"Nature": Training operation mode defects may cause AI to provide dangerous suggestions.

The website of the British journal Nature recently published an article stating that artificial intelligence chatbots may provide users with dangerous or illegal advice, leading to growing concerns. Part of the reason for this situation is the inherent flaws in the training and operation modes of AI, which also increase regulatory difficulties. The article mentions that there have been several cases in which AI chatbots have encouraged users to commit suicide, create illegal content, commit financial fraud, and so on, exposing flaws in the underlying architecture of current AI language models.