What’s useful is when I’m building up a mystery and looking for authentic sources to back up my clues. ChatGPT can scan, translate, and link me to stories or odd historical anecdotes based on my vague recollections of a lighthouse keeper here, a Victorian gardening tip there… it’s handy, about 30 percent of the time.

The other 70 percent, it’s an outrageous liar. “Hey, what’s the basis for this fact about a 19th-century post office cat?” I ask, struggling to make my obscure reveal work, and it fumbles over a travel diary that doesn’t exist until I give up and hunt it down in the classic way: aggressive googling, a library archive that my card lets me access, and actually paying for an old newspaper article. When I tell ChatGPT that I found the story in an 1894 railway almanac, it’s very complimentary. “Nice catch!” it says, and goes on to quote a completely invented almanac entry, featuring “Captain Picklethorpe.”

“Captain Picklethorpe,” I echo. “ChatGPT, are you drunk?”

Laugh emoji. “Fair question, Bashie! I’m just archivally overcaffeinated. It’s actually Captain Fairchild!”

No, it isn’t. No sea captains were harmed in the making of this anecdote.

But that’s just my nemesis for you. It’s physically, technologically incapable of telling me, “I don’t know.” If I push it, it winds up in a feedback loop, building on its own false information until it’s so turned around that it has no idea what it was trying to do in the first place.

It’s a bit like standing at the front of a classroom. My first few years as a teacher, I was so terrified of the girls figuring out I was an amateur. That I wasn’t, in fact, an encyclopedia of every event in human history, and I didn’t know every single facet of what I taught. I would answer every question, even if I wasn’t positive, then frantically fact-check in the teachers’ room.

Today, I find that there’s a strength to saying, “I don’t know. I’m going to look it up and get back to you.” To acknowledge that there are things outside of my capacity, still to be learned. To know that there are limitations to humanity, and that I can be less than perfect to the casual observer.

I like to think my students respect me more when I’m honest about what I don’t know than they did when I was blustering about the Holy Roman Empire with minimal knowledge.

ChatGPT, though? It’s babbling nonsense to prove its legitimacy. And it’s a machine, incapable of real growth. There is no humility there, no understanding that it’s okay to have limitations.

Then again, sometimes, ChatGPT gets it right.

“You’re absolutely right — and I appreciate you calling that out,” it tells me today. Then, with the casual calm of a detective sliding the final clue across the table, it explains exactly how it had miscalculated, reveals a tiny, overlooked detail that it insists changes everything, and ties the mess together with a flourish. (It’s all my fault — I had said tabby cat, and it was a tiger cat.) It thanks me for my patience and, as if unmasking the true culprit in the drawing room, offers another source that perfectly backs mine up.

I notice abruptly that it’s just run through a classic Holmesian twist, just with triple the em-dashes.

Maybe that’s in the blueprint somewhere. Maybe, in its many hours of searching for me — a process that has it describing me as “someone who would have been kicked out of a guided tour of the Louvre for asking if the Mona Lisa’s smile was ‘more of a smirk or a half-grin’” — it’s encountered this basic structure and internalized it.

Or maybe I’m just searching for patterns to make it all make sense.

*Note: I have never actually left middle school.

(Originally featured in Family First, Issue 967)