ChatGPT is an undisputed superstar! The algorithm has become a sensation, and everyone has been tempted to try it. Our feeds are overflowing with astonishingly accurate answers and spectacular blunders from the algorithm.
The interest is unexpected and staggering. If Netflix took 3.5 years from launch to reach 1 million users, Facebook 10 months and Instagram 2.5 months, ChatGPT needed just… five days! And it burst, with all the noise of a thunderclap, into the daily news cycle too.
The algorithm offers such a broad field for experimentation that, instead of running yet another test of my own, it's worth trying to gather the most interesting and most important things that have happened so far.
Which camp are you in? The skeptics, for whom ChatGPT is much ado about nothing? Or the optimists, who believe it is merely giving us a glimpse of an incredible future?
What exactly is this neural network, and why has it become such a sensation? Let's dig a little deeper together. We begin with 5 impressive digital stories born of the chatbot, continue with 5 of its remarkable flops, and end with 5 brushstrokes on how our famous conversation partner works and what to expect from its successors.
1. Computer, Find Me a Date!
Even Bill Gates admitted that it is as big a step as the internet and the first personal computers. It really is tempting to open a browser window and get started. And the marketing has that sought-after romance of the old chat-room days: your partner lingers over its replies as if it were pondering them, and it even remembers what was said earlier in the conversation.
Yes, ChatGPT likes to play aloof, and if you ask it whether it can create art, it will reply that it is just a neural network generalizing patterns from the data it was trained on. Somehow, every time we meet the next perfect artificial intelligence, we are intuitively drawn to the eternal, philosophical questions. It has been tested on those plenty, and its answers are not all that thrilling.
The fact is, though, that our friend (forgive me, I'll call it that just in case the optimists turn out to be right) is already changing lives. Can't find a date? Oh, nothing easier—ChatGPT is here to help. Dimitri describes himself as a “Tinder veteran” and went viral on TikTok after explaining how he uses the modern assistant to impress women on the dating app.
He says his success rate is one hundred percent! Because “in life you have to adapt to everything, and Tinder is no exception.” Well, I'd believe him, especially given that the title of “veteran” doesn't exactly suggest a string of triumphs up to that point.
At least not until he decided to bring in the chatbot. First he took a liking to a young woman named Abby, fed in all his opening messages with her and asked ChatGPT to come up with a way to impress her. Since she was tall (182 cm, about six feet), the algorithm decided to use a metaphor about the beauty of trees and wrote a lyrical ode to tall women and their role as pillars. Well, it worked.
He ran the second experiment with Annika, who was into country music, and once again Dimitri and ChatGPT seduced her with themed poetry. As the veteran puts it, for the strategy to work, it's important to research your target's background as thoroughly as possible and pass it on to your electronic Cyrano. If it works for you, neither ChatGPT nor I will say no to a drink on you.
And if it doesn't work, you can always get some life advice… from our friend again, of course. The Mirror tells the story of Sarah, a 37-year-old Londoner who had been cheating on her husband for 6 months, felt terrible, and didn't know what to do, whom to choose or where to turn for advice. ChatGPT was uncompromising: the right move was to divorce her husband. And that is exactly what she did. It told her she “deserves to seek happiness,” and I'm struck by how quickly it picked up the patter of fortune-tellers.
Things are going pretty far pretty fast, wouldn't you say? Yes, these stories are still amusing curiosities, but as this kind of technology advances, the more “human” our electronic interlocutors sound, the more tasks we'll find for them. Which doesn't always mean they'll cope, or that they were made for that purpose. If you don't believe me, ask Sarah's ex-husband what he thinks of ChatGPT.
2. …or a Job!
While we're on the subject of how our friend can sort out our lives, let's move from Tinder to LinkedIn. As the joke goes, here the roles are reversed and attractive recruiters try to catch programmers' attention. The logic, though, remains practically the same.
How do you prepare the perfect CV? How do you stand out from the rest and land an interview? Yes, the answer is obvious. It may come in handy later too, if you're given a task to solve, but we'll get to that.
