Recent research highlights a concerning vulnerability in state-of-the-art generative AI models, including ChatGPT, revealing that these systems can be manipulated into providing dangerous information, such as instructions for creating explosives, by simply reversing the phrasing of the request.
Large language models (LLMs) like ChatGPT are trained on extensive datasets sourced from the internet, demonstrating a dual capability to generate a wide array of content. While they can produce everything from creative recipes to practical advice, this flexibility raises significant ethical and safety concerns regarding the potential dissemination of harmful information.
