The Thrilling Conclusion!

“Any sufficiently advanced technology is indistinguishable from magic.” — Arthur C. Clarke

I began this series on generative AI because an irate reader had suggested that my blogs were written by a particularly doltish AI bot. Now that we know a lot more about generative AI we can decide whether my irked reader could even possibly have been right.

Bloggers, like columnists for newspapers and magazines, work constantly under deadline. Typically, we have to turn our essays in to an editor on a certain day and time – or else. The biggest problem we have by far is coming up with a topic to write about week after week.

Because it’s so hard to come up with interesting ideas, even the best writers have a pretty low hit ratio. But the same is true of baseball. Getting a hit only one out of every three at-bats sounds pretty dismal, but if you can do it for ten or fifteen years you’ll be in the Hall of Fame.

And the same with writers. Those who have a decent ratio of good essays to ho-hum essays rocket to the top of the profession. Those people eventually get to the point where they’re no longer working against an iron deadline but only publish when they have something interesting to say. (David Brooks, Maureen Dowd.)

Alas, those of us who toil near the bottom of the literary food chain have no such luck. We have to come up with an interesting topic every week and get our 1,000 words out on schedule. Naturally, our hit rate is low – so low some people think we must be bots.

As it happens, this essay is the 625th I’ve penned since I launched the blog back in 2013. Imagine how much easier all this would be if I could just ask ChatGPT to write my essays for me! But is it possible? Let’s find out.

The answer to all my problems would be an AI bot that could come up with an interesting topic every week and write the essay. So I opened ChatGPT and asked the bot to “Write a 1,000-word essay that is accurate and amusing.”

At first I thought I’d struck gold. The bot, out of the blue, wrote an essay on “the World’s Oldest Joke.” In case you’re interested, it was a joke told in ancient Sumer some 7,000 years ago and it went like this: “Person A: Tell me something that has never occurred in the history of the world. Person B: A young woman did not fart in her husband’s lap.”

Well, my wife thought it was funny….

But then, when I asked the bot for another 1,000-word essay that was accurate and amusing it came up with the same topic. I explained to the stupid bot that I can’t publish the same essay every week, that’s not how it works. But the world’s-oldest-joke essay was all the bot could offer.

That meant, of course, that if I was going to partner with a bot to write my blog it would be me that would have to come up with the topics – and that was by far the hardest part. But at least then the bot might write the essay for me. Or could it?

A few years ago I wrote a series of essays about a lady in Scotland, so I asked ChatGPT to “Write a 1,000-word essay about The Lady Jean Fforde, who owned the Isle of Arran and was broke. The essay should be accurate but amusing.” The result wasn’t pretty.

As other ChatGPT users have noted, the bot’s essay sounded authoritative, and if you knew nothing about Lady Jean Fforde you might have thought it was accurate. Here, for example, is the first sentence: “Lady Fforde was notorious for her financial struggles and her eccentricity.”

In the first place, her name was “Lady Jean,” not “Lady Fforde.” (Colonel Fforde, the ex, was a commoner.)

Also, while Lady Jean certainly endured financial struggles and was a harmless eccentric, she wasn’t “notorious” – she was “well-known” for her financial struggles and was a beloved character.

Then the bot descended into cliché: “Her life was full of ups and downs, and her story is a testament to the fact that even the wealthy can fall on hard times. … Her financial struggles … serve as a reminder that money can’t buy happiness.” Erg.

After that, the bot went off the reservation, inventing things that never happened (these errors are known in the AI community as “hallucinations”): “Lady Fforde (sic) was known for her frugality … Her thriftiness would eventually come back to haunt her, as she struggled to keep up with the expenses of running the island.” This is precisely backwards – it was Lady Jean’s lavish spending and inability to stick to a budget that led to her problems, not her “frugality.”

Lady Jean “had a bit of a gambling habit.” Wrong – that was Colonel Fforde.

“She was forced to sell the Isle of Arran to the Scottish government to pay off her debts. Lady Fforde was devastated to lose her beloved island, and she spent the rest of her life living in a small cottage on the mainland.” Wrong, wrong, wrong.

Lady Jean didn’t sell the island to the government, she was forced to turn her castle (Brodick Castle) over to the Scottish National Trust in lieu of estate duties. She then moved into a home (not a cottage), called Strabane, which was just down the road from the castle (not on the mainland). And so on and so on.

So the bot couldn’t come up with interesting subjects to write about and it couldn’t write a decent and accurate essay. Like a student who has to write a 1,000-word essay but has only 500 words of knowledge, the bot pads its responses with impressive-sounding but wrong facts. In short, my indignant reader was unfortunately wrong – despite my fond hopes, these essays weren’t, and at least for now can’t, be written by a bot, dim-witted or otherwise.

Of course, I was using ChatGPT-3, while GPT-4 was released in mid-March (although it was only available to users who were willing to pay for it, i.e., not me.) Maybe I’ll be replaced by version 4, or maybe 5, but for now I’m stuck writing these essays exactly like people wrote essays in ancient Sumer, 7,000 years ago.

Next up: Can the Fed be Saved?