The ChatGPT Honeymoon Phase is Over…Now What?

12 min readOct 17, 2023


Image made using Dall-E

It’s often said that the initial flush of infatuation fades, giving way to a more sober appreciation.

We’ve all been there — falling in love, that is. The excitement, the curiosity, the absolute enchantment. And then, as the charm starts to wear off and reality sets in, you see the imperfections, the quirks. That was me with GPT-4. But like a long-lasting relationship, the value of our bond goes beyond rosy beginnings. So it goes with my relationship with ChatGPT. It’s complicated…and my aim here is to explain why.

There was a time, not long ago, when every ping from ChatGPT fired off those dopamine receptors. But, as with all love stories that transition from intoxicating whirlwinds to grounded partnerships, the glitz and glamor of our early days has been replaced with a comfortable familiarity.

Our story together? A bit of a rollercoaster. Heart-racing highs, stomach-plunging lows, and the occasional moment when you scream because, well, you’re having that much fun — or because it’s downright terrifying. And while it might sting a bit to admit, our honeymoon phase? Yeah, it’s done and dusted.

Yet, this piece is more than a tale of woe. Quite the opposite. Although I’ve transitioned from starry-eyed admirer to a discerning user, my respect and adoration for GPT persevere. They’ve matured, like fine wine. In truth, I’ve come to grasp the nuanced difference between leaning on AI as a flashy crutch and harnessing it as a powerful tool. And I hope to support that notion.

Think the infatuation is over-hyped? Let’s talk numbers. ChatGPT won over a staggering 1 million users in a mere 5 days. With the claim of having 1 trillion parameters — though whispers in the digital alleys suggest the figure is closer to 220 billion — it’s no wonder GPT-4 turned so many heads. But not all gazes are adoring. For some, it’s the emblematic bubble soon to burst. For others, it’s the harbinger of a dystopian future, the first drumbeat in an AI rebellion. Dramatic? Certainly. But it underscores the polarized reactions this tech marvel has elicited.

And, oh, the AI renaissance it sparked! Suddenly, the market was awash with spruced-up, white-labeled GPT versions, each tailored for a niche. From Bing Chat to Jasper, it felt like ChatGPT had opened a Pandora’s box of AI offerings. And, well, you know how that story ends…

But let me backup a second. See, as a writer, I’ve been spoiled. Spoiled by the sheer thrill of feeding this digital beast bits of information to have it churn out pristine prose, devoid of those pesky grammatical errors. But it’s crucial to remember: While many harness ChatGPT for coding, data interpretation, and strategy, my lens is primarily trained on its applications in writing. And yes, its dazzling sibling Dall-E deserves a nod too — I mean, who do you think jazzed up this blog with images? Shout out to Dall-E. 🙌

So, dear reader, with our rose-tinted glasses set aside, let’s explore this post-honeymoon landscape, and uncover where the enchantment might have fluttered away.

GPT’s evolving evolution

A not-so-quiet shift has been unfurling across our digital spaces. To get specific, a recent poll from YouGov made me pause for a moment: almost 1/2 of Americans have heard of ChatGPT. Impressively, 2/3 of those familiar folks are the ones with postgraduate degrees. Yet, on the flip side, a notable 54% still draw a blank when it’s mentioned.

For the uninitiated, let’s break it down: GPT stands as the heart of an all-purpose AI chatbot using LLMs. When ChatGPT appeared, thanks to OpenAI’s understated launch in late November 2022, it quickly became an industry-changer. No big fanfare. No grand announcements. And, yet, its impact was felt everywhere. The San Francisco-based OpenAI, originally had modest expectations. But soon, they found themselves navigating an unexpected surge of interest.

Rewind a bit to December 2015. That’s when the foundational pillars of OpenAI were set by Sam Altman, Greg Brockman, Elon Musk, Ilya Sutskever, Wojciech Zaremba, and John Schulman. These individuals, each bringing their own unique expertise, aimed for one common goal: propel AI forward in a way that resonates with humanity’s betterment. Over time, dynamics changed — Elon took a step back and now, the captain’s hat sits firmly on Sam Altman’s head.

Financially? OpenAI’s journey has been remarkable. Now valued at a robust $29 billion, with investments soaring to $11.3B. Even tech giant Microsoft sensed the winds of change and bolstered its partnership with OpenAI, infusing a multibillion-dollar investment to fuel AI innovations worldwide.

Tracing the GPT lineage brings its own sense of wonder. GPT-1 was introduced in June 2018. With its 117 million parameters, it marked the first footstep towards what would become ChatGPT’s modern-day prowess. This was followed by GPT-2 in February 2019, which upped the ante with 1.5 billion parameters and delivered coherence in text generation that few could match. Concerns about its power led OpenAI to withhold it. Then, only to release it later in November after understanding its implications better.

GPT-3 in June 2020 changed the landscape. With its 175 billion parameters, this model wasn’t just about text generation — it showcased versatility, dishing out everything from emails to poetic verses, translations to programming. With GPT-3, people got hands-on with ChatGPT, realizing the vast opportunities this tech could stretch to.

