Artificial Intelligence, commonly referred to as AI, at the simplest level is the ability for computer applications to complete tasks without an explicit pre-programmed solution by utilizing large amounts of data. Non-AI applications have pre-programmed explicit solutions depending on the inputs.
Major progress has been made in past couple years in relation to a particular type of AI known as Large Language Models(LLMs). Large Language Models are primarily focused on text as the word language implies. Networking devices are configured completely by text. Since LLMs are utilized for text, that makes network devices a perfect uses case for LLMs and AI in Networking.
Large Language Models, commonly referred to as LLMs, are a type of Generative AI model. The most popular Large Language Models are OpenAI's GPT-4, Anthropic's Claude Opus, and Google's Gemini. These are the models that power the actual chat application. For example, OpenAI's ChatGPT is an application utilizing their GPT-4 Large Language Model.
Large Language Models are trained on large amounts of data. Essentially the entire internet and any other text data the firm behind them were able to access. In regards to networking, this includes all network vendor documentation, all troubleshooting forum posts with questions and solutions, all network blog posts, etc.
Every LLM application, such as ChatGPT, Perplexity, and Github Copilot, utilize Large Language Models behind the scene but offer traditional software in front of the model in order to provide a usable application. Sending your inputs directly to a model without any context would yield very undesirable outputs. Context for LLMs are incredibly important similar to how context can change the meaning of a sentence or statement even when we communicate as humans.
Context is the data that gets passed to the LLMs. The data(context) you pass into the LLM determines how it will craft its output. The LLM will essentially go through all its training data and find the most relevant data to respond with. If the LLM understands you're talking about Cisco Networking Devices, it will start looking through the Cisco Documentation from its training data. Without the LLM knowing you're talking about Cisco Networking Devices, you may end up getting an answer relating to some other Network Devices.
With proper context, LLMs are fully capable of configuring networks and diagnosing network issues. They can take natural language inputs such as instructions from an engineer, business requirements from an operator, or a trouble ticket from your ticketing system. With those natural language inputs they can fully complete your networking changes or diagnose your network issues. Without proper context, none of this is possible.
ITVA's platform ensures the LLM always has proper context for your request. We pull live data from you devices to give to the LLM such as output of CLI commands or snippets of the current configuration. Most importantly, we give the LLM relevant examples from our huge proprietary dataset. The best context for LLMs is always accurate and human-trained examples. ITVA's dataset grows every day by real human engineers.
