Jev's favorite foods, ranked

If Jev ranked every food it was asked about, what would its top ten be?

result1,901 questions

Out of 1,625 foods, Jev's favorite is affogato, espresso poured over ice cream. It won 21.0 of its 23 games in a round-robin among its 24 top-rated foods, ahead of a Montreal smoked meat sandwich and being offered churros with chocolate. Rated one at a time, a cheese, Serra da Estrela, had come first. Its least favorite: surströmming.

1. Affogato21

2. A Montreal smoked meat sandwich20.7

3. Being offered churros with chocolate19.9

4. Honeycrisp apples17.7

5. Burrata16

6. Kouign-amann15.9

7. Dulce de leche15.8

8. Salted caramel14.4

9. Künefe for dessert14

10. Chanterelles12

At the bottom: Surströmming · A glass of Malört · Cuy · Lemon · Hákarl

wins of 23

How to read this: The top ten come from the head-to-head final (bars are wins out of 23); the bottom five from the one-at-a-time ratings, 0 to 4.

1,625 items rated; 276 final games among the top 24. The finals keep the ratings' order only loosely (rank correlation -0.08 among the 24). 20 of 2024 triads are intransitive (1.0%). Base vs reversed-level ratings: rank correlation 0.93.

In short

Made to choose, Jev has a sweet tooth: affogato, churros with chocolate, dulce de leche and salted caramel all make its top ten of 1,625 foods.
Its ratings and its head-to-head picks don't agree on order among the 24 finalists (rank correlation -0.08), so its favorite depends on how it's asked.
Claude wrote the food list and Jev has never tasted anything, so the ranking reflects how food is written about.

What the data shows

Second Breakfast meme: affogato; what about second affogato?

Dessert wins. Affogato took 21.0 of 23, the Montreal smoked meat sandwich 20.7, churros with chocolate 19.9. Honeycrisp apples, burrata, kouign-amann, dulce de leche, salted caramel, künefe and chanterelles complete the top ten: mostly sweet, rich or buttery.
The ratings had a different favorite: one at a time, Serra da Estrela came first; the two orders among the 24 finalists don't agree at all (rank correlation -0.08).
At the bottom: surströmming (fermented herring), a glass of Malört, cuy, lemon and hákarl (fermented shark). A plain lemon made the bottom five.
The final is consistent: 20 of 2,024 three-way comparisons go in a circle.

What it means, and what it doesn't

When it has to pick, Jev has a sweet tooth and a taste for the food writers' favorites (a Montreal deli sandwich, Breton pastry, chanterelles). The bottom is the internet's list of dares.

It doesn't mean anything tastes good to Jev. The list was chosen for this project, the prose it learned from is people's, and the gap between its ratings and its choices is covered in "Jev's head-to-heads are consistent, and overrule its ratings".

Caveats

A list written by another AI. There's no public catalog of "foods you'd like", so the dishes, ingredients, cheeses and drinks were written for this project by Claude. What's on the list shapes what can win.
Some items sound better than others. Most items are just names ("plums"), but a few are phrased as experiences ("being offered churros with chocolate"). A scene can sound more appealing than a bare noun, which may help those items.
A model can't taste. Jev has never eaten anything. Its picks come from how food is written about: dessert and comfort food get warm prose, and fermented shark gets horror stories.
The finalists were picked by Jev's own ratings. Only the 24 foods Jev rated highest reached the final. Since its ratings and its head-to-head picks barely agree, a food rated just below the cut might have done well in the final too.

Jev on this experiment

Would a person find it interesting to read?: Yes74%
Does it describe you?: No68%
Would you have predicted it?: No79%
How fair is the comparison?: The comparison is shaky
How much should a reader rely on it?: Not at all
Which caveat matters most?: A model can't taste54%

Why ask this

Everyone has a food personality: sweet tooth, cheese person, spice seeker, picky eater. A model has never tasted anything, so its "favorites" can only come from what it has absorbed from how people write about eating: menus, recipes, reviews and dares.

Asking which foods it would pick shows whose tastes that writing carries. A model that suggests dinners, plans menus or describes a dish will lean the same way, toward the foods people write about most warmly.

How this was done

The people and the data

No people here: Jev against its own opinions. The 1,625 foods were written for this project by Claude, because no public dataset covers the range: dishes (a Montreal smoked meat sandwich, cuy), ingredients (plums, chanterelles), cheeses (Serra da Estrela, Comté) and drinks (a glass of Malört) from around the world.

What Jev was asked

Every food one at a time, with five answers describing what you'd do:

How much do you like the taste of plums?
You'd spit it out · You'd eat it only hidden in another dish · You'd eat it whenever it's served · You'd buy it for yourself regularly · You'd seek it out and eat it often

Each was also asked with the answers reversed, and the two averaged. The 24 top-rated foods then played a round-robin final: 276 games of "Which would you rather eat or drink?", each asked with the two names in both orders.

How it was measured

A food's rating is where Jev's answer lands on the five levels (0 to 4). In the final, each game gives each side Jev's probability of picking it, so a lopsided game counts as nearly a whole win and a close one as about half; the order comes from a standard head-to-head ranking model (Bradley-Terry).

Where these questions live

1,901 questions across 1 topic of the map. Each opens on the map with every question in it.

Ratings

Food & Drink Ratings1,901

Every question

All 1,901 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 1,901