Website "In the Weights" shows whether AI models know who you are

2026-06-19 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Fundamental Awareness, quick

Summary

The website "In the Weights" identifies individuals whose information is encoded within the numerical weights of large language models, indicating their relevance during training. This platform queries multiple AI models, aggregates the findings, and assigns a "strength score" to each person. For instance, scores range from 175 for Maximilian Schreiner and 262 for Matthias, up to a maximum of 996 for highly prominent figures like Mozart, Shakespeare, or Taylor Swift. Developed by former OpenAI employees Joey Flynn and Thomas Dimson, the site notes that appearing in smaller models, such as Meta's Llama with its billion parameters, signifies particularly high relevance. The creators acknowledge limitations, including the potential for models to hallucinate biographical details, the negative impact of typos on scores, and generally poorer results for common names.

Key takeaway

For AI scientists and data privacy professionals evaluating model training data and potential biases, "In the Weights" offers a unique lens into how deeply individuals are embedded in LLM knowledge. You should use this tool to assess the prominence of specific persons within various models, recognizing that smaller models indicate higher relevance. Be mindful of the site's noted limitations, such as hallucination risks and score inaccuracies for common names, when interpreting results for data governance or ethical AI considerations.

Key insights

The site reveals how deeply individuals are embedded in LLM training data, quantified by a "strength score".

Principles

• LLM weights encode knowledge of relevant individuals.
• Smaller models imply higher relevance for included individuals.
• LLMs can hallucinate biographical details.

Method

The "In the Weights" site queries several AI models, combines their results for a specific person, and then assigns a cumulative strength score based on these findings.

In practice

• Check personal "strength score" on "In the Weights".
• Evaluate LLM data retention for specific individuals.

Topics

• Large Language Models
• Model Weights
• Data Retention
• AI Hallucinations
• Personal Data
• In the Weights

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Scientist, Tech Journalist, General Interest

Related on AIssential

• Memory in LLMs: Weights and Activations - Jack Morris, Cornell — Current LLMs struggle with niche knowledge and context scaling; training knowledge into weights offers a path beyond RAG's limitations.
• When large language models are reliable for judging empathic communication — LLMs reliably judge empathic communication, matching experts and outperforming crowdworkers when properly benchmarked.
• Re-Centering Humans in LLM Personalization — LLM personalization evaluations relying on synthetic data significantly misrepresent real-world human experience and model limitations.
• Your AI Doesn’t Know Anything. And That’s Not the Model’s Fault. — AI hallucinations stem from infrastructure limitations, not model intelligence, necessitating robust knowledge layers like graphs.
• I created an LLM trained solely on Jeffrey Epsteins emails to see how messed up it becomes :) — Fine-tuning an LLM on a controversial, unannotated dataset can reveal unique behavioral shifts and spark technical debate.
• Contribution Weights: A Geometrical Analysis of Self-Attention Transformers — Contribution Weights offer a geometrically-informed, more faithful measure of token importance and reveal active roles for attention sinks in LLMs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.