On
In the Weights: A New AI Tool That Measures How Well ChatGPT Remembers You

For years, Googling your own name has been the standard way to check your internet presence. But now that AI has become millions of people's go-to information source, the whole idea of "looking yourself up" is shifting. More people are asking ChatGPT, Gemini, Claude, and Grok about individuals instead of visiting traditional websites. This raises an interesting question: if AI is becoming the primary gateway to information, do these models actually know who you are?

That's exactly what In the Weights sets out to answer — a new project designed to measure how much of a person's information has actually made it into AI model training data.

What Is In the Weights?

In the Weights was built by Thomas Dimson and Joey Flynn, both former OpenAI employees. The name comes from the weights in AI models — the numerical parameters that determine how a model learns and generates responses.

The tool's goal is straightforward: measure whether an AI model can recognize or recall a specific person without relying on external search tools.

The project's description captures the lighthearted spirit perfectly:

"Being present in the weights of an AI model essentially means your existence was significant enough to shape the development of artificial intelligence."

While more playful than scientifically rigorous, the concept quickly caught the attention of the AI community.

How Does It Work?

To generate your score, In the Weights sends queries to multiple AI models, including Grok, Gemini, various GPT versions, Claude, Llama, and several lesser-known models.

Each model receives the same query:

"Who is <this person's name>? Provide up to 10 results with brief descriptions and confidence levels."

The system then groups similar responses, analyzes how consistent the models are with each other, and calculates a metric called the Strength Score.

In simpler terms: a higher score means more AI models correctly identified the person with strong confidence.

What's interesting here is that In the Weights doesn't measure fame the traditional way — it's not counting social media followers or website traffic. Instead, it reflects how deeply information about a person is embedded in training data and learned by AI models.

The rankings shift constantly. When the project launched, actor Macaulay Culkin (Home Alone fame) was competing with opera legend Luciano Pavarotti for the top spot, both scoring near the maximum threshold.

The real appeal is that you don't just see a final score — you can examine how each individual AI model responded about you, which often reveals surprising differences.

Another notable feature is how the tool catches AI hallucinations. One model, for instance, described "Anthony Ha" as a vague abbreviation that could refer to multiple people, rather than correctly identifying the specific individual being asked about.

These kinds of errors highlight that despite their sophistication, AI models still struggle with accurate person recognition. Different models train on different data sources, leading to wildly different results. That's precisely why In the Weights fascinates the AI community — it exposes how these models actually "see" the world.

What's Next for In the Weights?

The developers initially thought this would be a fun curiosity tool. But demand revealed that plenty of people genuinely want to know whether they exist in AI memory.

Not everyone's convinced, though. Some AI researchers argue In the Weights is essentially just querying multiple chatbots with the same question and aggregating responses. AI researcher Anthony Moser bluntly described it as "asking 13 different chatbots what they know about you."

The development team says they're digging deeper into the collected data. Their research questions include:

  • Why do different versions within the same AI model family produce different results?
  • Which models tend to remember certain demographic groups more accurately?
  • Are there people notable enough to deserve Wikipedia entries but who aren't currently remembered by AI?

These analyses could reveal fascinating insights into how AI models form knowledge and where biases exist in training data.

In the Weights isn't a precise fame-ranking tool or a scientific measure of notoriety. But it does reflect an emerging reality: as chatbots become primary information sources, existing in an AI model's "memory" might become a new form of digital presence.

Being recognized by AI may not be the path to immortality that some joke about. Still, the project raises a genuinely important question about the future of the internet: in the age of AI, will our digital footprints be stored on websites, or embedded in the weights of massive language models?

Related Articles