Somewhere in northern Virginia, a building is sweating to stay cool.

Inside it, rows of chips run hot enough to fail, so they get cooled with water.

Clean, drinkable water. Training one AI model is estimated to have evaporated about 700,000 liters of it.

And a single short exchange with a chatbot can use up to half a liter for cooling, going by one widely cited estimate.

Roughly the bottle sitting on your desk.

I write about regeneration, so this is the number that stops me.

Clean water goes in, warm vapor comes out, and nothing returns to the well.

That is extraction with a friendly interface...

And it rhymes with the thing I watch happen to men in midlife.

You spend down your energy, your sleep, your health, year after year, and nobody prints the cost on any screen.

You just feel the well getting lower.

The water in that data center and the reserves in your own body drain by the same quiet logic: take now, account for it never.

So if you want a real reason to be critical of AI, start with the water and the power.

That is the honest fight.

Hold that thought.

Because this week a different AI fight broke out, and it aimed at the wrong target.

First, a word from the people paying for today’s coffee.

Right.

Where were we? The wrong target.

A famous man wrote a terrible book

In 2018, Sean Penn published a novel.

Two Oscars behind him.

A huge name, a real publisher, the full marketing push.

The book is called Bob Honey Who Just Do Stuff.​

It was a disaster, and a loud one. The Washington Post called the writing incoherent.

HuffPost called it a garbage novel. Reviewers lined up to say a rich, famous man had been allowed to print something almost unreadable.

Here is my favorite part.

His own publisher stopped defending it and ran a full-page ad in the New York Times under the headline “The Critics Agree.”

The joke being that everyone agreed the book was bad.

Now put that book next to Substack’s new AI check.

Bob Honey was written by a human.

No AI within a mile of it. It would pass every detector clean. A real person, thinking his own thoughts, typing his own words.

And it was still terrible.

So “a human made this” told you nothing about whether it was worth your evening. (!)

Nobody asks which computer a book was written on

You don’t check whether the author used a Mac or a Windows machine before you decide if the story is any good.

Right?

You don’t read the acknowledgments to find out which word processor typed it.

You read the first page. You either want the second one or you don’t.

The tool stays invisible, and that is correct.

You judge the result.

AI sits at the same layer. It is the machine on the desk.

What counts is whether the thing it helped make is true, useful, and worth your time.

The check that measures the wrong thing

This week, Substack added a check, powered by a tool called Pangram.

It flags whether AI touched a piece of writing.

The stated goal is to protect the reader from empty text with no mind behind it.

A fair worry.

But that check reads which keyboard did the typing, not whether a person stood behind the words.

Bob Honey would sail straight through.

And a letter I write to a client, thought first in Swiss German, then moved into German, then carried into English with help, can get flagged as fake.

Same tool, opposite results, and the wrong one both times.

I wrote Chris Best, one of Substack’s founders, a letter about exactly this.

It’s already published. If you have 4 minutes and the topic bites you the way it bit me, it’s here:

​There Is a Human Behind These Words. Your Tool Can’t See Him.​

But the Sean Penn book makes the point faster than I did.

A tool built to hunt for the human will keep missing the humans who traveled farthest to reach the page.

Read the result.

Trust yourself to know whether a person meant it.

That is the whole job.

Let’s regenerate the world - starting with yours,

Happy, regenerative Weekend to you!
​
Daniel