by Douglas Yeo (November 3, 2026)
Readers of TheLastTrombone know that while I believe artificial intelligence (AI) has many positive benefits, if one is not exceedingly careful, it is highly dangerous and misleading in very consequential ways. As a researcher and educator, I am very concerned about the role AI is increasingly playing in schools, as students rely on it to do research and write papers. My previous articles about this—ChatGPT and the trombone: Is this the end of writing and research? I don’t think so—and Why trombone players cannot trust artificial intelligence (AI) for historical information—outline some of my concerns.
Now I have another example to share with readers. If you are not suspicious of AI and you have a tendency to trust what ChatGPT or Claude tell you, please read on. Because what you are about to read is a story about a stunning failure of AI that is representative of one of the most dangerous aspects of this new technology: AI makes up stuff and represents it as truth.
My granddaughter is in sixth grade, her first year of middle school. Her history class is studying English history, and in particular, the Battle of Hastings. Fought on October 14, 1066, between the English army of King Harold II and the invading Norman army of William, Duke of Normandy, the Battle ended in Harold’s defeat and death, securing William’s claim to the English throne and paving the way for the Norman conquest of England. And the rest, truly, is history.
Last night, our daughter’s family came over to our house for dinner and my granddaughter and I had a really engaging conversation about the Bayeux tapestry (she has been learning about the tapestry in her class), which gives a spectacular visual account of the Battle of Hastings.
I don’t intend to give a detailed explanation of what happened on the battlefield in 1066—that’s not the main point of this article—but you can get a good summary HERE on the English Heritage website. This discussion of the Battle of Hastings employs many images from the Bayeux tapestry, an extraordinary 230 feet long embroidered cloth (technically it’s not a tapestry) thought to be made shortly after the Battle of Hastings. So, it is about 1000 years old. It is ordinarily on display in the Bayeux Museum in France but for a limited time from September 2026 to July 2027, it is on display in the British Museum in London.
David Wilson, The Bayeux Tapestry: The Complete Tapestry in Colour (London: Thames and Hudson, 2024).
Edited by Sue Brunning, Lloyd de Beer, Michael Lewis, and Naomi Speakman, The Bayeux Tapestry (New Haven: Yale University Press, 2026).
When my granddaughter and I were talking about the Bayeux tapestry last night, I paused our conversation and went to a shelf on my library and pulled down two of my books about the tapestry and the Battle of Hastings: David Wilson’s The Bayeux Tapestry: The Complete Tapestry in Colour (London: Thames and Hudson, 2024), and a new book edited by Sue Brunning, Lloyd de Beer, Michael Lewis, and Naomi Speakman, The Bayeux Tapestry (New Haven: Yale University Press, 2026). Both books have terrific production values and informative commentary.
Men looking at Halley’s comet, Bayeux tapestry panels 32 and 33
The tapestry is a fascinating historical document and full of famous segments. Among these is a group of men looking up at Halley’s comet (panels 32 and 33 in the tapestry, above) which streaked across the night sky for two weeks in April and May 1066 (over Harold’s head in the tapestry) and was viewed in the medieval world as a bad omen. The caption over this scene (in Latin) reads, “ISTI MIRANT STELLA”—”These men are marveling at the star.”
The death of Harold, Bayeux tapestry panel 57
Perhaps the most dramatic, important, and controversial panel on the tapestry depicts the death of Harold (panel 57 in the tapestry, above). The caption above reads, “HIC HAROLD REX INTERFECTUS EST”—”Here King Harold has been killed.”
Exactly how Harold died is a subject of ongoing controversy. Various sources give different ways in which Harold met his end but the tapestry appears to indicate that he died as the result of an arrow in his eye, and he was subsequently hacked by a Norman soldier’s sword. Harold appears twice in panel 57. (Keep in mind that the tapestry was meant to be viewed from left to right, like a long comic strip, and some characters appear more than once in a scene as the narrative unfolds.) First, the arrow hits Harold, although scholars disagree on whether the arrow was original to the tapestry or was added in the nineteenth century:
Harold is struck with an arrow, Bayeux tapestry panel 57 (detail)
Then, Harold is shown under a Norman soldier’s horse while being hacked with a sword. While you do this, note the small dots to the left of Harold’s head (in the red circle). Some scholars believe that these originally held the threads of an arrow in Harold’s eye that were later removed; others disagree with that conclusion):
Harold is struck with a sword, Bayeux tapestry, panel 57 (detail)
By now, you might be thinking, “Well, this is interesting, but what in the world does it have to do with artificial intelligence?” Actually, everything.
