Today's generation artificial intelligence models can create everything from images to computer applications, but the quality of their output depends heavily on prompts provided by human users.
Researchers at Carnegie Mellon University have proposed a new approach to teach everyday users how to create these prompts and improve their interaction with generative artificial intelligence models.
This method, known as requirement-oriented prompt engineering (rope), shifts the focus of quick writing from smart tricks and templates, clearly stating what AI should do. As large-scale language models (LLMs) improve, the importance of coding skills may decline, but rapid engineering expertise may increase.
“You need to be able to convey exactly what you want to the model. You can't expect to guess every customized need,” said PhD Christina MA. A student at the Human-Computer Interaction Institute (HCII). “We need to train humans with rapid engineering skills. Most people struggle to tell AI exactly what they want. The ropes help with that.”
Prompt engineering refers to the exact instructions (prompts) where the user provides a generated AI model to generate the desired output. The faster the user is engineered, the more likely the AI model will generate what the user intended.
In “What should I design at the prompt? Training humans with requirements-driven LLM use,” which has recently been accepted by the Association of Computational Mechanical Associations on Computer-Human Interactions, researchers will discuss the rope paradigm and the training modules they have created to teach and evaluate methods. Rope is a human partnering strategy in which humans maintain control of institutions and goals by specifying the requirements of the LLM prompt. This paradigm focuses on the importance of creating accurate and complete requirements to achieve better results, especially for complex and customized tasks.
To test the rope, the researchers ask 30 people to write a prompt for the AI model, completing two separate tasks as pretests. Create tic tuck toe games and design tools that help people develop content overviews. Half of the participants were then trained through the ropes, while the rest watched YouTube tutorials on prompt engineering. The group then wrote another chatbot prompt as a different game and a posttest.
When researchers compared the results of the exercise, they found that participants who received rope training outweighed those who watched the YouTube tutorial. Scores from pre-test to post-test increased by 20% for those who received rope training and only 1% for those who did not.
“We not only proposed a new framework for teaching rapid engineering, but also created a training tool that assesses how well participants work and how well the paradigm works,” says Ken Koedinger, a university professor at HCII. “It's not just our opinion that the rope works. The training module backs up that up.”
Generated AI models have already changed the content of their implementation programming and software engineering courses as traditional programming evolves into natural language programming. Instead of creating software, engineers can create prompts to instruct AI to develop software. This paradigm shift can create new opportunities for students, allowing students to tackle more complex development tasks early in their research and advance their fields.
Researchers did not design the ropes just for software engineers. As humans continue to integrate AI into their daily lives, clear communication with machines becomes an important aspect of digital literacy. Armed with knowing how to create AI models up to successful prompts and tasks, people with no background in coding or software engineering can create applications that benefit them.
“We want to enable more end users to the public to build chatbots and apps using LLMS,” Ma said. “If you have an idea and know how to communicate your requirements, you can write a prompt to create that idea.”
Researchers have opened saw training tools and materials with the aim of making rapid engineering more accessible.
/Public release. This material of the Organization of Origin/Author is a point-in-time nature and may be edited for clarity, style and length. Mirage.news does not take any institutional position or aspect, and all views, positions and conclusions expressed here are the views of the authors alone.
