Today's enerative AI models can create everything from images to computer applications, but the quality of their output heavily depends on the prompts provided by human users.

Researchers at Carnegie Mellon University have introduced a new approach to teach everyday users how to create these prompts and improve their interactions with generative AI models.

This method, called Requirement-Oriented Prompt Engineering (ROPE), shifts the focus of prompt writing from clever tricks and templates to clearly articulating what the AI should do. As large language models (LLMs) improve, the importance of coding skills may decline, while expertise in prompt engineering is likely to rise.

"You need to be able to precisely tell the model what you want. You can’t expect it to guess all your customized needs," said Christina Ma, a doctoral student at the Human-Computer Interaction Institute (HCII). "We need to train humans to master prompt engineering skills. Most people still struggle to accurately tell AI what they want. ROPE can help them do that."

Prompt engineering refers to the precise instructions (prompts) users provide to generative AI models to generate desired outputs. The stronger a user's prompt engineering skills, the more likely the AI model will produce the desired results.

In an article published in the ACM Transactions on Computer-Human Interaction titled "What Should We Design in Prompts? Training Humans in Requirement-Driven LLM Use," the researchers describe their ROPE paradigm and the training modules they created to teach and evaluate this approach.

ROPE is a strategy for human-LLM collaboration that emphasizes clearly defining the requirements of an LLM prompt, enabling humans to take control of goals and achieve them autonomously. The approach highlights the importance of formulating precise and complete requirements to achieve better results, especially for complex, customized tasks.

To test ROPE, the researchers asked 30 participants to write prompts for an AI model to complete two tasks as a pre-test: creating a tic-tac-toe game and designing a tool to help people create content outlines. Subsequently, half of the participants received ROPE training, while the other half watched YouTube tutorials on prompt engineering. Afterward, both groups wrote prompts for different games and chatbots as a post-test.

The researchers compared the results and found that participants trained with ROPE outperformed those who watched YouTube tutorials. ROPE-trained participants improved their scores by 20% from pre-test to post-test, while those without ROPE training improved by only 1%.

"We not only proposed a new framework for teaching prompt engineering but also created a training tool to evaluate participants' performance and the effectiveness of this paradigm," said Ken Koedinger, a professor at HCII. "The ROPE method is not only effective, but the training module also supports this."

As traditional programming evolves into natural language programming, generative AI models have transformed the content of introductory programming and software engineering courses. Engineers no longer need to write software but can instead write prompts to guide AI in developing software.

This paradigm shift creates new opportunities for students, enabling them to tackle more complex development tasks early in their learning and drive advancements in the field.

The researchers did not design ROPE solely for software engineers. As humans increasingly integrate AI into daily life, communicating with machines is clearly becoming a critical aspect of digital literacy. With the ability to write successful prompts and an AI model capable of executing them, even individuals without programming or software engineering backgrounds can create applications that benefit them.

Ma said, "We hope to enable more end-users from the general public to use LLMs to build chatbots and applications. If you have an idea and know how to articulate your requirements, you can write a prompt to realize that idea."

The researchers have open-sourced their training tools and materials, aiming to make it easier for non-experts to master prompt engineering techniques.