"In the Weights" queries multiple large language models to score how well-represented a real person is in AI training data.
A new website called “In the Weights” lets anyone check whether their name is encoded in the training data of large language models (LLMs) — the billions of numerical values that store what an AI model knows. [1]
The site was built by Joey Flynn and Thomas Dimson, both former OpenAI employees. [1] It works by querying several models to identify a specific person, combining those results, and assigning a numerical “strength score” that reflects how well-represented that individual is across the models. [1]
Scores range up to a maximum of 996, a level currently occupied by figures such as Mozart, Shakespeare, and Taylor Swift. [1] The creators note that appearing in smaller models is harder, meaning anyone who shows up in Meta’s Llama — which has one billion parameters — is considered highly relevant. [1]
The tool surfaces a practical question for developers and researchers building on top of LLMs: how reliably a model can recall factual information about a given person without augmentation tools such as web search. [1] The creators acknowledge several limitations of the approach, including that models can hallucinate biographical details, that name typos reduce scores, and that common names tend to produce less accurate results. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
