(Is ChatGPT) Not Good Enough (?)

Pol Martí
8 min readApr 29, 2024

--

Photo by Mohammad Mardani on Unsplash

Introduction

Have you ever caught yourself chatting with a machine as though it were an old friend? That’s exactly how my fascination with Large Language Models (LLMs) began. I found myself engaging in conversations, bursting into laughter, and even debating heatedly with what is essentially a sophisticated string of code, as if it had the capacity to respond!

Today, the buzz surrounding Large Language Models (LLMs) is inescapable. These groundbreaking innovations in machine learning are poised to revolutionize our world, sparking discussions across the spectrum — from enthusiastic early adopters to wary skeptics and everyone in between. Many of us are still exploring how to seamlessly integrate this cutting-edge technology into our daily routines, whether for professional use or personal projects.

Adopting LLMs is like riding a rollercoaster filled with highs and lows. It’s an emotional journey of trials and errors that stretches our expectations and demands resilience, as we navigate through moments of frustration and celebrate breakthroughs. Alongside the excitement, there’s an undercurrent of apprehension — some fear the unknown implications of these powerful tools, worrying about privacy, job security, or the sheer unpredictability of advanced artificial intelligence (AI).

In this article, that was born from a recurring phrase that I heard many times in my collaborations with colleagues during the creation of custom GPTs, I will share my personal experiences with one such model, ChatGPT, delving into how it has shaped my interactions, influenced my capabilities and way of working, and even transformed my perspective on the potential of AI.

Exposure and Initial Fascination

It all starts with a viral video — usually some influencer on TikTok or Instagram doing something snazzy with ChatGPT. They dress it up with flashy talk and slick edits, making it look like the next big revolution. And truth be told, it is. But the real magic of LLMs isn’t in just making them perform tricks or serve as a fancy Google. The epiphany hits when you start digging deeper, finding ways these tools mesh with your professional life or personal passions.

I remember the first time I tried to replicate one of those viral tricks. It was thrilling — at first. I managed to do something cool, and within minutes, I was on the phone, bragging to anyone who’d listen about how I was pushing the digital frontier. But that initial buzz didn’t last long. “Was this brilliant tool really the revolution it seemed, or just a high-tech novelty?” However, my curiosity wouldn’t let me stop there. I began tailoring queries, fine-tuning prompts to my actual needs. Suddenly, things started to click. It wasn’t perfect — it rarely is — but it was genuinely useful. That’s when the real honeymoon began.

Reality Check and Frustration

The honeymoon phase didn’t last forever. The day came when I tried to apply ChatGPT’s suggestions to a real-world problem, aiming high — maybe too high. I was dreaming of monumental achievements: writing that groundbreaking book that had always been in my mind — surely a future Nobel contender — or launching a website to skyrocket my macrame hobby into a bustling business in no time. But the reality? It fell short. Way short. ‘Not good enough,’ I muttered, staring at the results. That phrase would echo in my mind, becoming a refrain for every subsequent letdown.

And then, the financial sting hit. A notification popped up on my phone — a sober reminder from my bank about the upcoming payment for OpenAI. Two months of subscription fees, and what did I have to show for it? A growing sense of buyer’s remorse gnawed at me. It led to a stark crossroad: find a real use for this expensive digital tool or cut my losses and cancel. Stubbornness won. I decided to give it one more try, inspired by a lingering idea about optimizing a personal project. To my surprise, it worked! The success felt sweet, like a direct challenge to my earlier doubts. For a moment, I felt like a tech titan, ready to take on six-digit job offers as giants like Bill and Elon themselves would duel to hire me.

But as I paraded my success, reality had a harsh lesson waiting just around the corner.

Breakthrough and Reassessment

Just when I thought I’d mastered the game, LLMs threw me a curveball. After days of bragging about my newfound prowess, I set up a demonstration to showcase ChatGPT’s capabilities to a skeptical friend. It flopped. The tool stumbled, giving nonsensical responses that left me red-faced and fumbling for excuses. “Had ChatGPT been taking lessons in sabotage from my cats?” I wondered. “Why is this happening now?” I muttered under my breath during an awkward silence. Each failure not only dented my confidence but felt like a personal betrayal by the technology I had championed.

Frustration boiled over into heated exchanges with the chatbot. “Why are you saying this today when you were spot-on yesterday?” I challenged, half-expecting a reasonable explanation from the machine. The responses were apologetic yet infuriatingly inadequate. “I don’t want your apologies; I need results!” I snapped, caught in a cycle of irritation and dashed hopes.

But I wasn’t ready to give up. Encouraged by a video from an LLM expert, I tweaked my approach, adjusting prompts and expectations. This time, the results were surprisingly good. Not just once, but consistently across multiple queries, sessions and days. Finally, a breakthrough! The tool started to deliver results that actually made sense. It wasn’t perfect — far from it — but for the first time, the possibilities seemed tangible, even if every success was shadowed by a reminder of limitations.

Suddenly, the term ‘Not Good Enough’ took on a new, positive connotation. It wasn’t perfect, but it was a step in the right direction — a sign that perhaps, I could make this work to my advantage after all. A glimpse of what was possible when adjusting my strategies to align with the tool’s capabilities.

