Diogo Pedrosa

Automating Fallacy Detection: advancements With Open-Source Large Language Models

Argumentation is a fundamental element of human discourse, crucial in shaping societies and influencing decision-making processes. However, logical fallacies can compromise the integrity of arguments, leading to the spread of misinformation and harmful ideologies. This study explores the application of machine learning, particularly Large Language Models (LLMs), in detecting fallacies within natural language text. Using LLaMA-3.1 70B, an open-source LLM, we developed a methodology that leverages prompt engineering to achieve high accuracy in fallacy detection. Extensive evaluation with the LOGIC dataset resulted in an F1 score of 92.9%, significantly outperforming previous models designed for fallacy detection. This research highlights the potential of advanced LLMs to enhance critical reasoning through real-time fallacy detection, contributing to social networking moderation, misinformation mitigation, and political discourse analysis

The Future of Destruction: Analyzing the Potential Impact of Entropy on Technological Evolution in AGI, Autonomous Drones, and Applied Robotics

This paper discusses the rapid advancements in Artificial General Intelligence (AGI), autonomous drones, and applied robotics, highlighting their transformative potential across various sectors. It raises ethical concerns, particularly in warfare, using entropy as a metaphor for the potential chaos these technologies could introduce. The paper examines risks associated with AGI in conflict scenarios, analyzes the implications of autonomous drones on warfare tactics and ethical dilemmas, and explores applied robotics in military operations. It advocates for a cautious approach, proposing a framework grounded in science and ethics to guide the responsible development and deployment of these technologies while ensuring global stability.

Towards Universal Applied Supervised Machine Learning: A Multi-Agent Framework For Systematic Pipeline Executions

This paper introduces MADS (Multi-Agents for Data Science), an open-source framework for streamlined execution of supervised machine learning pipelines using multi-agent systems (MAS). It emphasizes the integration of various agent types, particularly large language model (LLM) agents currently employed for automating machine learning tasks. Future iterations aim to incorporate reinforcement learning (RL) agents to further enhance performance and adaptability. MADS aims to improve efficiency, scalability, and adaptability in supervised learning across domains by reducing manual effort and enhancing model performance through automation. This framework represents a significant advancement in machine learning, fostering broader adoption and collaborative development.