Are you in the weights? Testing whether AI knows your name
Category: AI & Gaming Trends
intheweights.com is a web tool that has been making the rounds in tech and gaming communities lately. You type in a name, it queries several AI models, and it tells you how strongly those models recognize that name from their training data. A score from 0 to 1000.
That’s the whole thing. But it turns out to be oddly compelling once you start running names.
How it works
You enter a name. The tool queries GPT-4, Claude, and a few others. Each model returns a weight score — a number that reflects how often that name appears in its training data and how confidently the model can associate it with real context. A high score means the AI has seen this name thousands of times in coherent, meaningful text. A low score means the model has no real picture of who this person is.
Gandhi scores 996. Rocky Balboa scores 986. Your coworker probably scores somewhere around 0.
What I tested
I ran a mix of names: historical figures, musicians, fictional characters, tech people, and a few gamers.
The obvious ones landed where you would expect. Gandhi (996), Dolly Parton (993), Angelina Jolie (992), W.B. Yeats (990), John Coltrane (990). These are names that appear in millions of documents — biographies, articles, academic papers, fan sites, Wikipedia entries. The model has absorbed them at a deep level.
The fictional characters were the surprising part. Stewie Griffin scores 986. Rocky Balboa scores 986. Daft Punk scores 986. They aren’t just famous — they’ve generated enough coverage, fan writing, and forum discussion that the model treats them as fluently as it treats historical figures.
The uncomfortable part: you can test living people, public figures who haven’t done anything particularly notable, or names from niche communities. Some of those scores are sobering.
Why this matters for gamers
Game characters score surprisingly high. The more documented a game world is wikis, reviews, forum threads, lore debates — the higher its characters score. A major franchise character from 2015 probably outscores a minor historical figure from the 18th century, purely because of text volume.
This has real implications for how AI models understand and discuss games. A character with a 950+ score is someone the model can reason about confidently — plot, motivations, connections to other characters. A character scoring below 500 might get hallucinated details because the model is guessing.
Worth trying
Go in with a mix of names you actually care about: real people, game characters, musicians, obscure figures from your interests. The gaps between expected and actual scores are where it gets interesting. A name you assumed was universally known might score 400. Something you thought was niche might sit in the 900s.
It’s a window into what these models actually absorbed from the internet — not what we told them to learn.