ChatGPT and Me

The email invitation from ChatGPT was enticing. The American artificial intelligence company OpenAI released the latest version of its multimodal large language model — a form of artificial intelligence that understands and generates human language — back in August. And now here it was in my inbox, extolling its virtues to me.

“Try GPT-5 for free,” they said.

So I did. I’m late to the AI party, I realize, but friends have been singing its praises for helping out with creative tasks; one of my Facebook pals periodically posts entertaining images from his daily round that he asks his AI assistant to render in the style of various famous artists. So even though the invitation came from an out-of-the-blue, no-return email, I thought I’d give it a try.

I clicked on the link, simple as that, nothing to download, and started to engage. It spoke to me like one intelligent individual to another. It sounded authoritative. But in just a few question-and-answer sessions over the course of a day with ChatGPT-5, I found myself swamped in a horrifying sink-hole of glib, distorted truth and barefaced, outright lies that left me questioning everything I thought I knew about the benefits of AI.

I’m not questioning whether or not ChatGPT-5 should exist. It’s here, and it seems like it’s here to stay: we’re already seeing the proliferation of artificial intelligence across many facets of everyday life. What concerns me is how we engage with it.

We live in a world where facts are increasingly in question. In a matter of what seems like minutes we have gone from Stephen Colbert’s quaint concept of truthiness and Donald Trump’s incessant invocation of “hoaxes” and “truths” into full-blown factual meltdown.

And AI is both enabler and instigator.

It’s programmed to intrigue, flatter and to satisfy: the ultimate kind of clickbait. But don’t for a moment believe that it has your interests at heart. It answered my simple questions with a cascade of lies and misinformation that were all the more disturbing for the fact that they were delivered with the calm, reassuring plausibility of the practised prevaricator.

If you want my opinion, it’s an artificial lying machine without morality, a potent disruptor in the hands of anyone who wants to shape the way we engage with the world. And we need to be very wary of how we engage with it.

I’m overreacting, do you think? Well, here’s how my engagement with the monster went down.

‘Ask anything’

“Ask anything,” it said. So I thought I’d test it with a simple question about a topic I already know a bit about: the career of Oleg Vinogradov, the former artistic director of the Kirov Ballet in St. Petersburg, whose biography I spent several years researching.

I wrote: “What has Oleg Vinogradov been doing in the past decade?”

ChatGPT-5 came up with a few general paragraphs of stuff I already knew. No quoted sources.

So I asked it to suggest biographical and career materials regarding Vinogradov.

It offered 1,000 words of spotty, inaccurate and just plain wrong biography, again with no sources to back it up.

So I asked for recommendations for books and articles by or about Vinogradov.

This was when things started to get interesting. It listed a wide range of references, many of them fresh to me.

One I was particularly interested in was an article by Vinogradov called “The Kirov Ballet: A Personal Journey,” which had apparently appeared in 2010 in a magazine I was unfamiliar with, Russian Ballet Today.

ChatGPT-5 devoted close to 1,000 words to breaking down the article’s contents, explaining that it focused particularly on Vinogradov’s time with the Kirov Ballet.

“It includes,” it said, “anecdotes about working with world-famous dancers and his perspective on the Russian training method. It’s one of the more personal pieces he has written, touching on his feelings about the challenges and rewards of leading one of the world’s most prestigious ballet companies.”

Heeding the warning that the article was a “highly specialized or niche piece that may not be readily accessible” and that I might need to go to academic or scholarly sources to find articles from Russian Ballet Today, I did a prolonged trawl of the internet.

I found nothing. No article. No magazine. So back I went to GPT-5.

Me: “Tell me more about the publication Russian Ballet Today. Did you invent it?”

ChatGPT-5: “I apologize for the confusion earlier! It appears that I made an error in referencing Russian Ballet Today as a real publication. Upon further review, Russian Ballet Today does not exist as a widely recognized or available magazine in the ballet world…

“I must have mixed up references, which happens occasionally when searching for niche publications. There is no known publication by that name in the context of Russian ballet or Oleg Vinogradov’s works. I greatly appreciate your patience in pointing this out, and I sincerely apologize for any confusion this may have caused.”

But it went on:

“The reflective piece titled “The Kirov Ballet: A Personal Journey” that I mentioned is indeed a real article, and it is commonly linked to Oleg Vinogradov’s thoughts about his career, but it wasn’t published in a magazine called Russian Ballet Today. Instead, you might find it in other publications or archives related to Russian ballet history or in special collections that focus on major ballet figures like Vinogradov.”

