Six months have passed after the release of ChatGPT, and its influence has spread to every sector of the economy and workplace. Thus, interest in the area of “prompting artificial intelligence (AI),” which entails adjusting inputs to produce the most efficient outcomes, has recently increased. Even now, some businesses are starting to employ AI “prompt engineers," who are experts at efficiently interacting with generative AIs (GenAI), which generate a wide range of data. Although the idea behind ChatGPT and prompting may appear simple, employing them effectively may seem far from easy. In order to provide readers with insight into how to obtain the most accurate results, The Granite Tower (GT) consulted four Korea University (KU) experts who have dedicated part of their academic careers to the technology enabling ChatGPT, offering readers a glimpse into how to achieve the most accurate results. As the release of ChatGPT has left everyone in suspense, these insights may help KU students unlock the full potential of this powerful tool.

The followings are the questions asked:

What are the difficulties you have experienced while using ChatGPT? What are your tips for prompting ChatGPT wisely?

Professor Nam Ho Seung (Department of English Language and Literature)

I am the NAMZ’s lab leader, and our lab researches speech processing, which includes speech recognition, speech synthesis, text analysis, conversation processing, and pronunciation evaluation. As we develop GenAIs in NAMZ, I have not experienced any problems or difficulties while using them. So, to provide some tips for prompting, I would strongly advise students to learn how to code. As far as I can tell, there are disparities between students who have knowledge in coding and those who do not. I have also noticed there were apparent differences in prompting ChatGPTs between students who have learned coding and those who have not. I think studying the concept of coding enables students to conceptualize their ideas and give better input sentences to GenAI models since they have acquired deductive reasoning. In addition, I would advise students to consider what the AI would want as input, which might be slightly different from typical human conversation.

Oh Jong Min (’16, Computer Science and Engineering)

I am a member of KWeb, which created KLUE (KU’s course review website), the Mensa Korea developer society, and the Google Developer Student Club (KUGODS). Thus, I communicate with various developers regarding GenAIs. I experienced difficulties when I received answers that were inconsistent or unclear about certain facts. To prevent this, accuracy and sufficiency with natural, structured sentences would be important in prompting. Simply put, this means that users should give information to ChatGPT about what they already know and what they desire to know. It is important to maintain a consistent writing style. For example, if a casual expression is used at the beginning of the question, and a formal tone is used at the end, the flow of the text will be unnatural, which can cause confusion. Additionally, verifying the authenticity of the answer is an essential point. Since ChatGPT is not competent in every situation, even if you have considered all the above cases and prompted, you should still be cautious and double-check the answer that is given.

Kim Yoon Jung (’21, Linguistics)

I am a member of ChatGPTers, an online community that shares information and insights about GenAI, and I have given presentations on it several times. To share my experience, I once asked ChatGPT to provide the bibliography of a certain argument. As a language model using a generative pre-trained transformer (GPT), it is tuned to give me several links as an output; however, none of the links were accurate. Thus, I believe demanding ChatGPT to provide insights would be an inaccurate prompting, considering its original purpose. We should acknowledge that GenAI is not suitable for creating new information. Considering this, I believe we should prompt by giving structures and utilize it in various situations, simply called nogada. Instead of prompting with long, unstructured text, I prompt ChatGPT by structuring the input into two parts: system and point. In the system part, I provide ChatGPT with what I want as a result, and for the point part, I emphasize some important keywords or perspectives regarding the system part.

An Jun Hyun (’16, Materials Science and Engineering)

I am the Vice President of the KU Computer Club (KUCC), arranging computer-related sessions, studies, and Hackathons. Since its release, I have used ChatGPT for a variety of things, but primarily for making and uploading YouTube shorts. To begin with, it was difficult to come up with the exact response I had in mind. The output was different from what I had anticipated because it offered information through texts. Another difficulty was that it was hard to backtrack once the conversation got awkward. Since ChatGPT remembers the conversation history in a dialog format, it might be better to start a new conversation in those situations. Therefore, I recommend providing clear templates, as if assigning tasks to a very intelligent new employee who is not familiar with working. For example, if you want to utilize ChatGPT to study thermodynamics as a student, you can provide a clear role and template: “I’m a university student, and I’m going to study thermodynamics. Please answer my questions as if you were my professor,” or “My report is in <format>. Please use <content> to write a report.” Furthermore, asking in English can be more effective in prompting as well. In fact, I refined my writing using ChatGPT!