Because the Bayeux tapestry is, today, a part of many conversations around the world, I thought I would put artificial intelligence to the test. By now, I’m sure that AI has been fed an uncountable number of books, articles, and commentaries about the tapestry. I’m equally certain that hundreds—or thousands?—of people ask AI questions about the Bayeux tapestry every day. So, I decided to ask AI a simple question and see what it came up with. Here’s my question (below are actual screen shots from my online interaction with ChatGPT):
It’s a simple question and I wondered what AI would do with it. I even gave AI some help, and let it know that it might find Harold depicted more than once. I expected it would direct me to an online image of the Bayeux tapestry; provide me with a link; perhaps lead me to some commentary about the scene. ChatGPT worked for 1 minute and 10 seconds and delivered this image (with no commentary); the image was had the file name Bayeux Tapestry_ Two Harolds Identified.png:
The death of Harold on the Bayeux tapestry according to ChatGPT
Uh oh. Compare this to the actual tapestry:
The death of Harold, Bayeux tapestry, panel 57
Even a sixth grader can see that the two images don’t line up. They are completely different. Yes, both show a character that ChatGPT identified as Harold getting an arrow in his eye and then getting hacked with a sword. But everything is wrong. AI invented new characters; it changed the images of the tapestry’s bottom border; it even changed the direction Harold was facing when the arrow flew into his head. AI made up a completely new scene in the style of the Bayeux tapestry and represented it as the real thing. AI delivered a completely fictional image, an EPIC FAIL. It—to use the word that is popularly used to describe incorrect information that AI delivers as truth—hallucinated. Big time.
I wasn’t going to let ChatGPT off the hook. Of course, when we interact with AI, we’re not interacting with a person; we’re interacting with a machine. A highly developed machine. But it is not human; it does not have emotions and feelings. It does not care about you. Still, I decided to engage with AI about its failure. Game on, machine! I wrote the following statement, full of hyperbole because I wanted the machine’s algorithms to understand that I was annoyed at what it did:
This is a danger with AI that many people do not understand. AI makes up stuff. ALL. THE. TIME.
ChatGPT then replied to my scolding, and wrote this (part 1 of its response):
Part 1 of ChatGPT’s response to my criticizing it for hallucinating
You’ll see several things here. First, the machine is not apologizing. It’s a machine. It can’t apologize. It did, however, recognize that it is wrong. Its “apology” was remarkably honest in describing its failures. And these failures happen with AI ALL. THE. TIME. If you’re a lazy student and you ask AI a question, you are virtually guaranteed to have a visit to the Dean’s office if you don’t test EVERYTHING AI gives you. Because AI makes mistakes and propagates disinformation in the guise of truth ALL. THE. TIME.
ChatGPT then continued with its conclusion:
Part 2 of ChatGPT’s response to my criticizing it for hallucinating
Here, ChatGPT recalled that I have criticized it in the past. It recalled that I have repeatedly told it that when I learn historical information, I always work to triangulate it and find multiple sources that support a claim. AI does not do this. It makes pronouncements which are often unsupported. All researchers need to be honest enough to differentiate between evidence, inference, and speculation. (Actually, I never use the word “speculation” when I discuss these three types of conclusions; I always use the word “possibility.” But, hey, it’s AI. It’s no surprise it got this wrong.)
ChatGPT acknowledged that it created a counterfeit image, and it concluded its mea culpa by recognizing that what it did in creating a fictional image and representing it as an image of the Bayeux tapestry is, “precisely the sort of failure that makes AI-generated material dangerous in historical research.” Preach it, machine. I think it must be reading TheLastTrombone. (Actually, it probably is. AI data centers read vast quantities of what’s on the internet..)
Friends, I’m not making this up. This really happened. And it really happens. Please, please, please: NEVER accept anything that AI delivers to you as factual on its face. ALWAYS verify anything that AI delivers. ALWAYS. Without exception. Because if you don’t, you may find yourself in a dangerous place, much more dangerous than the Dean’s office. You may find yourself in a place as dangerous as Harold found himself at the Battle of Hastings. People are dying and becoming severely ill because they trust AI. I am not kidding you. Read about it HERE, HERE, HERE, and HERE. It’s not only that AI fails in trombone research, and other historical research. It’s not only that when you have AI write a paper for you, it will be full of incorrect information delivered with immaculate spelling and punctuation. AI fails at anything and everything. ALL. THE. TIME. Verify, verify, verify.
My sixth grade granddaughter knows that AI is not to be trusted. Now, what do YOU make of all of this?