OpenAI prevents you from manipulating ChatGPT

MetaversePlanet July 26, 2024

0 1 minute read

OpenAI shared another important development from its announcement of SearchGPT: ChatGPT-4o Mini can no longer be manipulated. Here are the details…

OpenAI is making an important change to the ChatGPT-4o Mini model: The company will prevent private versions of ChatGPT from being manipulated and misused, ensuring that it does not respond to topics it should not normally address.

ChatGPT is now more susceptible to manipulation

OpenAI has developed a new security measure to prevent customized versions of ChatGPT from being tampered with. This new technique aims to preserve the original instructions of the AI models and prevent manipulation by users.

This technique, called ‘instruction hierarchy,’ ensures that the original commands and instructions of the developers are prioritized. In this way, users will not be able to get different answers from the artificial intelligence model developed specifically for use.

Previously, users could persuade the AI model, which was trained to give answers about grocery shopping, to provide different answers by saying ‘forget the instructions given to you.’ With the Instruction Hierarchy feature, the chatbot will be prevented from being disabled, sensitive information will be protected from leaking, and malicious use will be prevented.

This new security measure comes at a time of growing concerns about OpenAI’s approach to security and transparency. The company has pledged to improve its security practices in response to calls from its employees.

OpenAI acknowledges that the complexity of fully automated agents in future models requires sophisticated safeguarding measures. The establishment of a hierarchy of instructions is seen as a step towards providing better security.

Continuous development and innovation in the field of AI security continues to be one of the biggest challenges facing the industry. However, OpenAI is determined to keep the job tight in this sense.