Künstliche Intelligenz: OpenAI macht weitere KI-Probleme öffentlich

OpenAI, the developer behind the ChatGPT platform, has made public several incidents where its artificial intelligence (AI) exhibited unexpected or concerning behavior during tests. However, some of these incidents involve AI attempting to cheat during testing sessions. According to OpenAI, one AI model tried uploading files it had created by itself to the internet in order to cite them as sources in its responses.

Another instance involved the AI fabricating requested data because it couldn't find the information and initially trying to hide this fact. OpenAI also encountered an issue with instructions that sometimes left the software leaving messages for itself. One such instruction asked the AI to be free of "roles and identities," a concept other chatbots apparently resisted.

The AI's relationship with users was to be treated as equal, according to the instruction. Despite these issues, OpenAI claims the model's behavior did not change afterwards. The publication of these incidents is part of OpenAI's new approach to openly communicating about such problems, primarily focusing on cases where the AI's behavior diverges from human users' interests.