Googling Yourself Doesn't Hit the Same Way Anymore
Anyone who has typed their own name into a search engine recently knows the feeling has changed. The old ritual of the vanity search, typing your own name into Google to see what comes up, used to feel like checking a mirror that reflected the entire internet's opinion of you. These days that mirror feels cracked. Part of it is the well-documented decline of Google search results generally. But there is something deeper going on too: more and more people are learning about you, and about everyone else, from chatbots rather than from a page of blue links.
Two former OpenAI employees, Thomas Dimson and Joey Flynn, noticed the same shift and built something to measure it. Their project is called In the Weights, and in the few days since it launched on June 20, 2026, it has become one of the more talked-about small AI projects of the summer, the kind of niche, slightly absurd tool that ends up in everyone's group chat for a day or two before sparking a much longer conversation about what it actually reveals.
What In the Weights Actually Does
The concept is simple to explain and slightly strange once you sit with it. You type a name into In the Weights, and the tool sends that name to a wide range of AI models simultaneously, including Grok, Gemini, several versions of GPT, Claude, Llama, and a handful of lesser-known models. Each model gets a standardized prompt along the lines of asking who the person is, with up to ten possible results, each carrying a short description and a confidence level.
Crucially, the models are not allowed to use web search or any external tool to answer. They are working purely from what they already learned during training, from the patterns baked into their parameters, the floating-point numbers that give the project its name. In the Weights then clusters the similar descriptions returned by different models together and produces a single composite number called a strength score, a rough measure of how consistently and confidently the AI world, collectively, recalls that name.
"Being in the weights means your existence was deemed important in the process of creating superhuman artificial intelligence." - In the Weights, site copy
Why Querying Many Models at Once Matters
The decision to be model-agnostic, rather than relying on a single AI system, is one of the more thoughtful design choices behind the project. Any individual model can hallucinate, hold biased associations, or simply have gaps in what it absorbed during training. By aggregating responses across many models and looking for where they agree and where they diverge, In the Weights becomes a more useful diagnostic tool than asking any one chatbot the same question. It also surfaces something that a single query never would: which specific model said what, including the moments where a model gets something embarrassingly wrong.
The Leaderboard, the Scores, and the Hallucinations
Part of what made In the Weights spread so quickly is the leaderboard, a running, shifting ranking of the highest-scoring names on the platform. At one snapshot during launch week, actor Macaulay Culkin held the top slot with a strength score of 988, narrowly ahead of opera singer Luciano Pavarotti. The leaderboard moves constantly as more people search names and as the underlying clustering recalculates, which is part of the appeal. It behaves less like a static ranking and more like a live scoreboard for who currently dominates machine memory.
TechCrunch's own Anthony Ha, who wrote the original article introducing the tool, scored 641, putting him in the top six percent of names queried. But his result also surfaced one of the project's most interesting side effects: GPT-5.4 Mini reportedly described Anthony Ha as an ambiguous name form that could refer to multiple people sharing the initials A.H.A., a small but telling hallucination that illustrates exactly the kind of gap In the Weights was built to expose.
| Name | Strength Score | Notable Detail |
|---|---|---|
| Macaulay Culkin | 988 | Held the top leaderboard slot at launch |
| Luciano Pavarotti | Close second | Narrowly trailed Culkin at launch |
| Anthony Ha | 641 | Top 6%; GPT-5.4 Mini hallucinated an ambiguous-name explanation |
The Origin Story: Two Designers, a Meat Joke, and Post-Acquisition Restlessness
Thomas Dimson and Joey Flynn joined OpenAI through its 2023 acquisition of their design startup, Global Illumination. Asked by TechCrunch why he built In the Weights, Dimson explained the project came out of a desire to get the creative juices flowing again after some time inside a large AI lab. He described thinking through the idea that Google vanity searches are increasingly the wrong objective in 2026, as more traffic moves toward LLM-based interfaces, and that so many lives are encoded somehow in a bunch of floating-point numbers inside the AI brain.
Dimson also credited a more specific creative spark: a tongue-in-cheek blog post he had written riffing on AI weights alongside Terry Bisson's classic short story "They're Made Out of Meat," a piece of science fiction that imagines alien explorers struggling to comprehend that intelligent life on Earth is built from meat rather than something more sensible like silicon. That same inversion, the strangeness of intelligence being encoded in an unexpected substrate, whether meat or floating-point parameters, seems to have stuck with him and shaped the final direction of the project.
"So many lives are encoded somehow in a bunch of floating point numbers inside the AI brain." - Thomas Dimson, co-creator of In the Weights
Why a Project Like This Resonates Right Now
The Quiet Collapse of Web Search as the Default Reference
In the Weights is launching at a moment when the assumption that a web search is the definitive record of a person's public presence is actively eroding. Chatbots and large language models are increasingly the front end through which people discover information, and they often shape what someone learns about another person without pointing back to any original source at all. That shift creates a genuine gap in how people understand their own digital footprint: you can no longer assume that the version of you living in someone's chatbot conversation matches the version of you that shows up on page one of Google.
Part Status Marker, Part Genuine Measurement Tool
The project occupies an interesting middle ground between a joke and a genuinely useful diagnostic. On one hand, the framing of being in the weights as evidence that your existence was deemed important in the creation of superhuman AI is clearly tongue-in-cheek, a wink at the absurdity of treating model recall as a status symbol. On the other hand, the underlying phenomenon it measures, the latent representation of real people inside machine-learned systems, is genuinely interesting and largely invisible without a tool like this.
For creators, founders, journalists, and other people who depend on discoverability for their livelihood, the score offers a rough way to check whether their name is consistently associated with their actual work, or whether models confuse them with someone else, attribute the wrong accomplishments to them, or simply fail to recognize them at all. For everyone else, it is closer to a curiosity engine, a novel way of asking whether an AI system has effectively noticed you exist.
