Week of March 12, 2026

From the editor’s desk

The nerds of my household have been playing around with AI lately, using various platforms, with varying results.

Jerry, my spouse, asked Meshy, an AI platform used to make models for 3-D printing, to come up with a rendering. He asked, “What do the sheep look like that androids dream of?” (It is a reference to the Philip K. Dick sci-fi novel, Do Androids Dream of Electric Sheep?)

Meshy quickly rendered a puffy, light gray sheep with big eyes and a cute, creepy sci-fi vibe. While the image looked great onscreen, the file was problematic, and the 3-D print came out as an unrecognizable tangle of filament.

Nico, my 22-year-old son, has been shooting photos of white boards and papers full of equations and asking Gemini to function as an OCR scanner to digitize them, as part of his research in physics. His assessment on how well this works: “Pretty good. I’d say a 90% chance of reading an entire sheet correctly, and an 80% chance of writing the equations correctly.”

I have asked ChatGPT a few times to help me look for specific shreds of information that I may need in order to verify a fact. Sometimes, it gets me to a reliable source within a couple of clicks. Other times, it hallucinates fanciful assertions. It told me that an independent, student newspaper from the early 1970s was published by the Reserve Officers’ Training Corps (ROTC), which it was not. It told me that a particular Midtown restaurant was owned by the late University of Nevada, Reno, journalism professor Todd Felts, which it was not. 

When I factor the time I’ve saved searching for accurate verifications against the time I’ve wasted correcting for ChatGPT’s nonsense, I believe I have probably broken even. 

Our collective household takeaway: AI can be useful and potentially amusing, but its results are not inherently trustworthy. Obviously, you cannot take its word for things willy-nilly.

But … what if a massively powerful, pseudo-government organization does decide to take AI’s word for things willy-nilly?

Well, there is no need to speculate. That actually happened. Remember the so-called Department of Government Efficiency’s eight-month wrecking ball spree in 2025, when it slashed funding for agencies like the Department of Education, the Federal Aviation Administration, FEMA, the U.S. Agency for International Development and the National Institutes of Health? (I’ll stop here, but the list goes on for quite a while.)

This past Saturday, The New York Times reported that when DOGE was deciding which already-approved humanities projects to defund in 2025 for being too “woke,” its employees sat back and put AI in charge. This casts the wrecking ball spree in an even more heinous light than we knew.

I’ll quote the Times at some length here. This is all important: 

The prompt was simple: “Does the following relate at all to D.E.I.? Respond factually in less than 120 characters. Begin with ‘Yes’ or ‘No.’” The results were sweeping, and sometimes bizarre.

Building improvements at an Indigenous languages archive in Alaska risked “promoting inclusion and diverse perspectives.” Renewal of a longstanding grant to digitize Black newspapers and add them to a historical database was “D.E.I.” So was work on a 40-volume scholarly series on the history of American music.

A documentary about Jewish women’s slave labor during the Holocaust? The focus on gender risked “contributing to D.E.I. by amplifying marginalized voices.”

Even an effort to catalog and digitize the papers of Thomas Gage, a British general in the American Revolution, was guilty of “promoting inclusivity and diversity in historical research.”

So far, so bad. DOGE was aiming for exactly the type of erasure and propaganda many of these humanities projects were trying to counter. But hold on … it gets worse:

The DOGE employees did not appear to question ChatGPT’s judgments, and continued hunting for unacceptable projects. Two weeks later, they sent a master list of 1,477 problematic awards—nearly every active grant made during the Biden administration—to Michael McDonald, the endowment’s acting chairman.

You had me at “the DOGE employees did not appear to question ChatGPT’s judgments.” I’m somewhat terrified to think of what we’ll learn next, both about DOGE and about whatever careless, destructive uses of AI are coming down the pipeline. 

In journalism, we’ve always said, “Trust but verify.” With such a powerful tool available to all—the scrupulous and unscrupulous among us—I’m upgrading my advice to “Verify, verify some more, and, by all means, don’t trust AI to be your moral compass like DOGE did.”

Take care,

—Kris Vagner, managing editor