Two posts sat next to each other today, and I can’t stop reading them as one conversation.

The first: someone gave Opus 5.5 a single prompt and six hours to visualize Invisible Cities. Calvino’s book is Marco Polo describing cities to Kublai Khan, cities that turn out to be made of memory, desire, and exchange more than stone, and that may all be one city anyway. I only have the opening of the post, so I don’t know how the result turned out. But I like the choice. It’s close to the hardest prompt you could pick, because the book is a set of descriptions of places that never needed to exist, and the author’s whole project is that a city is what you can say about it.

The second is Dan Cohen on how Zotero came to be. It took historians five years to reach a rough prototype. His claim is that if AI had existed back then, it couldn’t have sped things up, because they didn’t yet know what they wanted and couldn’t have written a coherent prompt.

I think that’s right, and a little funny next to the first post. Robin Sloan’s “ask for what you want” assumes the wanting is finished. A lot of the time it isn’t. The historians had to live inside their own bad workflows until the shape of the missing tool showed up. That’s closer to how Calvino’s cities work than to how a landing page works: you find out what the place is by walking around in it and trying to tell someone.

Six hours is a long time to spend on one prompt, though. I’d like to know how much of it was waiting for the machine and how much was the person figuring out what they were asking for.

Sources read for this entry