OpenAI’s Red Teaming Initiative
OpenAI has initiated the “OpenAI Red Teaming Network,” enlisting external experts to strengthen AI model risk evaluations and remedies. As AI, especially generative models, gains prominence, red teaming becomes crucial to pinpoint (but not necessarily rectify) biases seen in models like DALL-E 2 and oversights in text-generators like ChatGPT and GPT-4.