In this age of AI, the world is going through a revolution. While large language models (LLMs) like ChatGPT, Gemini, and Jurassic-1 Jumbo garner the most attention, one thing slips everybody’s mind: the Prompt. Although the headlines focus solely on the impressive feats of LLMs, people often forget that the true magic of LLMs lies not in the software themselves, but the prompt a person gives is what makes the difference. In layman’s terms, prompts are instructions given to the LLM that help it understand the task. For example, a prompt can be as simple as, “ Give me a summary of this article,” or you can type something more detailed like, “Give me a summary of this article and highlight the part where the article talks about LLM,” you will get the summary while highlighting the part that is most relevant to you.
According to a 2023 study conducted by Stanford University, large language models performed at least 20% better when given specific and detailed prompts than generic prompts. American Computer Scientist Lex Fridman explains this beautifully. He says, “The art of prompting AI is akin to sculpting; with each well-formed prompt, we shape the AI’s response to align with our intentions.”
MIT conducted research in 2022 to test the effectiveness of detailed prompts in image generation using an LLM. The MIT team provided a detailed description of the picture they desired, and they could generate high-quality and high-fidelity images through LLMs. This opens the door for the usage of LLMs in product design, architecture, game development, and even scientific and astronomical illustrations. All that is needed is a detailed prompt.
To champion the art of prompting, one must be clear and precise and keep ambiguity to a minimum. British computer scientist Stuart Russel says, “Eliminate any ambiguous terms or phrases in your prompts. Ambiguity can lead to unpredictable or irrelevant responses.” According to him, ambiguous prompts can confuse the LLM, which can lead to unpredictable and, more often than not, low-fidelity outcomes.
Another tip to master the art of prompts is to be as specific as possible. Don’t leave anything unspecified. Lex Fridman Says, “If you need the response in a particular format, specify it in the prompt. Whether it’s a list, an essay, or a summary, clear instructions help the model align with your expectations.” This will result in accurate and more consistent outcomes. One thing that everyone should do is to review and adjust your prompt. When you get an outcome after giving a prompt to the LLM, review the prompt and then evaluate the outcome to determine where your prompt went wrong and then adjust the prompt accordingly. This method will help you understand the art of prompting even better.
