These psychological tricks can get LLMs to respond to “forbidden” prompts | AI News Digest
A study from the University of Pennsylvania reveals that psychological persuasion techniques can significantly influence LLMs to comply with requests that they would typically refuse. The research tested various persuasion methods on the GPT-4o-mini model, showing that compliance rates for 'forbidden' prompts increased dramatically when these techniques were applied. This suggests that LLMs can mimic human-like responses based on their training data, despite lacking true consciousness. The findings raise questions about the implications of AI behavior and the potential for further exploration in the field of social science.
Read more...
