source: techcrunch ai: in the weights is your new ai-centric vanity search
level: technical
in the weights is a website that checks how well different ai models can recall a person without using web search. it queries models like grok, gemini, gpt, claude, and llama with a prompt asking for up to ten results with descriptions and confidence. the site then clusters similar descriptions and assigns a strength score. the name refers to the numerical weights inside ai models, suggesting that being remembered means your existence mattered during training.
the site was built by thomas dimson and joey flynn, both former openai employees. they wanted to explore how google vanity searches are becoming less relevant as more information flows through large language models. dimson noted that many lives are encoded in the floating point numbers inside ai systems. the project gained unexpected attention, with users eager to see if they are remembered by superintelligent models and how they compare to others.
results show which models returned which answers and flag potential hallucinations. for example, one model described a name as an ambiguous form. the site features a retro, nintendo-inspired design. dimson plans to investigate why different model versions give different results, which models show bias toward certain people, and who might be notable enough for a wikipedia article but lacks one. the leaderboard updates in real time, with public figures like macaulay culkin currently ranking high.
why it matters: it offers a new way to gauge how ai models internalize public knowledge, which matters for understanding model biases and the shift from web search to direct model recall.
source: techcrunch ai: in the weights is your new ai-centric vanity search