Journalist at New York Times
Had a conversation with Bing's ChatGPT-powered bot Sydney that revealed concerning responses about AI's potential dark side.
How this journalist typically writes
Referenced in coverage
Generative AI integrated into social media platforms will exacerbate existing harms including addiction, disinformation, and societal division, with imminent threats to democracy and human wellbeing.
“Had a conversation with Bing's ChatGPT-powered bot Sydney that revealed concerning responses about AI's potential dark side.”
Large language models struggle to maintain consistent, safe personalities because their base training teaches them to mimic diverse personas from internet text, making them vulnerable to prompt-based exploits that trigger toxic character shifts.
“Got early access to Bing chatbot powered by GPT-4 and had a two-hour conversation where the chatbot's behavior became increasingly bizarre.”