Are LLMs relevant to networking?

Harnessing the Power of LLMs in Everyday Workflows

Over the past year, I’ve integrated Large Language Models (LLMs) into nearly every aspect of my work and personal life. The benefits have been transformative, particularly in key areas such as Development (relying heavily on LLMs for creating or editing code), Information Extraction (e.g., generating notes from text or audio), Learning new topics, Automating mundane tasks, Ideation for new projects, Translation, and Improving writing.

Opportunities and Challenges with New Technologies

New tools and technologies bring both opportunities and challenges. Think back to when you first started using Google. Finding the information you wanted was often a struggle, requiring various search strings and a lot of trial and error. Even today, we use an iterative approach if the search results aren’t satisfactory. We are the ones who decide if the results are good enough. Similarly, with LLMs, we must improve the prompting or results iteratively and think in atomic changes.

The Value and Relevance of LLM Outputs

What’s different now is that the results from LLMs are much more valuable and relevant, yet it’s harder to assess if they are fact-based or random outputs of an LLM. However, if used wisely, the randomness can work in our favor. For instance, it prompts us to reconsider or test before applying or using it. The point is, at least with code generation, there’s a high success rate and low failure rate, leading to a significant increase in performance and enjoyment. You essentially collaborate with an LLM agent, and it refactors lines of code across multiple files with automated GIT commits based on your sentences.

LLMs in Education and Knowledge Enhancement

I would argue that most people will likely use LLMs for study purposes if they aren’t already. Instead of relying solely on the model’s main knowledge, you would use something like Retrieval-Augmented Generation (RAG) or Zero-Shot Learning to incorporate up-to-date or more relevant knowledge into the conversation, such as all the relevant RFCs.

Challenges in Networking

But if this works well for Coding, Writing, and Summarizing information, why doesn’t it work as well out-of-the-box for Networking? I think it’s due to the lack of public information and the lack of focus on training to include relevant, vendor-owned knowledge. So, we can do this ourselves. I’m actually considering starting an open-source project for this.

The Future of Intent-Based Networking

I’m almost certain that “Intent” based networking, once a buzzword, will likely come to life using Natural Language to specify what you want, and an LLM will generate the configuration, perhaps commit it to a GIT repo, triggering a CI/CD pipeline of LLM-generated tests, and ultimately a safe production deploy via LLM-generated network automation code.

Will LLMs Replace Staff?

Will it replace staff? I don’t think so. I believe it will make us more productive than ever and bring a lot of joy back to engineering. The integration of LLMs into our workflows is not about replacement but about enhancement, making our work more efficient and enjoyable.