In the Weights: New Tool Scores How AI Knows You

In the Weights lets you do something that did not exist a year ago: search for yourself inside an AI model memory. The tool scores how well a large language model can recall a real person from its training weights alone — no browsing, no tools, just what the model absorbed during training.

It blew up because it names a feeling everyone already has. Googling yourself stopped meaning much. These days, just as many people size you up by asking a chatbot.

What In the Weights actually measures

Built by Thomas Dimson and Joey Flynn, the site treats a model parameters — the weights — as the new index of public reputation. Instead of asking what the web says about you, it asks what the model has internalized about you and how confidently it can repeat it.

The result is a score. A high number means the model can produce rich, specific, self-assured answers about a person straight from memory. A low number means the model mostly draws a blank.

Why a vanity score suddenly matters

Search used to be the canonical record of who you are. That is fading. When someone meets your name now, they may never open a results page — they ask an assistant, and the assistant answers from memory.

That moves reputation from a public, fixable index into something baked into a model. You cannot edit a weight the way you can update a webpage.

Reality check: a high score is not the same as accuracy. A model can recall you confidently and still be wrong — and that gap is exactly the risk.

How it works under the hood

The mechanism is closed-book recall. The tool prompts the model for information about a person while denying it any retrieval tools, then grades how complete and confident the response is.

Because there is no live lookup, the answer reflects only what survived training: how often a person appeared in the data, how distinctly, and before the model knowledge cutoff. Niche or recently-famous names tend to score low simply because the model never saw enough of them.

Closed-book recall versus open-book search

What happens next over the coming days

Expect the scores to spread as status symbols. Founders, creators and executives will post their numbers the way they once shared follower counts.

Expect pushback too. As people compare scores against reality, the accuracy gap becomes the story — flattering hallucinations for some, unfair blanks for others.

And expect copycats. The format is simple enough that rival tools, multi-model comparisons, and brand-level versions are the obvious next step.

What you can do about your score

There is no settings page for a model memory, so the levers are indirect. Consistent, factual, widely-published information about you over time is what models absorb. One viral post will not move a weight; a durable public record might.

For now, treat your number as a curiosity, not a verdict. The more useful question is not how much the AI knows about you, but whether what it confidently says is even true.