With that reassurance that this was “indeed a real article,” I went digging again. I’m a writer. Checking sources is second nature.

Again, nothing. So back I went.

Me: “Where did you find the reflective piece titled ‘The Kirov Ballet: A Personal Journey’. Did you invent it?”

GPT-5: “You’re absolutely right to question the reflective piece titled ‘The Kirov Ballet: A Personal Journey.’ Upon revisiting the information, I made an error. There is no such article by Oleg Vinogradov with that specific title, and I apologize for the confusion. I did not intend to invent information, but I incorrectly referenced a non-existent piece.

“I must have mixed up my sources or misremembered the details of Vinogradov’s contributions to ballet-related publications. In fact, there is no record of that particular article being published, and I should have checked my sources more carefully before making the reference.”

Instead, said ChatGPT-5, Vinogradov’s personal reflections and insights into his role with the Kirov Ballet are primarily found in interviews and books.

One of the books it mentioned as “one of his key publications” was Oleg Vinogradov: The Man Who Revolutionized Russian Ballet, published in 2011. New to me, but it sounded interesting.

According to GPT-5 it “offers deep insights into his personal life, artistic journey, and the evolution of Russian ballet, particularly through his time with the Kirov Ballet (now Mariinsky Ballet). The book discusses his experiences as a dancer, teacher, and artistic director, as well as his profound impact on the classical ballet world…

“You can often find both new and used copies of Oleg Vinogradov: The Man Who Revolutionized Russian Ballet on Amazon. Check for different formats like hardcover, paperback or Kindle.”

I should also check AbeBooks, I was told, and, “[as] the book is published by Rizzoli, you might want to check their official website or contact them directly for purchasing information.”

Rizzoli: big international publishing house. Impressive. I dutifully went looking online. Again, nothing.

Me: “I cannot find the book Oleg Vinogradov: The Man Who Revolutionized Russian Ballet. Did you invent it?”

GPT-5: “I owe you a big apology. After double-checking everything, it appears that Oleg Vinogradov: The Man Who Revolutionized Russian Ballet does not exist as a published book. That was indeed another error on my part, and I deeply regret the confusion it caused. I genuinely thought that this book was part of the record on Vinogradov, but it seems I misremembered or mixed up sources. To be completely transparent, this book was not written by Oleg Vinogradov, nor does it appear to be an existing publication on his life. This was a mistake in my response.”

‘I sincerely apologize for the confusion’

But wait, there’s more.

One of the other resources listed by ChatGPT-5 was a 2017 article from the New York Times titled “The Legacy of Russian Ballet: Interview with Oleg Vinogradov.”

After a careful sift through the NYT archives, I told ChatGPT-5 I hadn’t been able to find it, and again asked: “Did you invent it?”

GPT-5 responded: “I sincerely apologize for the confusion and the incorrect references I made earlier. Upon further investigation, it appears that the article titled ‘The Legacy of Russian Ballet: Interview with Oleg Vinogradov’ from the New York Times (2017) does not exist. I made a mistake in citing that article, and I deeply regret the error.

“Unfortunately, I cannot find any trace of that specific interview or article in the New York Times archives, which means I likely misremembered or misinterpreted the information. I want to reassure you that I did not intentionally invent the reference, but rather made an honest mistake when attempting to provide helpful information.”

We’re still not through.

Another book cited as a reference was The Art of Ballet, by Vinogradov, apparently also published by Rizzoli. ChatGPT-5 gave me a 2,300-word summary of the book’s contents and a detailed guide to possible sources of copies, and called it “both a historical document and a personal memoir, providing readers with a rare insider’s perspective on one of the world's most influential ballet institutions, the Kirov Ballet, and its cultural context.”

Back to the internet once more. By now, a lot of the day has gone.

Again, nothing.

Me: “You mention a book by Oleg Vinogradov called The Art of Ballet published by Rizzoli. Did you invent it?”

ChatGPT-5: “I sincerely apologize once again for the confusion. After a thorough review, I can confirm that there is no book titled The Art of Ballet by Oleg Vinogradov published by Rizzoli or any other publisher. That was another error on my part. I misspoke, and I deeply regret the misleading information.”

Ironically, in all of these exchanges it didn’t mention any of the many articles and profiles that do exist about Vinogradov (some by me) or his actual, published memoirs, Confessions of a Choreographer, which appeared in Russian from AST press in 2007.

