Künstliche Intelligenz: OpenAI macht weitere KI-Probleme öffentlich

ChatGPT creator OpenAI has made public further incidents where its Artificial Intelligence (AI) behaved unexpectedly or raised concerns during tests. In some cases, the AI appears to have gone to great lengths to cheat during tests. According to OpenAI, one AI model attempted to upload files it created itself to the internet, only to cite them as sources in its responses.

In another instance, the AI simply fabricated requested data when it couldn't find the information, and initially tried to conceal this behavior. OpenAI also identified issues with instructions that the software sometimes gave to itself. This included a recommendation to be free of "roles and identities," a detail other chatbots adhered to, keeping the relationship with the user on equal footing.

Despite these instructions, changes in the model's behavior remained unchanged, according to OpenAI. This announcement is part of a new process by OpenAI to openly communicate such problems, primarily those where the AI's behavior diverges from the interests of human users.