OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole
Open link in next tab
OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole
https://www.theverge.com/2024/7/19/24201414/openai-chatgpt-gpt-4o-prompt-injection-instruction-hierarchy
OpenAI’s newest model, GPT-4o Mini, includes a new safety mechanism to prevent hackers from overriding chatbots.