A new website called 'In the Weights' allows users to check if they are recognized by large language models. The site queries multiple AI models to determine if a person is stored in their weights, which are billions of numerical values that encode knowledge. If a person appears in the model's weights, it means the model considered them relevant enough during training to recall without external tools like web search. The site combines results from several models and assigns a strength score to indicate how well a person is recognized.
The creators, Joey Flynn and Thomas Dimson, both former OpenAI employees, explain that smaller models make it harder for individuals to appear in results. For example, appearing in Meta's Llama, which has a billion parameters, counts as highly relevant. The site also highlights the limitations of large language models, such as the potential for hallucination of biographical details, typos that can lower scores, and common names often producing less accurate results.
According to the website, the current strength scores for the creators' colleagues, Matthias Bastian and Maximilian Schreiner, are 175 and 262, respectively. The maximum strength score is 996, reserved for well-known figures like Mozart, Shakespeare, or Taylor Swift.
Source: thedecoder
