In the last three years, Large Language Models (LLMs) have transformed multiple fields, including higher education. While their understanding of human instructions continues to evolve rapidly, advancements in this field are driven by cutting-edge commercial LLMs developed by tech companies with varying privacy policies. To address the growing demand for privacy in the use of LLMs, local services relying on smaller language models have emerged. Working with smaller models presents unique challenges compared to their larger counterparts. Notably, smaller models are particularly sensitive to the quality and structure of user prompts. This underscores the importance of crafting effective prompts.
Prompt engineering has become an essential skill for the future. Numerous studies have explored advanced prompting techniques. These studies have identified the key skills required for effectively using LLM-based applications. Among these skills are understanding prompt structure, prompt literacy, mastering various prompting methods, and employing Critical Online Reasoning with an LLM. These elements revolve around two central ideas: knowing what to expect from an LLM and understanding how to achieve the desired outcomes.
Several prompting methods have proven particularly effective. These include few-shot prompting (providing the LLM with examples of similar solved tasks to guide its responses), chain-of-thought prompting (encouraging the LLM to reason step by step in a "think-aloud" manner), and role modeling (asking the LLM to assume specific roles or personas to approach tasks from a particular perspective). In this workshop, we will introduce participants to these prompting techniques and provide practical guidance on their application in academia and beyond. We will also highlight common pitfalls and mistakes to avoid when employing these methods, ensuring participants can leverage LLMs effectively and responsibly.