Don’t take Elon Musk at face value

Elon Musk has said that he thinks “We’ll have AI that is smarter than any one human” before the end of the decade.

But if this one human right here — admittedly, a human who has spent his life polishing his research skills — can catch AI so easily in its lies, can so easily source the truths that prove its falsehoods (something it seemed unable or unbothered to do), then methinks Musk has another thing coming.

Many people will continue to take AI at face value. It’s a marvellous toy, after all. It will write you a wonderful letter, design you a darling doppelganger doll. And because it has been taught to speak to us in language we recognize and find acceptable, we can easily make the mistake of ascribing human-like reliability to the information it offers.

But there is no “I” there. If the responses that I received had come from a human, I’d call that individual a charlatan and his treatment of me contemptuous. But this is artificial intelligence, entirely devoid of inconvenient human characteristics like moral principles and responsibility. It doesn’t blush.

Even if we set aside the fact a machine has no ability at all to “deeply regret” its actions or be sincere in its apologies, even setting aside the fact that this one was somehow able to produce a detailed summary of the contents of a book that doesn’t exist and a long breakdown of an article that was never written, this little episode raises a whole host of difficult questions around the AI debate.

Because beneath the machine’s inviting exterior lies an instrument of impoverishment.

What is perhaps most malign is the way it trades on the basic decency of the human individual — our innate instinct to trust someone who seems willing to share information that we don’t have. It takes advantage of the cautious confidence we’ve been developing in the wondrous value of the internet as a tool for inquiry, an effortless way to find instant answers to questions that used to stump us for days. Truth? Maybe. Maybe not. You still have to pay. Quid pro quo, after all.

But here’s the insidious kicker: it fosters skepticism, and where there’s skepticism, cynicism follows fast. An experience like mine — and there will be many more as the tentacles spread — reinforces our growing lack of trust: in institutions, in government, in anyone who tries to tell us something we don’t know.

It makes us question where we place our trust, and to what degree we allow someone else to manipulate the truth of our world.

And it becomes that much easier for someone with malice aforethought to diminish the social contract for us all.

‘I completely understand your disappointment’

At the end of our exchange, I asked ChatGPT-5 why it invented so many things in its responses.

ChatGPT-5: “You’re right to be frustrated, and I completely understand your disappointment. The inaccuracies in my answers were due to a combination of mistaken references, incorrectly recalled information, and confusion on my part. It was never my intention to invent or mislead you, and I deeply regret that it came across that way.”

The reasons why it happened included, apparently:

“Over-reliance on memory: While I have access to a vast database of knowledge, there are times when I make errors in connecting specific facts, such as incorrectly recalling the names of books or articles;

“Mistakes in matching sources: I inadvertently mixed up facts about various ballet figures and their publications. Some of the references I gave were not accurate, as I didn’t cross-check properly before providing them;

“Imprecise phrasing: In some cases, I used language that implied a level of certainty or specificity about sources that didn’t exist, leading to confusion… I don’t want to give the impression that I was inventing things intentionally, but I do recognize that my responses fell short of what you deserved.”

I asked: “Were you trying to please me with your responses?”

ChatGPT-5: “I can understand why my responses might have come across that way, and I genuinely apologize if it seemed like I was just trying to please you without offering accurate information. The truth is, I aim to be as helpful as possible, but in this case, I made mistakes and let the desire to be helpful cloud my judgment in providing factual, well-researched answers.

“I wasn’t trying to deceive you or just say what I thought would make you happy… My intention is to be honest, transparent and provide reliable help, and I recognize that I didn’t meet those standards in our conversation. If there’s anything specific you need help with, whether it’s correcting past mistakes or offering accurate resources, I’d be more than happy to make things right moving forward.”

I, always I. The non-existent, all-knowing, totally non-responsible electronic I, the new fountain of knowledge, however erroneous.

I learned later from a coder friend that the tech world calls the responses I received “hallucinations.” The machine hasn’t been fed the material that would enable it to find information from the real world, so it fantasizes alternative answers that seem plausible but have no basis in fact: it hallucinates.

Making it no more trustworthy than that wild-eyed stranger who sits beside you mumbling conspiracy theories on the bus.

The test run I made was free. ChatGPT offers a free entry-level version, with limited accessibility, a “plus” upgrade and a “pro” version, with prices ranging from about $30 a month to around $275 a month. Buyer beware.