Former OpenAI team members Thomas Dimson and Joey Flynn have launched In the Weights, a playful AI-focused platform that measures how often a person appears in the "memory" of large language models without relying on web search.
The idea is simple: the site asks several models, including Grok, Gemini, GPT variants, Claude, and Llama, to identify a person and provide short descriptions. It then groups similar answers and assigns a strength score, creating a ranking that reflects how strongly an AI system seems to "know" someone.
According to the creators, the project is meant to explore how identity is represented inside model weights, the numerical parameters that shape AI behavior. The result is part experiment, part digital mirror, and part conversation starter about how people are surfaced in an era where chatbots increasingly shape discovery.
The site has quickly attracted attention for its leaderboard-style format and retro-inspired design. Some names score surprisingly high, while the results also reveal how different models can produce inconsistent or ambiguous descriptions for the same person.
Dimson says the project was also a creative reset after leaving OpenAI, and that the team is now interested in deeper questions around model bias, naming patterns, and why some people are more easily recognized than others. As AI systems become a larger part of how information is organized, tools like this may help redefine how digital presence is measured in the future.