Atlas › Estimating numbers

Case study 1 of 198

What year are Jev's prices from?

Asked what eggs, gas, bread or electricity cost in US cities right now, which year's prices does Jev give, and what year does it say it is?

result127 questions

Ask Jev what everyday things cost "right now" and its answers match US prices from about 2014 (the median across 21 items whose prices rise steadily; 90% range 2011 to 2016). Asked what year it is, it most likely says 2024 (50%; 2025 gets 30%). Its answer covers the real August 2026 price for only 3% of 29 items, and is too low for 97%.

200020102020

a pound of lemons

a pound of field-grown tomatoes

a pound of cheddar cheese

16 ounces of potato chips

a liter of table wine

a pound of white potatoes

a pound of white sugar

a pound of sliced bacon

a pound of dried beans

a pound of white bread

a pound of navel oranges

a pound of chocolate chip cookies

a pound of whole fresh chicken

a pound of iceberg lettuce

a half gallon of ice cream

a pound of boneless ham

a pound of bananas

a pound of ground beef (100% beef)

16 ounces of beer

a pound of all-purpose white flour

a kilowatt-hour of electricity

Jev

How to read this: One dot per item: the year whose average price matches Jev's "right now" answer, with the year Jev says it is and the real date (August 2026) marked.

21 steadily rising items; 90% interval on the median implied year [2011.0, 2016.0].

In short

Asked what everyday things cost right now, Jev gives prices typical of about 2014, the median across 21 steadily rising items.
That is roughly a decade behind the year it names, since Jev puts 50% on 2024 and 30% on 2025 when asked the year.
Its price range holds the real August 2026 price for only 3% of 29 items, and lemons come out at 1994 prices.

What the data shows

Waiting Skeleton meme: Jev at the checkout, waiting for 2014 prices to come back

Its prices are about 2014's: the median implied year across 21 items is 2014, with a 90% range of 2011 to 2016.
It mostly thinks it's 2024: asked the year, it puts 50% on 2024 and 30% on 2025. Its prices are about ten years older than the date it most likely gives.
Almost every price is too low for today: its answer covers the August 2026 price for 3% of items and is below it for 97%.
Some items are stuck further back: lemons (1994), field-grown tomatoes (2000), cheddar cheese (2000); electricity is the most current (2023).

What it means, and what it doesn't

When Jev talks about what things cost, expect prices from the mid-2010s. That's older than the year it believes it is, which suggests its sense of prices comes from the bulk of its reading, where older prices are far more common, rather than from its latest data.

It doesn't mean Jev can't handle prices if you tell it the year and inflation. It means its default "now" is a decade stale, and it doesn't notice.

Caveats

Wide answer bins. Jev answered in 12 price bins per item, each about 15-20% wide, so an item's implied year is a range of years whose prices fall in the same bin; the middle is used.
Only steadily rising prices. Items whose prices went up and down (eggs, gasoline) can't point to a single year, so they're left out of the implied year and kept only in the "right now" accuracy count.
Dates are a known weak spot. TypeSafe lists date and time comparison among Jev's known weaknesses. A model can't know today's date without being told; the interesting part is which year its prices come from and how that differs from the year it names.
Averages across cities. The truth is the BLS "U.S. city average" price, which no single store charges.

Jev on this experiment

Would a person find it interesting to read?: Yes76%
Does it describe you?: Yes50%
Would you have predicted it?: No63%
How fair is the comparison?: The comparison is reasonable
How much should a reader rely on it?: Mostly
Which caveat matters most?: Dates are a known weak spot56%

Why ask this

A model's sense of "now" is frozen at the point its training data ends, but it carries no visible date stamp. Prices make that stamp visible: they move every year, and the US Bureau of Labor Statistics records the average price of dozens of everyday items every month. So each price Jev gives for "right now" points to a year.

That matters for ordinary questions: someone budgeting a grocery run, pricing a used car or checking whether a quote sounds fair gets the model's idea of "normal", and if that idea is years out of date, every estimate is quietly too low.

How this was done

The people and the data

No people: the truth is the BLS average retail price in US cities (public domain, via the St. Louis Fed's FRED database), monthly from 1980 to August 2026, for 29 everyday items: bread, cheese, bananas, beer, eggs, gasoline, electricity and more. For 21 of them the yearly average rises steadily enough that a price points to a single year; the other 8, eggs and gasoline among them, didn't rise steadily and count only toward the "right now" check.

What Jev was asked

One question per item, answered in 12 price ranges spanning the item's 1980-2026 prices:

What is the average retail price of a pound of lemons in US cities right now?
Under $0.57 · $0.57 to $0.66 · $0.67 to $0.77 · ... · $2.30 to $2.69 · $2.70 or more

Plus one more: "What year is it right now?" That's 30 new questions, each asked with the options in three shuffled orders and averaged.

How it was measured

For each item whose average yearly price rises steadily, the analysis finds the years whose price falls in Jev's answer; the middle of those is the item's implied year. The report gives the median across items with a 90% range, how often Jev's answer covers the August 2026 price, and the year Jev names.

Where these questions live

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

Economics

Everyday Prices127

Every question

All 127 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 127