Our latest protagonist, GPT-4, elevated the narrative even further. Improved alignment with user intentions, increased factual accuracy, enhanced steerability, and the avant-garde feature of real-time internet search capabilities make it a marvel. Ultimately, it’s all about the qualitative leaps like these.

Source: ChatGPT

However, beneath the cool facade of ChatGPT lies a fascinating architecture. It tries to grasp your prompts and produces responses based on a vast expanse of training data. Contrasting traditional supervised learning, ChatGPT functions on a generative pre-training method where it processes a universe of unlabeled data, gleaning patterns and relationships from almost everything on the open internet. Tokens, or chunks of text, form the core of its understanding. While GPT-3 processed roughly 500 billion tokens, GPT-4, with its enhanced prowess, has remained an enigma in exact numbers.

All this information, all these tokens, come from the myriad content created by humans. So, when you think of ChatGPT’s operation, think of it as a highly advanced version of the “fill in the blanks” exercises. It comprehends your input, processes it, and then delivers an output. But remember, “understand” and “comprehend” are merely the closest terms we have. To be clear, it’s not genuine cognition.

GPT-4: The magic, the myths, and the tech behind the curtain

Image made using Dall-E

There’s a strange allure to things that promise the world. Their vast capabilities and almost mystical aptitude draw us in. Yet, anything with such immense power inevitably has its limitations. And Generative AI is NO exception.

Reflecting on my experiences with GPT-4, I recall the wonderment I initially felt. It was reminiscent of the Wizard of Oz — a grand entity with untapped potential. But after spending considerable time with it, I’ve come to see it more as an eager intern. One whose assistance I value immensely, still who also has a lot to learn.

The web has always served as a societal mirror, reflecting our virtues and vices alike. With the rise of Language Learning Models (LLMs), these digital reflections risk becoming permanent. ChatGPT, for instance, is trained on vast textual data from the internet. This means that while it’s undeniably smart, it can sometimes regurgitate biases, prejudices, and even glaring inaccuracies. It might churn out answers that sound correct but miss context or relevancy. For the record, it’s crucial to verify any information GPT provides. Especially when details and context matter.

In a particularly jarring episode, 2 lawyers in a Manhattan federal court claimed they were “tricked” by ChatGPT into including fabricated legal research in a court filing. Such episodes are eye-opening reminders of the unforeseen challenges and responsibilities of using such tools.

Yet, while I recognize these limitations, I have always opposed GPT for information gathering of any kind. Why? The risk of absorbing and sharing misinformation is too significant. And even if GPT sourced accurate data, I trust my own analytical prowess to meld that information.

Additionally, ChatGPT, though conversational, misses the mark in emotional intelligence. More often than not, it fails to capture the subtle nuances of human conversation. This might pose a challenge for SEO too. Google’s Spam policies are strict against “auto-generated” content that lacks originality. Given that a significant chunk of website traffic is from organic searches, relying heavily on AI-generated content could prove detrimental. Consider the fact that 53.3% of all website traffic comes from organic search. Likewise, damaged SEO health can be hard to pull yourself out of.

Let’s not worry too much about that yet though. 😬

The world of marketing seems to be especially inching towards a repetitive loop. Many claim that there are no new ideas to be found. After all, if GPT-written content floods the internet, will the AI just end up recycling the same data over and over? We could be witnessing a digital Ouroboros — AKA the serpent forever devouring its tail.

On the technical side, while ChatGPT can craft coherent sentences, long-form content still challenges it. It lacks the organic flow that a human writer brings to the table. Provide it with too little information, and it spins wild tales. But overload it, and you might lose depth and context. It’s a balancing act. And it’s not an easy one either.

Moreover, recent murmurs indicate a decline in ChatGPT’s response quality. A startling study by researchers from Stanford and UC Berkeley unveiled that GPT-4’s performance, on a specific task, plunged dramatically over a few months. This hints at the influence of “behavioral drift”, a phenomenon that can disrupt the workflows and reliability of products built on LLM AI APIs. While some fluctuations might introduce diversity in content, their unpredictability can pose challenges.

As with any tool, ChatGPT has its strengths and weaknesses. It’s neither a flawless sorcerer nor a mere rookie. It lies somewhere in between. It’s an entity teeming with potential but also bound by its limitations. I value its contribution to my work, but with a watchful eye, always mindful of its innate imperfections. After all, isn’t that how we should approach all tools and technologies? With respect, understanding, and a healthy dose of skepticism?

Reflecting on both utility and artistry

There’s no denying the myriad ways to employ GPT-4. And I’m not alone in being swept up by this technological masterpiece. Among employed adults acquainted with this tool, a distinct pattern emerges, according to a study from Pew.

Those under 50 have a higher inclination than their elders (18% vs. 10%) to use ChatGPT for work. Then, there’s the educational divide. Adults with just a high school diploma who have stumbled upon ChatGPT tend to use it less than those with higher educational qualifications — especially for learning.

Collectively this brings up an interesting question–how might GPT4’s impact vary by profession? Age group? Will software engineers, graphic designers, and journalists find their roles shifting because of chatbots?

