Leveraging LLMs for K-12 Education Assessment: An Open-Ended Question Answering Use Case

In a context of great explosion of large language models (LLMs) applications, education is destined to be one of the areas where these new technologies can generate the greatest impact. Novel LLM-based solutions could enhance the scalability of educational systems, tackling with the scarcity of teachers and generating more valuable time to dedicate on their students learnings. In this article we focus on an application where LLMs can have a significant impact, such as automated assessment. For this purpose, we built a dataset in Spanish with 1224 responses to open-ended questions and their respective reference ratings. Based on these data, we compared the performance of state-of-the-art LLMs and different kind of prompts, also in Spanish. The results show significant differences between models, where the most advanced ones are capable of solving the task with an accuracy over 90%. Furthermore, we also evidence the relevance of the prompt details to get the most out of the LLMs in the specified task.