From the course: Building Secure and Trustworthy LLMs Using NVIDIA Guardrails

Unlock the full course today

Join today to access over 24,000 courses taught by industry experts.

Behind the scenes: How Guardrails enforces LLM safety

Behind the scenes: How Guardrails enforces LLM safety

From the course: Building Secure and Trustworthy LLMs Using NVIDIA Guardrails

Start my 1-month free trial Buy for my team

Behind the scenes: How Guardrails enforces LLM safety

“

- [Instructor] Before we start implementing Guardrails in Python, let's take a look at how they work behind the scenes. The primary goal of this discussion is to help you understand how Guardrails maintain the integrity and safety of LLMs. We'll explore the processes and mechanisms behind these safety measures and why they are critical in the development of reliable and ethical AI. So let's get started. NVIDIA's NeMo Guardrails system supports five main types of safety rails. These are input, dialog, output retrieval, and execution. Each type of rail plays a unique role in maintaining the safety and integrity of interactions with LLMs. Input rails, validate and potentially alter or reject user inputs to ensure safety right from the start. Dialog rails maintain the flow of conversation and decision-making processes while retrieval rails handle accuracy of data retrieval, ensuring that only reliable information is used. Execution rails oversee the inputs and outputs of custom actions…

Contents