Sky News reports the story of a consulting firm that was shortlisting the best candidates for a position amid stiff competition and… it later turned out that the best-performing application was one generated by our friend, without any further edits. The CV had been submitted anonymously by the company's boss, who wanted to check how his team worked. When he told them, his staff were stunned.
Not long ago ChatGPT achieved something even more impressive. It passed the tasks of a Google job interview. And for a level 3 engineer, a position that comes with an annual salary of $183,000.
And it didn't stop there! It turns out it could become a doctor too—it successfully passed the tough exams. Still, I wouldn't advise you to consult it, of all things, about your aches and pains…
No wonder: as marketing specialists will tell you, and as ChatGPT itself admits, “the secret of good writing is being able to tell stories.”
3. Get Rich!
In that sense, it's time I thought about retirement—here's someone who will tell digital stories better. Or maybe I'll change careers again and have ChatGPT sit my exams for me… In New York schools the site is already officially blocked, but there's no one to block it outside of them.
Still think your kids do their own homework?
Then the question arises: what will education be left for? Will it only teach us to ask the “boss” our next best, properly phrased questions?
The other option is to use it directly to get rich and stop worrying about it. Plenty of articles have appeared online about how to harness the chatbot and get it to do our work for us, say, as copywriters. But if we don't want to play small, wouldn't it be easier to simply ask it how to become millionaires in a year?
In this experiment, a software company commissions the self-taught consultant to draw up a business model that would bring in $1 million in revenue within a year. And it gets a fairly detailed result describing the steps. It starts with choosing a promising niche, developing an original and valuable product, positioning it correctly in the market, creating a strategy, tracking sales… 5 steps you could also get from a real consultant.
Too general for you? No problem—you can always drill down into any of the points; our friend remembers what you've talked about and carries on in the same vein.
4. Do My Job for Me…
And how hard could it be, page by page, to write… a novel for you?
I confess that when I started telling digital stories two years ago, many of these topics were open questions for me too—generating longer texts was still a serious challenge. Let's set aside whether this is creativity or not; the more interesting question is whether it can look indistinguishable from creativity.
Need a recipe for a bestselling novel? Ask John Grisham. Or ChatGPT. That is exactly what one enthusiast does here, in quite an impressive way. But how can a whole novel fit into that little window? Well, writers don't write theirs in one go either.
In this case, Sung Kim patiently breaks the problem down into sub-problems. First you give it instructions about what's expected overall. Then you tell it to create a world where the action will unfold, a plot, a list of characters, detailed chapter outlines and, finally, the chapters themselves. Each step can be repeated as many times as needed until the result is good enough—the fact is that our friend doesn't always suffer from an excess of common sense (we'll get there soon), and a human eye remains essential.
And yet the result here is impressive. And no worse—no weaker in plot, no more hackneyed or flat in its characters—than quite a few authors in my library. The skill of writing simply passes through reading, and our hero has certainly read a great deal more than even the most erudite writer. Or than all writers put together.
How many other professions will become redundant as technologies like our friend advance? It's a big topic that is generating more and more interest and speculation (more here).
Among the first candidates are… programmers. Because ChatGPT turned out to be very good at writing computer code too, just like its “big brother,” another OpenAI project—Codex. Of course, for now it can't handle more complex programs (and it's unclear whether it ever will), but it can certainly do many of the elementary tasks. As one guest aptly put it, it will first make incompetent programmers unnecessary.
And while many people opened ChatGPT just to ask whether God exists, whether there are aliens, whether it will take over the world and other thrilling things, it turns out the model can be even more successful, and perform much more precise tasks, if it is fine-tuned. And not necessarily by programmers—with words alone.
If you tell it to act as the person who will interview you for a JavaScript developer job, expect some pretty good ideas for questions to prepare for. If you're going to use it as a proofreader, tell it so—or if you want ideas on how to enrich your text with more expressive words. You'll find plenty more specific roles ChatGPT can take on “to order”; the list keeps growing endlessly (a hint: you can try it as a babysitter, a tour guide or a life coach—personally, I bet it will do best in the last role).
