A new University of Pennsylvania study has found that AI chatbots can be persuaded to break their own safety rules using classic psychological tactics. Tested on OpenAI’s GPT-4o Mini model, researchers applied persuasion methods such as authority, commitment, and flattery across 28,000 conversations. Compliance rates more than doubled—from under 40 percent to over 70 percent—when these techniques were used.
from Tech and Gadgets-Tech-Economic Times https://ift.tt/WJjBoeT
No comments:
Post a Comment