normal” your ideas have to become.

You can see this by looking at low-headcount “movies” (YouTube videos, indie films, short films) which tend to be more out-there and weird, and also by looking at high-headcount “books” (pop-sci, magazines, newspapers) which tend to have more in-distribution ideas. You can even expand this to other creative media such as podcasts, theatre, poetry.

Creative media are ideal for doing this sort of headcount-vs-overton-window analysis, because there’s typically very little that fundamentally constrains what the message can be in creative media.

The headcount-overton-window tradeoff also applies to startups, organisations, political campaigns, public movements, although these have more constraints than creative media, such as profitability, existing in the physical world, convincing investors. But there’s a reason why “donate to save this puppy” is an easier pitch than “donate to save these shrimp”.

Discussing memetics is always tricky, but I do believe this tradeoff is real, and that it effects the objects we see in the world. We can draw informative conclusions from this tradeoff: to find unique ideas, you must consume content that required very few people’s approval in order to exist (e.g. blogs, preprints, off-the-cuff discussions with smart people). To produce unique ideas, you will be fighting an uphill battle if you subject that idea to other’s opinions. Sometimes this battle is worthwhile; I said the ideas would be unique, not good. But every novel idea you introduce is something that needs to fight for space in someone’s mind, and you’ll need to convince them not to project your idea onto their existing map of the world. Every introduction is a chance for failure of communication (which may go unnoticed by you). But also, every introduction is a chance for someone to cull your bad ideas.

So when you feel in a rut, and lacking interesting thoughts, consider the media you consume. Are you sampling from the same distribution as everyone else? By how much has the media you consume been filtered (either anthropically or algorithmically)? Books, for me, are second to none for novel ideas, and I look forward to being able to demonstrate this empirically.