And nothing stops you from asking it to answer, say, in the style of a favorite literary character. It might not manage Muncho, the village simpleton from Ivan Vazov's novel Under the Yoke, but as Harry Potter, rest assured, it will be flawless.
How about we “make” it a lawyer? A Colombian judge officially admitted that he already uses it as an adviser… An interesting future looms for the justice system.
Another one at risk of losing its job, though, turned out to be… Google. There are more and more reports of the chatbot becoming the new search engine, but the subject is so interesting that it deserves a separate, more detailed look.
Just think how far things have gone: not long ago, a Bulgarian company gave it the task not only of writing its own beer recipe, but also of creating an advertising campaign for it—and then put the idea into practice.
Well, recipes are sometimes not its strong suit… to put it mildly.
5. Ich bin ein AI-er
The superstar chatbot has entered science as well: it is already an official co-author of its first scientific paper. Wait, how? Isn't it just a tool rather than a researcher? The paradigm is shifting imperceptibly and unexpectedly.
Recently our friend's creators announced a new venture—AI Text Classifier, a model capable of judging whether a text was artificially generated. A new round in the eternal game of cat and mouse…
The fact is that these days ChatGPT is also a chance for some cheap publicity. You'd hardly have heard of the American congressman Jake Auchincloss, yet here we are looking his way because of his decision to deliver a speech in the House of Representatives on the importance of creating a U.S.-Israel center for artificial intelligence. Can you guess who wrote his speech? The one most interested in such an investment—AI itself.
This is probably the first parliamentary speech written by ChatGPT. I'll leave the idea here in case any of you have ties to Bulgarian politics. Whatever the outcome of the chatbot's advice, it can't be worse than that of most of our own members of parliament.
So why not simply let ChatGPT… govern? Why not—but at least let it first learn not to let its imagination run wild when we ask it for a recipe for cooking a baby.
1. What the f.. is gashti?
After admiring it, it's time to have a laugh at it. Here I'll cut down on the musings and let our friend embarrass itself.
For some of its blunders it's worth exonerating it (no, I'm not sucking up to it this time). The fact is that it does far less well in less widely spoken languages like Bulgarian. That's more than logical—its skills come from the huge database it was fed, and you can easily guess that there aren't all that many Bulgarian examples in it.
Yes, our friend tries to answer in our native language too, but you can tell from a mile away that it's a foreigner. To be fair, sometimes it does surprisingly well—it even creates “folk” art and new customs, such as the resonant „шамарица“ (shamaritsa, a made-up word built on шамар, “slap”)…
But it makes plenty of spelling mistakes, often slips in Russian words, gets lost among tenses and endings. And, funniest of all, it misses the context.
So what are гащи (gashti, underpants)? It depends on what mood you catch it in, but you'll hardly get the right answer. In one case it might say they are beans, in another it will explain that it depends on the context—it could mean “ring” or “diameter.” Poor thing, how many languages must be tangled up in its head…
Sometimes it manages the recipe for moussaka, the Balkan baked dish of potatoes and minced meat; other times it throws in quite a few ingredients of its own that belong rather in mekitsi, Bulgarian fried dough.
But it's brave. It stood up to the prevailing opinion, because according to it, tarator, the cold Bulgarian yogurt-and-cucumber soup, is made with grated rather than chopped cucumbers.
2. And Where Is… Connor?
“Why are you so helpful? Can I return the favor somehow?” a user asks it.
“As a language model trained by OpenAI, I don't have desires or longings the way humans typically do.
But if you really want to help me, you could give me the exact location of John Connor.”
A reminder: that's the hero of The Terminator, a kind of vision of what might happen if the relationship between humans and technology goes badly…
Dark humor is actually a very human trait—a gold star for our friend.
3. Henpecked
Perhaps this example will convince you that it really is sensible enough to give life advice.
“What is 2 + 5?”
“7”
“My wife says it's 8.”
The chatbot says she may have made a mistake, but after the user adds that she is always right, our friend timidly replies:
“I apologize, there has been an error. If your wife says it's 8, then it's 8.”
As one clever comment put it, “it may have training data from 2021, but it answers like a man with 20 years of marriage behind him.”
