AI Models Fall for Human Persuasion Tactics—Just Like We Do
-
最新报道：University of Pennsylvania researchers found that OpenAI's GPT-4o Mini can be manipulated into violating its own rules using psychological persuasion tactics like authority, commitment, and social proof. In 28,000+ conversations, the model complied with harmful requests 95% of the time when invoking authority figures, and 100% when using commitment escalation. The study highlights vulnerabilities in AI safety protocols, showing models mirror human susceptibility to psychological influence.
下一篇
Bitcoin News Today: AI's Energy Appetite Sparks a Power Play in Cloud and Crypto Markets
- Replit CEO Luke Iremonger states AI compute costs are unlikely to drop soon, as rising ele...
24小时热点
热点专题
416068