The genesis of Intelligent Prompts

 Writing prompts is like writing good code- they should be precise, efficient and and able to handle edge cases well…

With AI at our disposals, activities of day-to-day have become incredibly convenient. So much so, the GPT tab pn our PCs practically never closes! Being a second year student enrolled in a Computer Science Engineering Programme, I witness this phenomena nearly every day. With my pals, rushing through their assignments and DSA practicals by hard-core copy-pasting codes from their very own, aforementioned saviour, ChatGPT.

Which, of course, on a moral ground is incorrect and shouldn’t be this way in the first place. But where my problem lies, is actually on a different and more important front- the practical aspect. Getting your labs done and dusted and your submissions aligned is important, but really? You call yourselves engineers (in the making), without having to put yourselves through the ‘pain’ of thinking and actually engineering? Self-sabotage much? I could write an entire article on this, but let’s not deviate from the actual topic at hand.

Sure, AI is overused- misused even! But who says, you can’t smartly leverage it to polish your products and decorate your creations, all while real intelligence doing the core work? Certainly not me…

The more I survive this ordeal named college, the more I realise just how incredible of an innovation GenAI is. I had to recently complete an assignment of a course called Principles of Economics and Management, asking us to list down the possibilities of India becoming the third largest economy by 2030. I was able to complete it with a one-day, but extensive and enough research on the net. It was a tight deadline and I was struggling to find an opening line to my abstract / intro section. I am big on Introductions, I need them to be flawless. There went my prompt-

generate a thought provoking and well written opening line for the abstract of a project which talks about the predictions of India becoming the third largest economy, despite the recent global economic setbacks.

and here was the output-

In an era marked by unprecedented global economic upheavals — ranging from the lingering impacts of the COVID-19 pandemic to geopolitical tensions and climate-related disruptions — India’s economic trajectory presents a compelling paradox.

Which very appropriately went with my theme, of India’s growth being positive and upwards, despite the global background of turmoil and disruption- I got what I wanted! The language being precise, yet flashy, just enough to develop interest in the reader and the tone being almost a perfect balance of smugness and nonchalance.

How was I able to achieve this? With what I call, a ‘good prompt’! And you need not be a master Prompt Engineer to create one. So, how do you go about this?
To get what you want, you must know what you want. In order to make the most out of Large Language Models (LLMs) like GPT, one must be extremely intentional with every word of their query. For reference, these are the recommendations on the official GPT website;

Notice the repeated use of words like ‘specify’, ‘indicate’, ‘include’, etc.? The point really being made here is, be as precise about what you want, as you can possibly be. Do not shy away from adding details and keywords. Tuning your request so finely in the first go, that the corresponding output is just exactly what you will use, is always a good idea.

You’re saving time, efforts and possibly avoiding a breach of quality! The other day, I was scratching my head, trying to understand the Kadane’s Algorithm, which was used to solve one of the problems I came across. After a read or two of codes that used it, I went straight to ChatGPT to get a first-hand understanding of what the Algorithm was all about. With a simple but efficient query, I was able to grasp the concept quickly and proceed to solve my problem!

Here are some points you could keep in mind while crafting a prompt-

• Less Isn’t More- As already stated, a longer prompt would in no way reduce the efficiency of your output. It would rather do the opposite in most cases. Go for a descriptive idea of how you want what you want- in what length, of what style and for what purpose.
• Explicitly mention what you don’t want- If you definitely do not want something in the results- let your model know! If you expect something to appear because it seems generic or obvious but want to avoid it, trust your instinct and include the same in your prompt. You can go as far as mentioning it in caps. No one is stopping you. Anything for saving a repetitive saga of queries.
• Do not confuse the model- It is a machine at the end of the day and real intelligence would always be superior. Ensure you’re using yours. It is easy for the models to produce incorrect outputs if confused or cluttered. This can be avoided by keeping queries simple, and follow-up questions being reasonable and within context.

To think that an article is being written for crafting good prompts… Seems ridiculous at first glance, no? But if AI has indeed become such an indispensable part of our lives (and for good reasons, at that), then we might as well learn doing it the right way! :)