Generative AI tools are being used for a multitude of applications. For the use of the GenAI tools input from humans is always needed in the form of a prompt. The first GenAI tool I experimented with was ChatGPT. Everyone around me was already using it and raving about how helpful and convenient its use can be but did tell me to be careful of mistakes in the output it gave. So, when I started using ChatGPT, I assumed the hardest part would be evaluating the outputs. Instead, I quickly realized that the real challenge lied in carefully prompting my input. I had to make sure to ask the tool the right questions to get the desired output. A vague request often resulted in generic or mismatched outputs, while carefully crafted prompts with the right wording and structure could produce more precise results. Sometimes it was especially frustrating when it seemed like the GenAI tools kept misinterpreting what I was trying to ask of it.
I expected an instant efficiency boost by using ChatGPT but bad and inaccurate outputs seemed to add extra work because I had to go back and forth with ChatGPT by adding to the original prompt to get my desired output or tweaking the given output to match my intent. It became clear prompting was a new skill set I had to master to effectively use GenAI tools. The most important points to write a successful prompt are to be specific and complete, since such tools will take what you say very literal.
GenAI tools have gotten significantly better and overall less precisely worded and structured prompts now get a decent output. However, to have GenAI tools produce their best output and make their use as efficient as possible, it is still of importance to perfect the skill set of prompting.
Such a relatable post! I’ve struggled with this issue a lot myself. During one of my bachelor’s classes, we had an assignment where the essay had to be written entirely by AI (including references). At first, I thought that it’s going to be so easy. But I was completely wrong. I gave ChatGPT the initial prompt, and while it produced an essay, the quality was far from what I expected. I kept refining my instructions, giving more and more specific prompts to improve the output, but it often felt like it just didn’t understand what I meant. It also gave me completely made-up references when I asked it to cite sources. Part of the assignment was also to submit not only the final essay, but also the prompts we used and ChatGPT’s responses. By the end, my prompt history was over 60 pages long (!). Mostly because ChatGPT either didn’t follow the instructions or misinterpreted what I meant. I completely agree with you that successful prompts need to be specific and detailed. But even then, the answers don’t always fully meet my expectations.