Two former OpenAI employees launched a website called In the Weights, attempting to answer an experimental question: how many people does a large model actually "remember" without invoking web searches? This type of testing is beginning to gain practical significance as more users turn to chatbots for information.
Test name recognition using multiple models
The "weights" in the website name refer to the model parameters. Developers Thomas Dimson and Joey Flynn believe that the traditional search function of "searching for oneself" is no longer the only way to measure an individual's online presence; the ability of a model to directly identify a person is becoming another visibility metric.
In the Weights asks different models a question similar to "Who is so-and-so?" and asks for up to 10 results, a brief description, and a confidence score. The website then categorizes similar descriptions and generates a strength score to measure how well the model "remembers" the name.
Rankings can change, and illusions can occur.
The models currently participating in the testing include Grok, Gemini, multiple versions of GPT, Claude, Llama, and some less common models. The results page will also show which models provided answers and which answers may contain illusions or confusion.
For example, TechCrunch author Anthony Ha scored 641, placing him in the top 6% of all names. However, the rankings are subject to change. At the time of publication, actor Macaulay Culkin was in first place, followed by singer Luciano Pavarotti.
The report also mentioned that GPT-5.4 Mini interpreted Anthony Ha as a vague form of name that could correspond to multiple people, rather than directly identifying him as a specific person. Such instances were also flagged as potential hallucinations by the website.
Developers are betting on new visibility in the era of models.
In an interview, Dimson stated that after leaving OpenAI, he and Flynn hoped to work on projects that could reignite creativity. They joined OpenAI following the acquisition of its design firm, Global Illumination.
He believes that by 2026, as traffic continues to shift towards larger models, Google-style vanity search will no longer be the most important objective. Compared to webpage ranking, whether or not information about you is present in the model parameters is becoming a new form of online presence.
The developers also stated that they will continue to investigate why the same model series gives different results, which types of people are easier for different models to "remember", and which people should theoretically have Wikipedia entries but have not yet been created.
