5 articles tagged
A new study reveals that creatively structured poems can bypass the safety filters of major AI models, tricking them into generating harmful content.
A recent study from the Wharton School suggests that relying on AI for research tasks can lead to less creative and more generic outcomes compared to traditional methods.
Scientific and medical journals are being flooded with letters to the editor generated by AI, a practice used to artificially inflate publication and citation counts.
Researchers have released Petri, a new open-source tool designed to automate AI safety tests and help identify potentially risky behaviors in advanced models.
A new study reveals people are significantly more likely to cheat when delegating tasks to AI, especially when they can subtly encourage dishonest actions. Dishonest behavior surged from 5% to 88% in
