In the Weights Tests Whether AI Models Recognize Your Name

In the Weights is a name-recognition experiment that asks several AI models who a person is, compares their answers, and produces a score based on how consistently they identify that name. The result can be entertaining and revealing, but it should not be treated as a scientific measure of fame, reputation, or importance.

What In the Weights measures

The project, created by Thomas Dimson and Joey Flynn, takes its name from the numerical “weights” that encode patterns learned by an AI model. Its central question is simple: can a model identify someone from its learned knowledge without relying on a live web search?

That is different from searching a name on Google. Search results reflect pages currently available and ranked on the web, while an AI model may respond from information represented in its training data. A correct answer suggests that the model encountered enough consistent information to associate the name with a person; it does not prove exactly where that information came from.

How the Strength Score is produced

The tool sends a name-based prompt to a selection of models from families such as GPT, Gemini, Claude, Grok, and Llama. It asks each model for possible identities, short descriptions, and confidence estimates.

The system then groups similar answers and looks at how strongly the models agree. It presents the result as a Strength Score: a higher score generally means that more models identified the same person with greater confidence.

How to interpret a result

A score is best understood as a snapshot of responses from the models tested at that time. It is not a direct reading of a model's internal parameters, and it does not measure social-media followers, web traffic, professional reputation, or public sentiment.

The detailed model responses are often more useful than the headline number. They show whether the models agree on one identity, confuse people with similar names, or invent unsupported biographical details. Those errors illustrate why AI-generated profiles should be checked against reliable sources.

Why scores and answers can differ

AI models are trained on different collections of data and are updated on different schedules. Even versions in the same model family can produce different answers. Common names, limited public information, conflicting sources, and ambiguous spellings can also reduce consistency.

A low score therefore does not mean a person has no digital presence. Likewise, a high score does not guarantee that every detail in the responses is correct.

What the experiment is useful for

In the Weights offers a quick way to compare how chatbots represent public identities and where their knowledge breaks down. It can highlight mistaken identity, uneven coverage, and hallucinated details. Used with those limits in mind, it is an interesting demonstration of how differently AI systems can answer the same apparently simple question.