Two former OpenAI employees have built a website that answers a question many people have wondered about but few could measure: do AI models actually know who you are?
The site is called "In the Weights," and according to The Decoder, it reveals which people AI models can recall purely from their training data. In other words, it tests whether a person is embedded deeply enough in what a model has learned that the system can summon details about them without looking anything up.
The tool assigns a strength score of up to 996, indicating how deeply a given person is baked into the model. The higher the number, the more thoroughly that individual is represented in the AI's underlying knowledge.
The Decoder reports that the most deeply embedded figures are unsurprising household names. Mozart, Shakespeare, and Taylor Swift top the list, reflecting how heavily documented and widely written-about those people are across the text that AI models learn from.
The name itself is a nod to how these systems work. An AI model's knowledge lives in its "weights" — the internal values it adjusts as it trains. A person who appears constantly across that training material ends up firmly encoded; someone rarely mentioned barely registers at all.
Why it matters: As AI systems increasingly mediate how we look up and learn about people, a tool that exposes who the models "know" — and who they don't — offers a rare, concrete glimpse into the otherwise hidden contents of their training data.