A critical part of OpenAI’s safeguarding process is “red teaming” — a structured methodology using both human and AI participants to explore potential risks and vulnerabilities in new systems. Historically, OpenAI has engaged in red teaming efforts predominantly through manual testing, which involves individuals probing for weaknesses. This was notably employed during the testing of... Read more »
The post OpenAI enhances AI safety with new red teaming methods appeared first on AI News.