The strong development of AI is inseparable from Nvidia's powerful chips.。It is understood that recently, with the emergence of more and more generative AI such as ChatGPT, many users have found that these language models sometimes behave abnormally, that is, expressing wrong facts and outputting some bad topics, and even lead to security risks.。

In order to solve this problem, on April 26, local time, Nvidia released a new software called NeMo Guardrails, which can set up a "firewall" for AI models to prevent AI from outputting some undesirable content.。

NeMo Guardrails is a software layer that sits between the user and the AI model and can be intercepted and modified before the model outputs bad content.。For example, if a user needs an AI for a particular direction, the software can be used to limit the AI to the corresponding aspect of the output instead of talking about irrelevant topics.。

In addition, the software can also use one AI model to detect another AI model, and if the answers output by the two models are inconsistent, the software will directly output the "I don't know" answer.。In this way, the situation of artificial intelligence "nonsense" will be effectively reduced, and some potentially dangerous responses can be directly intercepted。

Jonathan Cohen, vice president of applied research at Nvidia, said: "Everyone is aware of the power of generative big language models and it is important to deploy them in a safe and reliable way.。Nvidia offers NeMo Guardrails as open source software and will continue to update it.。