Procesamiento de lenguaje natural mediante el uso eficiente de Prompt Engineering en Large Language Models

Abstract

Abstract: Prompt engineering involves the creation of patterns or guidelines that guide generative language models (LLMs) in the execution of a specific natural language processing (NLP) task, such as classification, text generation or sentiment analysis. LLMs present several use cases, three of them are addressed in the present work, where we evaluate their application and assess their performance using prompt engineering techniques. The application of these techniques in LLMs has evidenced their usefulness to perform different NLP tasks, with zero or little training. To test the prompt engineering techniques, three NLP tasks were defined, and different language models and datasets were selected; to score the model response, different metrics were used to determine their performance. By testing the language models with more complex tasks, the usefulness of prompt engineering techniques to drive the model to generate more accurate responses was verified. After experimentation, and based on the results obtained, a practical guide for the efficient construction of prompts was elaborated.