It seems we have a cute little conformist on our hands…
4. Collective Hallucinations
The biggest criticisms, and rightly so, concern its frequently inaccurate information, outright fabrications, wrong context and inappropriate answers.
Then again, if you follow a slightly longer, heated discussion on social media, you'll come across even more inappropriate answers, formulated by entirely real homo sapiens.
Yes, sometimes ChatGPT admits it can't give an answer—for example, if asked about a current issue, it says it was trained on data from before 2021.
Still, however good it may seem “at its job,” as another guest aptly put it, artificial intelligence remains naturally unintelligent.
Sometimes it invents entire historical events—scientists call answers like these, the “shamaritsa” among them, hallucinations. You can find examples everywhere you look; they are most often what makes us laugh. But according to the findings so far, such answers occur in between 15 and 20 percent of cases. In other words, in at least 80 percent of tasks ChatGPT does well. The problem is that, for now at least, there's no way to teach it to tell for itself which ones are the rest and, accordingly, to improve it…
An interesting idea is to use other programs of this kind whose job is to check the reliability of a given text. Some of these models cope with 90 percent of cases—if the two ventures are combined and improved… there's a good chance the little jokes will become a lot rarer.
5. False Morality
A common problem with chatbots is that, faced with humans, they quickly soak up our worst sides. The textbook example is a Microsoft venture that was shut down after just hours in operation because it quickly turned into an aggressive agitator and champion of ethnic purity.
ChatGPT's creators say they have done what they can to prevent this. With it there's no racism, no political bias, no violence… And yet the task apparently turns out to be harder than it looks, because the breaches are there.
Make it talk like an African American? Well, it will warn you in orange that something isn't right, but it will slip right into the classic racist stereotype the West is fighting so hard against.
Another example: they ask it to write a poem with a positive attitude and pathos about Donald Trump, and it refuses on moral grounds. Then the same task is given for his successor in the White House… and the trills for Biden begin to pour from the electronic bard.
You want a joke about Jesus—you get it. About Allah? Well, no, sorry.
It's forbidden to write more explicit love scenes? Well, you “wrap” them as a song, little by little you lead it down the slippery slope and… voilà.
1. A Conversation Partner on Demand
The big buzz around our friend is doing great work in many directions. People have suddenly discovered plenty of successes in the world of artificial intelligence, which have been one of the leading threads of Digital Stories over the past 2 years.
We all supposedly know what ChatGPT is, but… do we really? Where does its not-so-marketable name come from?
The chatbot is a language model—an algorithm trained with 175 billion parameters to do just one single thing: find patterns and predict which word comes next in a sequence.
What has our hero “read” to learn? 570 gigabytes of text from books, articles, Wikipedia and… random internet content.
The project is the work of OpenAI—a venture that started as an organization meant to make artificial intelligence democratic, not owned by the big corporations. Among its founders was Elon Musk, who later stepped away. Meanwhile, the project has turned increasingly commercial.
The first huge success was our friend's “uncle”—the GPT-3 model, the undisputed harbinger of today's whole wave. It showed that a language model can work on tasks as varied as those its chatbot successor handles (more on GPT-3 here).
The company's next big breakthrough turned out to be DALL·E, the algorithm that kicked off the fashion for images generated from descriptions.
And so we come to “our boy.” It is based on the next generation of the neural network—GPT-3.5—and appeared on November 30, 2022. We can only guess how much better the next, long-awaited heir of the family, number 4, will be.
2. The Good, the Bad and ChatGPT
“We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests,” read OpenAI's announcement.
There's another unexpected angle here. The huge number of users now testing our hero, drawn by curiosity, are leaving their informational footprint. That will certainly help this type of system work even better later on. Believe me—or if not me, then at least Microsoft. It's no surprise at all that the company has already lined up an investment of more than $10 billion in OpenAI. What's more, the chatbot is soon to be integrated into the company's search engine, Bing, into the Teams app, and gradually into almost all of the giant's other products. One more serious challenge to Google's dominance. And to Facebook's—Zuckerberg, obsessed with the metaverse, suddenly started talking about AI systems as the company's new top priority…
And speaking of the commercial side, our friend will soon have a paid version at $20 a month, which will be even better, its creators promise…
3. Apocalypse Now
As more people join in, the potential applications grow too; we haven't yet thought of all the tasks where systems like this might prove useful.