Many seem to think so.

I still passionately believe in GPT-4’s value, both personally and professionally. The other night, with a hodgepodge of groceries, my fiancé conjured up a delightful recipe with GPT’s guidance. But like any tool, your success with GPT-4 depends on several factors:

Knowledge of GPT: There’s no user manual for GPT–which is actually OK, because it’s more important to understand its idiosyncrasies. Be wary of its cliches — yes, the overuse of symphonies and tapestries, among others.

Good writing judgment: Can you tell a compelling narrative from a dud? It’s essential to weed out the fluff and look for varied sentence structures.

Articulation: Be clear in your instructions. GPT isn’t a mind-reader. The more specific and direct you are, the better results you get.

Management skills: Offering feedback effectively and spotting any concerning tendencies is crucial.

Tech-savviness: Although I certainly don’t claim supreme tech expertise, those who do can harness the power of GPT-4 even more efficiently.

Now, in full transparency — communication with ChatGPT requires a certain finesse. My personal writing quirks, such as using three adjectives to set the tone or avoiding a string of negative constructions, are essential. And yes, there are times GPT loves to pull from its bag of marketing cliches or overused analogies. Being precise in instructions helps. For instance, I often emphasize the inclusion of all information and quotes. Particularly for longer blocks of tips.

Another tip? Be kind. Odd advice for interacting with AI, perhaps, but treating it courteously has often resulted in improved outcomes.

I’ve also found that refinement is key. Sometimes nudging GPT in a different direction yields better results than starting from scratch. And always provide context — it’s like handing someone a map versus just an address.

Last, but not least, of course, there’s no dearth of advice on “prompt engineering.” LinkedIn might be awash with self-proclaimed experts offering the definitive guide. However, remember, everyone’s experience with GPT-4 will be unique, molded by individual objectives and preferences.

Source: Search Engine Journal

Now, for the big elephant in the room. AI detectors. Do they work? In my personal tests, the results have been…mixed, mainly pointing towards no. While they can spot direct outputs, a touch of refinement often fools these detectors. Plus, recent studies corroborate my observations. The rapid development of Generative AI seems to be outstripping the pace of detection technology. The implications? Non-native English speakers, for instance, could find themselves unfairly penalized due to the limited linguistic expressions these detectors recognize. Ethical AI anyone?

Final thoughts on the paradox of progress

Image made using Dall-E

In the brisk march of modern tech advancements GPT-4 has roused polarizing opinions. I mean, no surprise there, right? There are the purists who scoff and dismiss, corporations that gleefully invest, and the nihilists who view it as the ticking time bomb of the apocalypse. I, however, find myself oscillating between awe and introspection, much like a moth enchanted by the flickering glow of a candle, wondering if it’s about to get burned.

But let me ask you this, isn’t AI’s method of creativity a mirror to our own? The annals of history are replete with instances of human creativity blossoming from the seeds of imitation. The Renaissance painters were besotted by Classical Greek art. Centuries later, we still unearth Shakespearean echoes in the tales we tell. It’s uncanny. Yet still, a recent study nudges us towards a confronting truth: pure originality is perhaps an illusion. Creativity is built on the foundation of memory and control. It merges past experiences in strategic alignment to craft the “novel.” Sound familiar? It’s uncannily similar to the digital musings of our AI counterparts.

There’s a prediction by the savvy folks at Epoch AI that’s been haunting my thoughts lately. Come 2027, and we might see the sunset of the AI creative surge, starved of quality human content. Some would lament. Others might shrug. Oddly enough, I find a comfort in that projection.

You see, in the sprawling bazaars of AI-generated content, there’s something unmistakably alluring about a piece handcrafted by human thought. Much like that handmade vase that stands out amidst assembly-line decor, human musings have that unpredictable, irreplaceable quirk. It’s the imperfections, the rough edges, the pauses, and the crescendos of human narrative that keep us coming back for more content.

The increasing capabilities of AI can be seductive, luring us into a sense of complacency. Still, remember, just as a car needs a driver, AI, for now, needs us. We endow it with direction, purpose, and soul.

So, ultimately, stepping out of my AI honeymoon phase has been surprisingly liberating. Refreshing even. By not placing it on an unreachable pedestal, I’ve shed my naivety, allowing room for pragmatic optimism. It’s no longer the magical Oz — just some geeks training a computer behind a curtain. And what a formidable tool it is, banishing the tedious to grant us the luxury of time.

Our relationship with AI, particularly GPT-4, is an ongoing, ever-growing one. We moved past the infatuated first glances, settling into a partnership of mutual respect. My idealism may have waned, but it has carved space for a grounded appreciation. Because while AI can weave tales, it’s our stories, our songs, that pierce the heart and linger in memory.

We are the architects of this digital marvel. Shouldn’t we then steer this relationship with the wisdom of creators? For in the end, the tools we forge are but reflections of our aspirations, fears, and dreams.




Zeitgeisty = your compass to the culture cosmos. 🔭 Ideal for the marketer's muse, the journalist’s journal, or if you just love being culturally in the know.