NVIDIA, the multinational technology company, has launched a new tool called NeMo Guardrails to assist developers in creating safer and more accurate AI models. This comes after the release of NVIDIA’s AI Foundations service, which provides businesses with the ability to train large language models (LLMs) on their proprietary data since March. NeMo Guardrails enables software engineers to implement three types of limits on their in-house LLMs to ensure accuracy, appropriateness, and safety. The tool allows companies to establish “topical guardrails,” “safety limits,” and “security limits” to prevent their apps from addressing untrained subjects, pull accurate information, and link to secure apps.
NVIDIA claims that NeMo Guardrails is compatible with all LLMs, including ChatGPT, and nearly any software developer can use the software. In addition, since NeMo Guardrails is an open-source tool, NVIDIA says it will function with all the tools enterprise developers already employ. The company is integrating the tool into its existing NeMo framework for creating generative AI models, which is available to business customers through its AI Enterprise software platform and AI Foundations service.
The launch of NeMo Guardrails follows the scrutiny of high-profile generative AIs, such as Microsoft Bing and Google Bard, which have been accused of disseminating false or “hallucination” information. According to NVIDIA, NeMo Guardrails is the product of several years of research and is open-sourced to contribute to the developer community’s efforts to work on AI safety. The company aims to keep their smart services aligned with safety, privacy, and security requirements to stay on track with the development of AI technologies.
For those interested in learning more about the workings of NeMo Guardrails, NVIDIA has published a blog post that offers a deep dive into the subject and shares information on how to get started with the software.