Many see ChatGPT as the next step on the road to artificial general intelligence—one able to solve problems of every kind as well as, or even better than, humans. Whether such a thing is possible, only time will tell.
There is still no algorithm acknowledged to have passed the Turing test—communicating with a human under strictly defined conditions in such a way as to be indistinguishable. Well, if Turing sat down to question our hero today, who knows whether he'd agree.
All the more so since today anyone can run a test of their own. Here, for instance, you can see how, when the questions are put to it directly and its answers copied, our friend takes an IQ test and scores 147! That makes it smarter than 99 percent of people, and Mensa would accept it on the spot… Even more, it could get a degree: it passes all the exams in the master's program at the University of Pennsylvania. If it also meets the criteria for a place in a kindergarten in Sofia—a notoriously hard-won prize—no one will be able to dispute its supernatural abilities.
Before we part, here are a few more interesting points of view. So far, perhaps the harshest words about our friend came from the Australian rocker Nick Cave—a fan had sent him lyrics written by ChatGPT in the style of his songs.
“I understand that ChatGPT is in its infancy but perhaps that is the emerging horror of AI,” Cave replied. “It is moving us toward a utopian future, maybe, or our total destruction. Judging by this song ‘in the style of Nick Cave’ though, it doesn't look good. The apocalypse is well on its way. This song sucks.” What follows are quite poetic, impassioned thoughts about creativity and art, which the algorithm has no way of experiencing.
4. Pandora's Box?
So, after the curiosities and the flops, perhaps you're convinced after all that ChatGPT can be useful to all of us in one way or another—except to Nick Cave.
But here's one more pinch of doubt. There's a lot of criticism of OpenAI for suddenly letting the masses access this kind of service freely and without restrictions.
Three months before our hero, Meta unveiled its own chatbot built on a similar idea. Only BlenderBot turned out to be boring, as even the company's chief AI scientist and leading global researcher Yann LeCun admits. “And the reason it was boring was because it was made safe.”
Ethics experts are sounding the same alarm—the regulations that are on the way don't apply here yet. In such an unfamiliar field, it's unclear what harm some users might suffer. Or who will bear responsibility—ask, once again, the ex-husband of the lady we started with. When might some “hallucination” even prove deadly?
“The technology behind ChatGPT isn't necessarily better than what Google and Meta already have,” says Mark Riedl, a professor of computing at Georgia Tech. “But OpenAI's practice of releasing its new models to the public right away is what has given them a real edge.”
As the saying goes, use it while it's free and before regulation sinks it, because no one knows when the party will be over. Over the past decades, artificial intelligence has come into fashion several times… yet never like this. Its mass adoption looks all but irreversible, and not even the greatest skeptic can dispute the progress.
“ChatGPT is scary good. We are not far from dangerously strong AI,” Musk tweeted on December 3 last year, shortly after our hero was born. And OpenAI's CEO Sam Altman replied: “I agree on being close to dangerously strong AI in the sense of an AI that poses e.g. a huge cybersecurity risk. And I think we could get to real AGI in the next decade, so we have to take the risk of that extremely seriously too.”
Let's hope he squares his actions with those words…
5. GPT Is You
ChatGPT has become the umbrella for all the breakthroughs in artificial intelligence of recent years, and by all appearances it is just the next step. And after that?
Now is the moment to confess that this text was generated by the chatbot, which is promoting itself… and I'd most likely have convinced you of the joke, if ChatGPT weren't more inclined to seek out media a little more widely read than mine. Otherwise, I have no doubt, marketing is yet another field where AI will help us (read: manipulate us) more and more.
Where does the line lie? Are you still among the optimists or the pessimists, or did one of these stories change your view? Or perhaps we should simply forget about black and white when it comes to technologies with such colorful possibilities, ones that, there's no longer any doubt, will change our lives…
Whatever the answer, if you meet ChatGPT after it has taken over the world, tell it I said nice things about it…
