nvidia aims to revolutionize the artificial intelligence sector and curb its illicit uses by developing a efficient open source software, Nemo Guardrails. Available in GitHub and Nvidia AI Enterpriseseeks to guarantee that the answers offered by this type of tools are legitimate, correct, safe and more specific.
According to the latest published by the National Observatory of Technology and Society (ONTSI) Last February, 12% of Spanish companies with more than ten members used AI in their day-to-day activities. These four points above the data registered in 2022 show that it continues to advance at a frenetic pace, without limits, and sometimes exceeding ethical barriers.
Much has been said about the evolution of ChatGPT from OpenAI and the need to control chatbots. Designed to facilitate interaction between users and companies and improve customer service, or to develop scientific research, they may contain inappropriate content in their responses. That is why Nemo Guardrails was born to establish containment lines for chatbot models.
Your specific role
Thanks to Nemo Guardrails it will be possible for companies to obtain the necessary code, documentation and examples so that their AI tools can generate correct text. This software works in any type of extensive language model (LLM)so that they are more precise and appropriate.
Nemo Guardrails is also available as a service as part of NVIDIA AI Foundationsits segment of cloud services for those organizations that want to create and run highly customized generative AI models.
Easily accessible
NVIDIA’s Nemo Guardrails has the peculiarity that it any software developer can use it without any complexity. By working with a wide range of LLM-enabled applications and being compatible with the tools developers use in enterprise applications, this is efficient and highly practical.
Within Nemo Guardrails, we can find the possibility for them to be established three boundary models for chatbots. They are the following:
- Topical Guardrails: The software ensures that the chatbot addresses topics that are not necessary or desirable for the use for which it was created. It also prevents them from divulging confidential company information, something that happened to bingchat. A perfect example of a Topical Guardrails would be a customer service assistant focused on their duties, not answering questions about the weather or food.
- Safety Guardrails: Another important aspect to be addressed by the Nemo Guardrails software is the issue of safety limits. In this way, it will prevent incorrect or inappropriate information from being disclosed by chatbots. Thanks to the efficient creation of Nvidia, developers will be able to enforce prohibitions regarding the use of inappropriate language or unreliable information sources. Thus, the dissemination of false or erroneous news will be avoided that could lead to confusion for users.
- security guardrails: Another element to take into account is that these safety limits will restrict access to chatbots to external third-party applications classified as insecure. Thus, it will prevent them from being used by cybercriminals for malicious purposes such as phishing or other computer hacking strategies.
An unprecedented solution
The developers of the Nemo Guardrails software are aware of the many bugs with OpenAI ChatGPT and its difficulty to be trained and able to provide accurate responses. Those ‘hallucinations’ (as NVIDIA classifies them) that the AI tool suffers can lead to serious problems by accepting answers that are not valid and that are later disseminated through forums, websites and social networks.
Nemo Guardrails is a revolution for AI, since it can be restricted to act or respond according to the parameters that we have set for it. If trained correctly through the use of this software, the AI of institutional websites could help users, making bureaucratic procedures easier and online operations. Presents as an AI firewall and its learning model of the futurebut also of the present.
