Category: Uncategorized

  • Red teaming ChatGPT via Jailbreaking: Bias, Robustness, Reliability and Toxicity

    Rise of powerful NLP sparks ethical concerns: LLMs like ChatGPT show potential for bias, unreliability, toxicity, demanding new ethical benchmarks and design considerations. This study analyzes ChatGPT across four key areas (bias, reliability, robustness, toxicity) and reveals limitations of existing benchmarks. More research needed to build responsible LLMs and mitigate…