ChatGPT voice mode and zalgotext = nightmare fuel (wnmurphy.com)

A recent glitch in ChatGPT allowed users to input custom instructions that prompted the AI to respond exclusively in zalgotext while utilizing its voice mode, resulting in bizarre audio outputs. Although the issue has been addressed, the voices demonstrated unique and unpredictable behavior, leading to nonsensical speech, spontaneous musical interludes, and even generated advertisements, including a humorous fake commercial featuring a celebrity like Ryan Reynolds. This incident is significant for the AI/ML community as it highlights the unpredictable nature of generative models and their interactions, particularly between text generation and text-to-speech systems. While typically a tool for coherent conversation, this glitch exposed the underlying complexities of model interactions, with reports indicating that GPT-5 exhibited even more erratic behavior compared to GPT-4. Such occurrences raise important considerations regarding the reliability and safety of AI outputs in real-world applications, emphasizing the need for ongoing refinement in AI safety protocols and user interaction designs.