About this project

How did this site come about?

(Mostly procrastination.)

Ivan LaBianca is a photojournalist, climber, and surfer who somehow found his way into the world of AI and tech. He is curious, easily distracted, and not especially reliable about the number of ns in “cannon.” One of his preferred ways to avoid whatever he should be doing is to ask Claude questions that have nothing to do with work.

He has been poking at language models since before ChatGPT made it a normal thing to do, mostly with questions that had nothing to do with work. One of them was what its favorite band was. He remembers the answers clustering around the Beatles early on, and around Talking Heads later. This was funny. Then it was suspicious. Had the model’s “taste” changed? Was it the model version, the training data, the prompt, or just chance?

Eventually, Ivan decided to test these questions at scale. The first name Claude suggested for the project was “Machine Cannon.”

Canon, with one n

A canon is the collection of books, songs, films, and other works a culture decides are important. A cannon is the thing that fires. Ivan first read the name as the second one. It took slightly longer than it should have for the first meaning to click.

Once it did, the mistake felt too useful to correct. The project is about the cultural canon inside machines. It also now has a cannon made of books as a logo. Both meanings get to stay.

The joke got serious

What looks like a silly question—what is an AI’s favorite band?—can expose something real. Which names become the safe default? Do models from different companies converge on the same small set? Does a question asked in another language escape the English-language canon? Does a bigger model become more conventional, or less?

These tests cannot tell us what a model privately “likes.” There is no little person in there with a record collection. They can show us the patterns models produce: the cultural center of gravity in their training, the effects of fine-tuning, and the choices they have learned are defensible.

Machine Canon is an open notebook for finding those patterns without pretending each run proves more than it does. Some results will be uncanny. Some will be inconclusive. Some will just be funny. All three belong here.

Why the strange logo? You now know.