University Park, Pennsylvania – Using services like ChatGPT or Microsoft Copilot can sometimes look magical. But like other complex systems, there is always room for improvement and optimization, according to Rui Zhang, assistant professor of computer science and engineering at Pennsylvania Electrical and Computer Science.
Zhang and his research group recently introduced a new approach to processing high-resolution images and wrote three papers that automatically encourage better responses from AI systems. The paper currently available online will be published in Vienna, Austria from July 27th to August 1st at the 63rd Annual General Meeting of the 63rd Association of Computational Linguistics. International Conference on Computer Vision in 2025, October 19-23, in Honolulu, Hawaii. The 13th International Conference on Learning Expression was held in Singapore from April 24th to 28th.
In the next Q&A, Zhang discussed his group's work, how it can improve the efficiency and usefulness of AI, as well as some strategies that individuals can employ to gain more value from their personal use.
Q: What is quick engineering? Is there anything specific that readers can do to write better prompts for AI systems?
Chan: Prompt Engineering is the process of designing effective inputs that guide AI systems like CHATGPT to guide AI systems to create better responses. These systems are sensitive to asking questions, so well-written prompts can greatly improve the output of the system. For example, instead of asking “Summary this article,” you might say “Summary this article in three bullet points for high school students.” The extra context helps AI to coordinate responses. For everyday users, the main strategy is to be clear, concrete and goal-oriented. Don't be afraid to try multiple prompt versions to improve your results.
Q: What are the benefits of automating and optimizing rapid generation?
Chan: While good prompt engineering can greatly improve AI performance, writing the best prompt requires time, experimentation and expertise on the subject matter contained in the prompt. Our research has developed a method called Great that allows AI systems to automatically generate and refine prompts using gradient-based optimization, using gradient-based optimizations.
We also developed GreaterPrompt, a user-friendly, open source toolkit built on a bigger way. This allows the model to automatically generate and refine prompts for a wide range of tasks. Automating this process means that AI can adapt to new tasks that reduce human input, improve accuracy, save time and reduce costs. This is especially valuable for users who lack subject time and expertise in order to come up with better prompts. By providing an open source toolkit that anyone can download, modify, or share, we effectively distribute access to our work for all interested users.
Q: How did you measure the effectiveness of Greater? Are there any actual tools that can be improved through implementation?
Chan: we A greater evaluation was given for inference and mathematical problem-solving tasks in various languages, including answering complex questions, solving logic puzzles, and performing mathematical calculations. The results showed a significant improvement in performance compared to standard prompts. Especially in small language models that normally wrestrate these tasks, as they are limited by special parameters of specific tasks and questions. In some cases, these optimized small models were comparable to models of much larger quality. Real-world applications that can benefit include AI-powered tutors, writing assistants, customer support agents, and tools that need to quickly adapt to a variety of users and topics without manual reprogramming.
