El potencial de los modelos grandes de lenguaje para mejorar el aprendizaje de probabilidad: un estudio sobre ChatGPT 3.5 y estudiantes de primer año de ingeniería informática.

This study evaluates the effectiveness of ChatGPT in solving common probability problems in computer engineering exams. For this purpose, a set of 23 exercises previously used in exams given to first-year computer engineering students at the Universidad Rey Juan Carlos, Madrid, Spain, were selected. The answers of the model were evaluated by five statistics professors. The results show that ChatGPT outperforms the average student in writing, organisation and logical reasoning and performs consistently well in Spanish and English. However, it has difficulties with basic numerical operations, although returning the solution via an R script helps it to overcome these limitations. Despite some deficiencies in probability reasoning, the model demonstrates high potential as a learning assistant due to its ability to provide high quality explanations and illustrate solutions in different programming languages.

UDÍAS Ángel;  ALONSO-AYUSO Antonio;  SANCHEZ MARTIN Jose Ignacio;  HERNÁNDEZ Sonia;  CASTELLANOS María Eugenia;  MONTES Raquel;  LÓPEZ CANO Emilio; 

2025-03-31

Dykinson

JRC133262

978-84-1070-836-5 (online),   

1560-4292 (online),