¿Quién quiere ser millonario?
Loading...
Official URL
Full text at PDC
Publication date
2025
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
Este proyecto presenta el desarrollo de una aplicación web interactiva que simula el concurso “¿Quién Quiere Ser Millonario?”, explorando la integración de large language models (LLMs) e inteligencia artificial (IA). El objetivo central fue diseñar una plataforma atractiva y funcional, abordando desafíos clave como la unicidad de las preguntas y la validación precisa de las respuestas mediante técnicas de procesamiento del lenguaje natural (NLP) y algoritmos de comparación. La aplicación implementa la mecánica del concurso, permitiendo a los usuarios responder preguntas de dificultad creciente.
Se integraron los modelos Gemini (para establecer la verdad fundamental de las respuestas) y DeepSeek, OpenAI y Llama (para el comodín de la llamada y como concursantes IA en el modo Máquina vs Máquina). Se describe la construcción y optimización de los prompts para DeepSeek, logrando una reducción significativa de la latencia de respuesta. De manera similar, se aborda la integración y el uso de prompts para OpenAI y Llama en el modo de enfrentamiento entre IAs. El trabajo presenta los resultados iniciales sin optimizar para DeepSeek y el impacto de las mejoras implementadas en su rendimiento. La aplicación desarrollada sirve como un entorno de prueba para la integración y comparación de diversas IAs en concursos de preguntas, ilustrando los retos y soluciones encontrados en el proceso de implementación y optimización de cada modelo.
This project presents the development of an interactive web application that simulates the “Who Wants to Be a Millionaire?” quiz show, exploring the integration of large language models (LLMs) and artificial intelligence (AI). The central objective was to design an attractive and functional platform, addressing key challenges such as the uniqueness of the questions and the precise validation of the answers through natural language processing (NLP) techniques and comparison algorithms. The application implements the mechanics of the quiz show, allowing users to answer questions of increasing difficulty. The Gemini models (to establish the fundamental truth of the answers) and DeepSeek, OpenAI, and Llama (for the phone-a-friend lifeline and as AI contestants in the Machine vs. Machine mode) were integrated. The construction and optimization of the prompts for DeepSeek, achieving a significant reduction in response latency, are described. Similarly, the integration and use of prompts for OpenAI and Llama in the AI versus AI confrontation mode are addressed. The work presents the initial, unoptimized results for DeepSeek and the impact of the improvements implemented on its performance. The developed application serves as a testing environment for the integration and comparison of various AIs in question-answering contests, illustrating the challenges and solutions encountered in the process of implementing and optimizing each model.
This project presents the development of an interactive web application that simulates the “Who Wants to Be a Millionaire?” quiz show, exploring the integration of large language models (LLMs) and artificial intelligence (AI). The central objective was to design an attractive and functional platform, addressing key challenges such as the uniqueness of the questions and the precise validation of the answers through natural language processing (NLP) techniques and comparison algorithms. The application implements the mechanics of the quiz show, allowing users to answer questions of increasing difficulty. The Gemini models (to establish the fundamental truth of the answers) and DeepSeek, OpenAI, and Llama (for the phone-a-friend lifeline and as AI contestants in the Machine vs. Machine mode) were integrated. The construction and optimization of the prompts for DeepSeek, achieving a significant reduction in response latency, are described. Similarly, the integration and use of prompts for OpenAI and Llama in the AI versus AI confrontation mode are addressed. The work presents the initial, unoptimized results for DeepSeek and the impact of the improvements implemented on its performance. The developed application serves as a testing environment for the integration and comparison of various AIs in question-answering contests, illustrating the challenges and solutions encountered in the process of implementing and optimizing each model.
Description
Trabajo de Fin de Grado en Ingeniería Informática UCM, Departamento de de Sistemas Informáticos y Computación, Curso 2024/2025












