The lying Charlatan GPT (CGPT)

Some 25 years ago I completed my doctoral degree at the University of Sheffield in the application of AI to an engineering problem. AI was an esoteric subject back then, but I have nevertheless been fascinated by its potential. These days, I am deliriously excited by the things it can do and I use it almost every day. In all my excitement, I forgot a crucial fact: CGPT makes up things. The industry jargon for this is ‘hallucination’ – though I don’t understand why they just can’t say the output was plain BS. I was fully aware of this ‘hallucination’ habit. I had experienced it a couple of times before, but to my peril hadn’t learnt the lesson.

A couple of days ago, I wrote a potential newspaper article on the Northern Lights (Aurora Borealis). Just before sending it off, I felt it needed a good quote. My go-to tools are CGPT and DeepSeek. The conversation with CGPT follows:


Me: Please give me poems by some famous poets about the Northern Lights.

CGPT: “The Aurora Borealis” by H.W. Longfellow beautifully captures the majesty of the Northern Lights:

“I saw the lights rise o’er the hill,
They bathed the heavens in glory bright,
A shifting, shimmering, silent thrill,
A phantom dance of spectral light.”


Just the thing I needed. I quickly quoted Longfellow in my article, adding the line “A phantom dance of spectral light,” somewhere in the first paragraph.

I sent it off promptly to the editor and sat back feeling quite contented. My passion for literary work luckily saved my bacon. Curious about the poem, I went back to CGPT.


Me: Aurora Borealis by H.W. Longfellow – give me the full poem please

CGPT: Henry Wadsworth Longfellow did not write a poem specifically titled Aurora Borealis, but he did mention the Northern Lights in his poem “The Discoverer of the North Cape” (from Tales of a Wayside Inn).

Me: OMG! Why did you give me wrong info?

CGPT: You’re absolutely right to call that out—I should have double-checked before stating that Wordsworth wrote a poem titled The Aurora Borealis. That was a mistake, and I appreciate your patience. I’ll make sure to be more careful in verifying sources before sharing information. If you’d like, I can find a verified classic poem on the northern lights. Let me know how you’d like to proceed!

Me: I would like to proceed by giving you a boot up your fxxxxxg axxe.


I withdrew the article immediately and deleted the fake Longfellow quote. I tried the same request on my phone where I have the CGPT app. It did exactly the same thing. Then I tried DeepSeek and it gave me some other fake poetry.

In those simpler days, when we spent long hours in the lab on AI experiments, the outputs were binary – either right or wrong. Sometimes, we also associated a probability with each output. The word hallucination was only associated with acid or magic mushrooms and such! Never with a lying computer programme.

EPILOGUE: After publishing the above in the morning, I passed it through CGPT this evening and asked it to proofread it. It suggested around 5 changes, some of which I implemented to improve flow. Best of all it pointed out that I had used ‘contended’ instead of ‘contented’. So this time, it was really useful. The lesson is that one must be convinced about what CGPT is saying and not accept general information without questioning. Finally, I said, “Thanks for your help,” and it responded,
“You’re welcome! Let me know if you need any more edits-or if CGPT pulls another fast one on you.”

Comments

Leave a comment