Despite significant advancements in Artificial Intelligence (AI) and Large Language Models (LLMs), detecting and mitigating bias remains a critical challenge, particularly on social media platforms like X (formerly Twitter), to address the prevalent cyberbullying on these platforms. This research investigates the effectiveness of leading LLMs in generating synthetic biased and cyberbullying data and evaluates the proficiency of transformer AI models in detecting bias and cyberbullying within both authentic and synthetic contexts. The study involves semantic analysis and feature engineering on a dataset of over 48,000 sentences related to cyberbullying collected from Twitter (before it became X). Utilizing state-of-the-art LLMs and AI tools such as ChatGPT-4, Pi AI, Claude 3 Opus, and Gemini-1.5, synthetic biased, cyberbullying, and neutral data were generated to deepen the understanding of bias in human-generated data. AI models including DeBERTa, Longformer, BigBird, HateBERT, MobileBERT, DistilBERT, BERT, RoBERTa, ELECTRA, and XLNet were initially trained to classify Twitter cyberbullying data and subsequently fine-tuned, optimized, and experimentally quantized. This study focuses on intersectional cyberbullying and multilabel classification to detect both bias and cyberbullying. Additionally, it proposes two prototype applications: one that detects cyberbullying using an intersectional approach and the innovative CyberBulliedBiasedBot that combines the generation and detection of biased and cyberbullying content.

In a previous paper we defined testFAILS, a set of benchmarks for measuring the efficacy of Large Language Models in various domains. This paper defines a second-generation framework, testFAILS-2 to measure how current AI engines are progressing towards Artificial General Intelligence (AGI). The testFAILS-2 framework offers enhanced evaluation metrics that address the latest developments in Artificial Intelligence Linguistic Systems (AILS). A key feature of this re-view is the “Chat with Alan” project, a Retrieval-Augmented Generation (RAG)-based AI bot inspired by Alan Turing, designed to distinguish between human and AI generated interactions, thereby emulating Turing’s original vision. We assess a variety of models, including ChatGPT-4o-mini and other Small Language Models (SLMs), as well as prominent Large Language Models (LLMs), utilizing expanded criteria that encompass result relevance, accessibility, cost, multimodality, agent creation capabilities, emotional AI attributes, AI search capacity, and LLM-robot integration. The analysis reveals that testFAILS-2 significantly enhances the evaluation of model robustness and user productivity, while also identifying critical areas for improvement in multimodal processing and emotional reasoning. By integrating rigorous evaluation standards and novel testing methodologies, testFAILS-2 advances the assessment of AILS, providing essential insights that contribute to the ongoing development of more effective and resilient AI systems towards achieving AGI.

Despite significant advancements in Artificial Intelligence (AI) and Large Language Models (LLMs), detecting and mitigating bias remains a critical challenge, particularly within social media platforms like X (formerly Twitter) and in addressing cyberbullying present on them. This research investigates the effectiveness of leading LLMs in generating synthetic biased and cyberbullying data and evaluates the proficiency of Transformer AI models in detecting bias and cyberbullying within both authentic and synthetic contexts. The study involves semantic analysis and feature engineering on a dataset of over 48,000 sentences related to cyberbullying collected from Twitter (before it became X). Leveraging state-of-the-art LLMs such as ChatGPT-4o, Pi AI, Claude 3 Opus, and Gemini-1.5, synthetic biased, cyberbullying, and neutral data were generated to deepen the understanding of bias in human-generated data. AI models including DeBERTa, Longformer, BigBird, HateBERT, MobileBERT, DistilBERT, BERT, RoBERTa, ELECTRA, and XLNet were initially trained to classify Twitter cyberbullying data and subsequently fine-tuned, optimized, and quantized for multilabel classification (detecting both biases and cyberbullying). The study's outcomes include a prototype of a hybrid application that combines a Bias Data Detector and a Bias Data Generator, validated through extensive testing.