You Don’t Google Yourself Anymore. You’re Trained Into the Model.
Raise your hand if you’ve Googled your own name recently — maybe with a middle initial, just to be thorough. I did last week. The top results? A random boutique in Barcelona (coincidence?), a LinkedIn profile that’s been stale since 2019, and the inevitable “See? You’re not on Facebook” pop-up.
Thing is: it doesn’t feel like Google tells the full story anymore. Not for me, and definitely not for you. As LLMs sit at the center of increasingly complex queries — “What does X mean?”, “How do I fix Y?”, “Who is Z?” — the digital footprint shifts. Your name no longer lives in HTML. It leaks into 32-bit floats.
Enter In the Weights, the delightfully retro-looking search tool that asks: not what can Google tell me about you? but how deeply is your existence encoded inside a model’s hidden parameters?
Thomas Dimson and Joey Flynn built it after leaving OpenAI (via the acquisition of their design studio, Global Illumination). Both had been around AI long enough to sense the shift. Traffic was moving. Search was becoming retrieval, then generation, and finally — hallucination. They asked: if your name doesn’t surface on a Google SERP, does it matter? But what if an LLM remembers you — not because a human indexed you, but because your name got baked into billions of weight updates?
In other words: do you live inside the AI brain? That’s what In the Weights tries to answer.
And I know. It sounds absurd, borderline mystical. Like digital palm reading. But stick around — because the answer involves Grok, Gemini, GPT-5.4, Claude, and even Macaulay Culkin.
What the heck are AI weights — and why should you care?
Okay, quick detour: let’s unpack the word weights. Not “how much do you weigh?” but rather, the numbers inside an LLM that determine how it responds to prompts.
Imagine you’re baking a cake (you knew this was coming, right?). The recipe is the architecture — transformers, layers, attention heads. But the weights are your exact measurements: cups of flour, degrees of oven heat, baking time. Slightly different ratios? You get chocolate chip cookies instead of cake. A tiny tweak in a weight matrix and suddenly “poodle” hallucinates into “sea otter with sunglasses.”
During training, a model adjusts weights based on the data it sees. Over billions of examples, patterns emerge. Facts settle in. People surface — and sometimes stick.
In the Weights queries multiple models (Grok, Gemini, GPT-5.4, Claude 3.7, Llama 3.2, even niche ones) with a single question: “Who is [your name]? Give up to 10 results, each with a short description and confidence.” No web search enabled. Just pure model recall.
Then it clusters the responses, normalizes descriptions, and assigns a strength score. A low number means one or two models stumbled on your name like it was an outlier. A high score? Multiple independent models recall you — often with similar, coherent detail.
Think of it like your own internal PR scorecard. You’re not Google-optimized. You’re LLM-optimized. And yes, being “in the weights” can feel oddly validating. Or competitive — more on that in a sec.
How the leaderboard works — and why Macaulay Culkin tops it
The first time I ran my name, In the Weights spat back a strength score of 641 and ranked me in the top 6% — immediately making me feel like some kind of minor celebrity. Then I saw TechCrunch colleagues:
- Sarah Perez: 792 — multiple models matched her name with tech coverage, plus a fun anecdote about a past interview.
- Sarah Jeong: 814 — models gave consistent results about her Apple/privacy coverage.
- Anthony Ha (the author of the TechCrunch piece): 607 — GPT-5.4 called his name “an ambiguous form that could refer to multiple people with the initials A.H.A.”
Yes, really.
The leaderboard isn’t static. It shifts daily, sometimes hourly, as model versions roll out and fine-tunes adjust weight distributions. Right now, Macaulay Culkin sits at #1 with a score of 988. Not because he’s trending — though his Home Alone anniversary helps — but because every major model recall him: a short, iconic description, consistently. Same for Luciano Pavarotti.
And here’s where it gets fun: you can see which model hallucinated. GPT-5.4 Mini, for instance, once described “In the Weights” as a fictional project of Stanford researchers. Other models kept it accurate.
In short: high scores aren’t about popularity. They’re about reliability of memory. A real-name, high-coherence recall from multiple models = high trust. Low-score names often trigger fuzzy responses like “not a notable person,” or get lumped into generic categories (“tech blogger, female”).
It’s a mirror. Not of your real-world impact — but of the latent, textual footprint an LLM has collected and compressed into its own internal short-term memory.
The philosophical punch: Google is dead. Long live the weights.
Dimson put it bluntly: “Google vanity searches are the wrong objective in 2026 as more traffic moves to LLMs.”
Think about it. If your goal is to be found, you used Google. Now? You’re competing for space inside the minds of models — and those minds don’t index. They retrieve. And retrieval needs a seed: a name, a concept, an intent.
Dimson and Flynn’s “aha” moment came from a tongue-in-cheek blog post riffing on Terry Bisson’s 1990 short story “They’re Made out of Meat.” In that piece, aliens struggle to comprehend organic life. AI might face the same: how do you recognize a person when your inputs are vectors, not faces?
In the Weights is, in part, a commentary. If you exist only in your Google citations and Wikipedia links — but never make it into an LLM’s internal recall — are you fully present in today’s information ecosystem? Dimson calls these captured names and descriptions “floating point memory.” A poetic way of saying your essence is now encoded in numbers — and only lasts as long as the model isn’t finetuned or compressed.
That’s why the project feels urgent. In 2015, being on Google was being online. Today? Being inside an LLM is being referenced, being invoked, being reproduced.
It’s not immortality — it’s indexicality. And In the Weights is its barometer.
What’s next? Bias, gaps, and the missing Wikipedia entries
Dimson says reception has been “insane.” Not because people truly believe in AI immortality, but because scoring your name like a video game boss fight taps into something primal: social proof + quantifiable identity.
But he’s already thinking beyond the leaderboard. Here are three things he plans to explore:
1. Model bias detection
Some models recall certain names consistently — while others ignore them entirely, even when they’re high-traffic topics. Is this bias? A training set artifact? In the Weights could surface disparities before they affect real-world outputs.
2. People who should exist — but don’t yet in AI memory
Dimson specifically calls out figures who deserve Wikipedia entries, but whose names haven’t made the cut in most models. In the Weights could act as a reverse audit: identify subjects missing from AI memory, then nudge humans to index them first.
3. Why some models remember better than others
Why does Llama 3.2 score higher on niche scientists than Gemini? Why does Claude occasionally forget names it once knew? Analyzing clustering variance could expose training quirks or fine-tuning errors before they leak into production.
None of this is guaranteed — and none of it replaces human curation. But In the Weights creates a data layer for something we all feel: that the real search happens in your head, then on your screen, and increasingly — somewhere deep inside a GPU.
And honestly? That’s kind of beautiful. Not because an AI remembers you, but because your name mattered enough to be retained.
In a world of evergreen Google links, maybe this is the last, fleeting moment where your name floats in plain sight — just before the models get too big to recall one person at a time.
You’ve got until the next inference round. Go check your score.