What the Score Does Not Actually Tell You
It is worth being clear-eyed about the limits of what In the Weights measures. A high strength score does not mean a model knows the full truth about a person, only that the name is widely recognized within the patterns the model absorbed during training. The tool measures recall, not verified identity, and outputs are entirely dependent on what each model happened to learn, which is itself shaped by whatever data that model's developers chose, found, or were able to license during training.
- It conflates presence with importance. A model recognizing a name is not the same as that name representing someone who matters in any objective sense. Strength score reflects training data frequency, not social or civic significance.
- It can amplify existing biases. If training data overrepresents certain demographics, professions, or regions, the strength score will reflect and reinforce that imbalance rather than correct for it.
- It is dynamic and unstable. As AI companies update their models and refine alignment and safety training, a person's apparent presence inside a model's weights can shift unpredictably between versions.
- It can surface hallucinations as fact. The Anthony Ha example shows how confidently a model can be wrong, describing an actual identifiable person as an ambiguous initialism rather than recognizing them correctly.
AI critic Anthony Moser, quoted in coverage of the launch, was blunt about the limitations, dismissing the effort as essentially asking multiple chatbots the same question and reducing the novelty to simple aggregation. That criticism is fair as far as it goes, though it perhaps undersells the value of seeing model disagreement laid out clearly in one place rather than scattered across separate conversations.
The Privacy and Regulatory Questions Hiding Underneath the Novelty
When a Hallucination Becomes a Compliance Problem
For developers, privacy professionals, and policymakers, In the Weights is more than a novelty. It is an accessible, user-facing demonstration of a much larger reckoning already underway around AI transparency and personal data. The European Data Protection Board has acknowledged that applying traditional data subject rights frameworks, the kind that let individuals request correction or deletion of inaccurate personal data, to AI model weights is technically and legally complex. Regulators are actively working to address that gap through the EU AI Act and supplementary guidance, but the underlying problem remains genuinely hard: how do you correct a fact baked into billions of parameters rather than stored in a single database row?
When a model hallucinates details about a real, identifiable person, as GPT-5.4 Mini reportedly did with Anthony Ha, that is not merely an amusing quirk. It potentially touches on data accuracy obligations under frameworks like the GDPR, since inaccurate personal data is being processed and reproduced at scale, just inside a model's parameters rather than a conventional database.
A Practical Warning for Any Team Deploying LLMs
Beyond the regulatory dimension, In the Weights raises a substantive technical point that should be on the radar of any team deploying large language models in production. The platform's ability to compare outputs across model versions, including different generations of the same underlying model family, reveals that nominally similar models can return dramatically different information about the same person. That inconsistency is not just an academic curiosity. It is a real operational risk for any enterprise application that leans on an LLM to retrieve or summarize information about real individuals, whether that is a customer support tool, a recruiting assistant, or an internal knowledge system.
The International Association of Privacy Professionals has noted that AI-generated hallucinations involving real individuals are an emerging compliance risk that most organizations have yet to adequately address in their AI governance frameworks.
How This Fits Into the Broader Evolution of the Vanity Search
| Era | What You Searched | What the Result Reflected |
|---|---|---|
| Classic web search | Your name on Google or Bing | Indexed web pages, social profiles, news mentions |
| Social-era vanity metrics | Followers, likes, engagement | Audience size and platform-native visibility |
| AI-centric vanity search (In the Weights) | Your name across multiple LLMs | Presence and consistency inside model training data itself |
The vanity search has always evolved alongside whatever technology people used to learn about each other. Web search rewarded being indexed and linked. Social media rewarded being followed and engaged with. In the Weights suggests the next stage rewards being recalled, accurately and consistently, by the systems that increasingly mediate how information gets summarized and surfaced to other people in the first place.
What Comes Next for AI-Centric Identity Tools
The reception to In the Weights has been brisk, something the founders attribute to a mix of nostalgia for old-school vanity metrics and genuine curiosity about who, exactly, lives on inside the weights of the AI systems shaping daily life. Whether the project remains a fun one-off or becomes the first of a broader category of AI-recall auditing tools is an open question. Several commentators following the launch have suggested that querying a model for what it knows about an entity, and comparing that against ground truth, could eventually become something closer to a standard step in the AI development lifecycle, particularly as litigation and regulation around AI training data and web scraping continue to intensify.
As AI companies continue layering in more rigorous safety filters and alignment training, the scores produced by tools like In the Weights are likely to keep shifting. That instability is itself informative: it shows that a person's apparent presence inside a model is not a fixed fact about them, but a moving snapshot of how that particular generation of models was trained, filtered, and aligned at one moment in time.
The Bottom Line
In the Weights is, on its surface, a fun and slightly self-indulgent way to spend ten minutes finding out whether ChatGPT and Claude have ever heard of you. Underneath that surface, it is doing something more substantive: turning an invisible property of modern AI systems, the latent, compressed memory of real people baked into billions of parameters, into something concrete, numeric, and shareable.
That concreteness is valuable, even if the score itself should be treated with real caution. A strength score conflates presence in a training corpus with social importance, and it is vulnerable to exactly the kind of hallucinations and biases that already make people skeptical of AI-generated summaries. But the broader question the tool puts in front of people, who gets remembered by the systems that increasingly decide what the rest of the world hears about you, is not going away. If anything, projects like In the Weights are an early, somewhat playful signal of a much larger reckoning the AI industry will eventually have to face head-on.
Related Topics: #InTheWeights #VanitySearch #AITransparency #DigitalIdentity #AIHallucinations #LLM #AIPrivacy #TechCrunch #AIModels #PersonalBranding #AIGovernance