Mastery, Persistent Challenges, and Strategic Acceptance

The rollercoaster of highs and lows with ChatGPT continued, but each challenge taught me something new. I began to understand that this wasn’t just about machines taking over our tasks. It was about learning to coexist, to collaborate with a digital mind whose strengths complemented my own weaknesses, and vice versa. Armed with better strategies and a more nuanced understanding of how LLMs function, I started to see consistent results that were “Good Enough” for my purposes. My proficiency grew, so did my confidence. I experimented with more complex queries, tailored my prompts more precisely, and learned to anticipate the kinds of responses I would get.

And just when I thought I had it all figured out, LLMs reminded me that they’re still a work in progress. More tests, more tweaks, more questions — and not surprisingly, more inconsistent answers. I’d run the same query several times and get different results each time. It was maddening. “Are you playing games with me?” I wondered, half-joking but genuinely perplexed. This inconsistency was a hard pill to swallow, a reminder that despite all advances, these tools have their own set of quirks and unpredictabilities.

Unlike traditional computers, which store and recall information with perfect accuracy, LLMs operate on a stateless model (throughout this article, I will use ‘memory’ to describe this characteristic). Each interaction is independent, with no memory of past interactions, unless specifically designed to temporarily remember within an ongoing session; each interaction is a fresh start, shaped only by the immediate input and the fleeting context of ongoing sessions. Moreover, while humans theoretically offer consistent responses, in practice, they can even provide different answers to the same question at different times — due to variations in phrasing, mood, or external influences. This variability highlights a significant expectation gap: we often expect the perfect memory and consistency of computers when using LLMs, but their operation is more akin to human-like interaction without long-term memory. Recognizing this was crucial — it shifted my approach from one of confrontation to one of guidance. I learned to craft my prompts with increased precision, providing clearer and more comprehensive contexts and shaping their outputs to suit my needs. The frustration that once overwhelmed me began to recede, replaced by a sense of control and predictability.

This realization didn’t just improve my results; it fundamentally changed how I interacted with the technology. I became more patient, more strategic, navigating its transient memory limitations strategically. The tool wasn’t just a source of answers but a partner in exploration, one that required careful handling to yield the best outcomes. This wasn’t a battle to be won; it was a dance, a collaboration between human creativity and machine computation, and if I wanted to dance, I had to adjust my expectations. Accept that LLMs aren’t magic wands; they’re tools, sophisticated yet fundamentally limited by their programming and current technology.

Reflection and (pseudo)Philosophical Insight

In my journey with LLMs, I’ve come to see their imperfections not as failures but as echoes of human nature. We criticize these models for ‘hallucinating’ or for not consistently delivering the perfect answer. Yet, isn’t that remarkably human? We too make mistakes, talk nonsense, and even our brightest moments can spring from misunderstandings or missteps. This, I believe, is the most human aspect of LLMs — their fallibility.

This realization has been liberating. It taught me that ‘Not Good Enough’ isn’t a final verdict but a starting point. These imperfections provoke thought, inspire adjustments, and sometimes lead to brilliant insights that a flawless machine might never provoke. Every ‘error’ forced me to engage more deeply, to blend my knowledge and intuition with the machine’s outputs. This synergy, this dance between human and machine, often leads to outcomes far richer than what either could achieve alone.

The beauty of working with LLMs lies in this partnership. It’s about using the 80% they provide as a foundation, then adding our own 20% — the personal touch, the human flair that transforms ‘Not Good Enough’ into great. In a world obsessed with perfection, embracing imperfection as a catalyst for creativity is a profound shift in perspective. It’s a recognition that sometimes, the path to innovation is paved with mistakes and that our responses to these challenges can turn the ordinary into the extraordinary.

Conclusion

Navigating the world of Large Language Models has been a journey of discovery, frustration, and ultimately, acceptance. What began as a quest for perfection has transformed into a profound lesson in humility and collaboration. Each phase of this journey has taught me invaluable lessons about the nature of this technology — and about myself. These digital entities, as sophisticated as they are, have shown me that perfection isn’t always necessary, or even desirable. Instead, it’s about leveraging their capabilities to augment our human creativity and insight. This experience is a narrative about technology as much as it is about human adaptation and innovation.

The relationship with LLMs is less about commanding a foolproof tool and more about engaging with a dynamic partner that challenges, confounds, and occasionally enlightens. It is the 20% — our creativity, context, and humanity — that we add which transforms the output from a machine into something uniquely insightful and profoundly human. This synergy is where true value lies, transforming ‘Not Good Enough’ into ‘Just Right.’ LLMs compel us to articulate our thoughts more clearly, challenge our assumptions, and expand our boundaries. This interaction is not merely about technology assisting us; it’s about technology inspiring us to excel. In this partnership, where machines handle the heavy lifting and humans add the finishing touches, we can achieve remarkable results.

Let us embrace ‘Not Good Enough’ not as an endpoint, but as a beginning — a catalyst for innovation and discovery that defines our shared future with AI.

--

--

Pol Martí

ADD brain behind the screen! Expect a kaleidoscope of thoughts on tech and leadership—always insightful, often unfinished.