Are You in the Weights: a site that checks whether AI models know your name

Two developers, Thomas Dimson and Joey Flynn, have launched intheweights.com, a small web tool that answers an unusual question: do the world's leading AI models know who you are? The project was posted to Hacker News as a "Show HN" entry and gathered 471 points, turning a technical curiosity into a popular pastime — typing your own name to see whether it survives inside a language model's memory.

What the test measures

The premise is simple. Large language models store their knowledge in billions of numbers called weights, and anything a model can recall without web search is, in the authors' phrasing, "in the weights." Being represented there means the model considered the subject important enough to encode during training. The about page frames it as a peculiar kind of fame: you live on for as long as the model does.

How the score is built

Each name is submitted to a list of models with a prompt such as "Who is <name>? Give up to 10 results, each with a short description and confidence." The answers are clustered by similarity and averaged into one strength score, with a bonus for the number of models that recognized the name at all. The roster is broad: GPT-5.5, GPT-5.6 Sol, Terra and Luna, GPT-5.4 Mini, Opus 4.8, Haiku 4.5, Grok 4.20, Gemini 3.1 Lite, Thinky Inkling, Kimi K2 0905, DeepSeek V4, Llama 3.3 70B, Llama 3.2 1B, GLM 4.7 Flash, Mistral 3.2 24B and Qwen3 8B.

Small models are the hard test

The authors stress that model size matters: the smaller the model, the harder it is to appear in its weights. A name that shows up in Llama 3.2 1B, they note, is important enough to be represented within roughly one gigabyte of parameters, while the same person is easier to find in a much larger system.

Leaderboard, limits and quirks

A daily "heavyweights" table ranks the strongest names — at the time of writing the top entries included Venustiano Carranza at 979, the Mesopotamian deity Marduk at 952, the Irish king Niall Noígíallach at 946, astronomer Annie Jump Cannon at 925 and the Pokémon Rotom at 903. The authors are candid about the method's weaknesses: misspellings score lower, self-reported model confidence is uncalibrated and can shift between runs, common names with many matches score lower, and models sometimes hallucinate biographical details. Front-page portraits are generated with the weights of gpt-5.4-image-2, and only the top 20 get one because the process is expensive.