AI Knows Who I Am. It’s Just Ten Years Behind.

I typed my name into a website called In the Weights and it handed me back something that looked like a personal trading card.

Pixel art, retro arcade font, the whole thing styled like a 1989 video game. Pew pew. My name was displayed across the top with a strength score of 773. That makes me the top 3 percent of every name the tool has ranked. And under that, the title the AI models had settled on for me: Internet Marketing Entrepreneur. One of them, Mistral, summed me up as someone known for his work in digital marketing and social media.

All of them were accurate. Sort of. At least they WERE accurate. That positioning for me is almost a decade old.

In the Weights was built by two former OpenAI people and went live this month. The premise is simple and a little unnerving. It asks a dozen or so AI models the same question, who is this person, and measures how confidently each one answers from memory alone, with no web search allowed. Then it scores you. The ceiling is 999, and the builders reserve that top tier for names like Trump,Mozart, Shakespeare, and Taylor Swift. Elon Musk came back around 992. Tim Cook landed in the high 980s. Several of the journalists who reviewed the tool scored in the low hundreds.

So 773 isn’t nothing. Plenty of models know my name and answer with confidence.

The interesting this is what they’re confident about.

The version of me sitting inside those models is the one who built a fart app.

I’m not embarrassed by that. iFart hit number one in the App Store and stayed there for 23 days, and I’d do it again tomorrow. The AdSense Code made the New York Times list in 2006. Twitter Power came out in 2009. That was the work that put enough of my name across the early web for an AI to encode it years later. The models aren’t wrong. They’re just behind.

The score measures one specific thing, and it’s worth knowing which thing.

When people talk about a model’s “weights,” they mean the billions of numbers locked in place when the model finished training. That’s the model’s long-term memory, fixed the day training stopped. In the Weights only reads from that frozen memory. It won’t let the models look anything up.

That’s a different thing from what happens when you ask ChatGPT or Claude about someone today and it searches the web before answering. Live search sees this year. The weights see whatever the world looked like when the training data was collected, which is often at least a year or two behind, sometimes more.

My current work, the keynotes, the AI translation for people who don’t write code, the book coming out with Morgan James this December, almost none of it has reached the weights yet. It’s too recent. What’s encoded is the dense, decade-deep record of the guy from the social media era. The new chapter is all over the live web. It just hasn’t been baked into anyone’s long-term memory.

I keep coming back to how different this is from the Google era.

For 20 years, your reputation was a list you could inspect. You searched your name, you read the first page, and if something was wrong you published a correction or asked someone to fix it. You could see the record and edit it. AI doesn’t work that way. The answer is generated, not listed, and it’s pulled from a frozen impression of you that you can’t open up and rewrite. You can’t email the weights. “Hey, Mr. Weights, this is what I’m doing now.”

Which means anyone who has reinvented is going to run into the same thing I did. Changed fields. Left a company. Started a new one. Picked up a craft at fifty. The machine introduces you as the person you used to be, and the introduction is confident and out of date, and it’s increasingly the first thing a stranger hears about you.

I’m not worried about it, and that surprised me a little.

The frozen memory always catches up. Every model that trains next year will read a web where the AI work outweighs the old material, and the encoded version of me will move forward a chapter. The lag is real, but it’s only a lag. The correction is already running. It’s just running on the training cycle’s clock instead of mine.

I’ve been on the wrong side of a curve that hadn’t finished bending before. It’s the same lesson every time. You are not behind. You are early. Again.

The next version of me the machine learns, I’ll have already moved past.