Who's in the weights?

How the 13 models score one person

Each model's confidence runs from 0 ("never heard of them") to<br>100 ("sure who they are"). It's not yes-or-no, and it's not one verdict —<br>for the same person, the 13 answers can land anywhere from 0 to 100. Here's one person —<br>a Georgian judoka:

The bars are the 13 models' confidence. Some are sure; several have no idea who<br>he is — pooled together, those make up his strength . Now do this for<br>291 people.

More looked up → better known, but loosely

Each dot is one of 185 people. Across the bottom: how often<br>they're looked up — their monthly<br>Wikipedia pageviews. Up the side: how well the 13 models know them, their 13 confidences<br>averaged into one score.<br>It climbs — household names sit near the top — but loosely: among the rarely-looked-up, the<br>models know some people surprisingly well and draw a blank on others just as obscure.

Pageviews are on a log scale — each step right means about 10× more monthly<br>views, so the famous and the obscure fit on one chart. Hover any dot for who it is, its score,<br>and a Wikipedia preview; click through to the article.

Some models are far more confident than others

Average each model's confidence across the people we ran, and the averages run from one model<br>that's sure of almost everyone (about 90 out of 100, top) down to<br>one that's blank on almost everyone (about 18, bottom). So how high<br>a person scores depends a lot on<br>which model you ask, not just on who the person is.

Bar length = that model's average confidence across those people, 0–100. Hover a<br>model's name for its size and knowledge cutoff.

But they mostly rank people the same way

Scoring people high and ranking the same people high are two different things — a model can<br>hand out higher numbers across the board but still put the same people on top. So set the<br>overall levels aside and compare orderings: line up each model's people from best-known to<br>least-known. By that test the models match moderately well: a typical pair ranks the people<br>about 0.65 alike, where 1 would mean identical orderings and<br>0 means no relation at all. So they mostly agree on who is better-known than<br>whom, even where they disagree on the exact scores. The clear exception is the smallest<br>model, whose ordering barely matches the others.

Your line of work barely matters — except for athletes

You might expect some kinds of people to live in the weights more than others. Mostly they<br>don't: split by occupation, recognition is strikingly flat — six of the seven groups sit<br>within a few points of each other, at broadly similar fame. The one clear exception is<br>athletes, who lag the rest. So apart from sport, what you did for a living barely changes<br>whether the models know you.

Average recognition (0–100) per occupation, across the people we ran; their fame<br>is broadly similar across groups, so this isn't just a fame gap. Athletes (orange) stand out.

A shared name costs you a little

Does sharing your name with other notable people hurt? We compared German footballers whose<br>full name is theirs alone on Wikidata to ones whose exact name is shared by<br>several other notable people, matched on fame — so the only real difference is the name:

The short version

Whether a model "knows" a person tracks how often the world looks them up<br>— loosely. A shared name hurts a bit. And while the 13 models broadly agree on who ranks as<br>better- or lesser-known, they differ enormously in how confident they are overall — so<br>whether a borderline person counts as "in the weights" still comes down to which model you<br>ask.

Every number here describes these 291 people — a deliberately wide<br>spread from famous to obscure, not a random or representative sample. Read them as<br>comparisons, not rates.