My fascination with AI began, like many, with a sense of wonder. I saw the power of large language models and thought, “This is it! I can automate so much!” My first big idea was a “gem” to get daily current affairs for my Kerala PSC students. I pictured myself effortlessly feeding them the latest, most relevant news.
My initial prompt was simple, almost naive: “Get me daily current affairs.”
The response? A weather report.
I stared at the screen, utterly baffled. Here was this incredibly powerful AI, trained on the entire internet, yet it was giving me the day’s temperature when I asked for vital news. It was like I was speaking a different language.
The Alien in the Box: Decoding How AI “Thinks”
That’s when I had my first big realization: AI is an alien that knows everything but knows nothing about me. It doesn’t “understand” current events in the human sense. Instead, it operates on incredibly complex statistical probabilities. It’s a master at pattern recognition, always predicting the next most likely word in a sequence based on the vast sea of data it’s consumed. My vague prompt, “Get me daily current affairs,” was just asking it to pick any statistically common “current event,” and a weather report fit that broad pattern perfectly. I wasn’t giving it the precise signals it needed.
This led me to my core discovery: the “phrase approach.” I realized I had to learn to speak its language, to use specific words and structures that would guide its pattern-matching process exactly where I wanted it to go.
My Breakthrough: The Power of Specific Phrases
My journey became one of experimenting with these linguistic signals:
- Finding the Focus (Specificity): My first attempt at refinement was adding “give me relevant to Kerala PSC.” Suddenly, the weather reports disappeared, replaced by exam dates and upcoming events. This was better! I learned to use phrases like “specifically, discuss…” or “focusing on…” and “detail…” to narrow the AI’s vast knowledge to my exact requirements.
- Shaping the Output (Format & Length): I needed specific formats for my students. I started explicitly stating: “in bullet points,” “as a numbered list,” “limit to 5 sentences.” These phrases weren’t just stylistic; they were instructions for the AI’s generative architecture, telling it what kind of pattern to output.
- Adopting a Persona (Role Assignment): For my explanations, I experimented with phrases like, “You are an experienced prompt engineer. Explain this…” This was fascinating! The AI’s tone and depth of explanation shifted, almost as if it had truly donned the persona, accessing patterns of language associated with that role.
- Adding Context (Background Information): The AI didn’t know I was a teacher for PSC students. Providing context became crucial. My internal dialogue shifted to thinking, “What does the AI not know that I do?” This led to adding phrases like “given the following context” or explicitly stating the “target audience is…” This grounded the AI’s responses, preventing wild, irrelevant tangents.
- Showing, Not Just Telling (Few-Shot Prompting): This was the real game-changer. When the current affairs were still not quite right, I took a leap. I provided the AI with actual recent Kerala PSC question papers. This is called few-shot prompting. I essentially said: “Here are two examples of Kerala PSC current affairs questions and answers: [Example 1], [Example 2]. Now, generate similar current affairs for today…” By showing it exactly what I wanted, I was giving its pattern-recognition engine a perfect template. This was the breakthrough. Suddenly, within 10 minutes, I could prepare incredibly relevant current affairs!
- Navigating Nuance (Cultural & Linguistic Specificity): My next challenge was translating these accurate current affairs into Malayalam. The initial translations were often “weird language,” grammatically correct but culturally awkward. This revealed another layer of AI’s pattern recognition: it needed cultural context. My new phrase became: “Please search Malayalam articles related to it and then use that word and styles.” This instructed the AI to draw on patterns from native Malayalam content, helping it bridge the cultural gap.
- Guiding Through Steps (Chain-of-Thought): For more complex tasks, I learned to break them down with phrases like: “First, identify X. Second, outline Y. Finally, state Z.” Or simply, “Think step by step.” This chain-of-thought prompting wasn’t just for me; it forced the AI to follow a logical progression, improving its accuracy on intricate tasks by guiding its predictive journey.
The Unwavering Need for Accuracy: My Fact-Checking Imperative
Despite these advances, one critical issue remained: accuracy. My students rely on precise data. Sometimes, the AI would “hallucinate,” confidently presenting plausible but utterly false information. I’d get things like “Operation Sindoor” instead of “Operation Sindhu,” an actual Iranian evacuation.
I tried prompting for “fact-checking,” but the AI, left to its own devices, couldn’t reliably verify its own internal patterns against external reality. It taught me the most crucial lesson: AI-generated content, especially for critical information, requires human oversight and external verification.
I even asked the AI for help with this, and it suggested Perplexity AI. So now, my workflow includes a swift check with Perplexity, followed by my own manual verification. This adds 20-30 minutes to my daily routine, but it’s a non-negotiable investment in accuracy and reliability.
My Evolving Dialogue with AI
My “gem” has transformed from a frustrating weather reporter into an invaluable assistant. This journey has been a masterclass in learning the subtle language of AI. It’s about recognizing that AI doesn’t understand in the human sense, but excels at pattern-matching. My role, as a prompt engineer, is to provide the clearest, most specific patterns possible through my “phrase approach.”
It’s an ongoing, iterative dialogue. I refine, rephrase, and add context based on the AI’s responses. This constant back-and-forth ensures that I’m not just typing at a machine, but effectively communicating with an incredibly powerful, albeit alien, intelligence. My experience has shown me that the true potential of AI is unlocked not just by its capabilities, but by our ability to learn its language.