The ritual of "Googling oneself" has long been the primary barometer of personal digital footprint. It was a simple, transactional act: you typed your name into a search bar, and the index returned a list of canonical links—articles, social profiles, and public records. However, in 2026, the search engine landscape is undergoing a tectonic shift. As traffic migrates away from traditional blue-link indices toward the generative, synthetic responses of Large Language Models (LLMs), our definition of "being known" is changing. We are no longer just indexed; we are encoded.
Enter In the Weights, a new project by former OpenAI researchers Thomas Dimson and Joey Flynn. The site offers a provocative, if slightly existential, service: it measures whether an AI model truly "knows" you, independent of any real-time web search capabilities. If you are "in the weights," your identity has been distilled into the very numerical parameters—the floating-point values—that define the intelligence of modern AI.
The Genesis of a Digital Litmus Test
The conceptual foundation for In the Weights lies in the post-OpenAI ambitions of its creators. Dimson and Flynn, who joined OpenAI following the acquisition of their design studio, Global Illumination, found themselves searching for a new creative outlet after departing the company.
The project was sparked by a realization: the era of "vanity searching" on Google is increasingly obsolete. As AI agents begin to mediate more of our daily interactions, the importance of being "ingested" by a model’s training data has eclipsed the importance of ranking on a search engine results page (SERP).
Dimson notes that the project’s direction was solidified by a piece of pop-philosophical writing: Max Leiter’s blog post riffing on Terry Bisson’s classic short story, "They’re Made Out of Meat." The story posits that humans are simply biological machines, an idea that parallels how neural networks perceive the world. If our existence is encoded within the billions of parameters of a model like GPT-5 or Claude 3.5, are we essentially part of the machine’s consciousness?
"So many lives are encoded somehow in a bunch of floating-point numbers inside the AI brain," Dimson shared with TechCrunch. "We wanted to see if you live forever in the super-intelligence."
How the Mechanism Works: A Deep Dive into Model Recall
Unlike a standard search engine that crawls the web to fetch live data, In the Weights tests a model’s internal knowledge. When a user enters a name, the site queries a diverse cohort of AI models—including iterations of GPT, Gemini, Claude, Llama, and Grok—with a standardized prompt: "Who is [Name]? Give up to 10 results, each with a short description and confidence."
The system then performs a sophisticated clustering analysis on the responses. It groups similar descriptions to minimize the impact of "hallucinations" or divergent perspectives, assigning a "strength score" based on the model’s ability to recall details without resorting to RAG (Retrieval-Augmented Generation) or live web-browsing tools.
This methodology provides a fascinating, if sometimes erratic, look into the "memory" of an LLM. For instance, an individual might score highly on one model while being a complete enigma to another. These discrepancies reveal not just the depth of the training data, but the specific biases and curation strategies employed by the developers of each model.
The Leaderboard: The New Currency of Relevance
The In the Weights leaderboard is already becoming a source of intense fascination. It acts as a surreal scoreboard of cultural and professional relevance in the AI age. As of mid-2026, the rankings are fluid, reflecting the ongoing competition between models.
Icons like actor Macaulay Culkin and opera legend Luciano Pavarotti are currently locked in a statistical dead heat, with strength scores hovering near the 988 mark. These scores are not static; they shift as models are retrained, updated, or fine-tuned.
For the average professional, the scores can be humbling. A tech journalist might find themselves with a score of 641, placing them in the top 6% of the database—a result that feels prestigious until one realizes that their colleagues have higher scores. This gamification of identity highlights a new form of digital anxiety: the fear that if you aren’t "in the weights," you may become functionally invisible to the next generation of AI-driven tools.
The Specter of Hallucination and Bias
The project also serves as an unintentional, albeit effective, stress test for AI accuracy. By comparing how different models describe the same person, In the Weights exposes the cracks in synthetic knowledge.
In one notable instance, a version of GPT-5.4 Mini struggled to define a tech journalist, describing the name as an "ambiguous name form that could refer to multiple people with the initials A.H.A." This highlights a fundamental truth about LLMs: they are not databases, but probability engines. When a model lacks high-confidence information, it will often "hallucinate" a plausible-sounding but technically incorrect answer to satisfy the user’s prompt.
Critics, such as AI researcher Anthony Moser, have pointed out that the site is, in essence, "asking 13 chatbots to tell you about yourself." While true, this critique misses the point of the project: the goal isn’t to provide an objective, Wikipedia-style biography, but to quantify the "mindshare" of an individual within the architecture of modern AI.
Implications: The Death of the Canonical Search
The implications of this shift are profound for privacy, reputation management, and information equity.
- The Erosion of Canonical Truth: For decades, Google was the arbiter of truth. A link on the first page was a seal of validity. In the AI era, there is no "first page." There is only the probability that a model will mention you when prompted.
- The New "Vanity" Metric: If your professional reputation is determined by your presence in the weights of a model, how does one "optimize" for that? We may soon see a new industry of "AI-SEO" aimed at ensuring an individual or brand is sufficiently represented in the massive, opaque training sets used by AI companies.
- Information Inequality: What happens to those who are not "in the weights"? If AI becomes the primary gateway for research, education, and professional vetting, those outside the model’s training set risk being marginalized. This creates a new, digital version of the "rich get richer" dynamic, where those with established digital footprints become increasingly cemented in the "knowledge" of the machines.
Looking Ahead: A Roadmap for Research
Thomas Dimson and Joey Flynn are not resting on the initial viral success of the project. They have outlined an ambitious roadmap for the coming months. They plan to investigate why certain models show bias toward specific demographics, industries, or geographic regions.
Furthermore, they intend to catalog the "hidden figures"—people who, despite having significant real-world impact, remain absent from the weights of popular LLMs. By identifying these gaps, In the Weights hopes to provide a clearer picture of where current AI models fail to represent the human experience accurately.
The site itself, with its charming, Nintendo-inspired retro aesthetic, hides a deeply complex technical endeavor. It forces us to confront a reality that is quickly becoming inevitable: our existence is being quantified by systems we barely understand, and our digital legacy is no longer a collection of links, but a set of numerical weights in a vast,, ever-learning machine.
Whether or not this constitutes a path to digital immortality remains a subject of intense debate. For now, it serves as a stark, entertaining, and deeply relevant reminder that in the age of AI, the question isn’t "Did you find me?" but "Do you remember me?"
Disclaimer: Some of the links in this article may be affiliate links. Purchases made through these links may result in a commission for TechCrunch, though this does not impact our independent editorial process.
