It used to happen every so often, but now it happens all the time: the feeling that by the time a technology is mastered, the next technology step has already arrived. Today, we see this with generative artificial intelligence (GenAI). The GenAI industry has the potential to be an open industry, with lots of models and frameworks available for evaluation. But the fast pace of GenAI advancement leads to a cycle of continuous reevaluation.

It’s about time

The extreme speed of GenAI technology evolution contrasts with the massive impact GenAI will have on the way we look at network automation. Waiting until GenAI technology matures feels like waiting for Godot because the trillions of dollars poured into the industry will continue to accelerate the speed of innovation. This raises the question: When is the right time to start?

Just dive in and learn to swim in the GenAI pool

The short answer to that question is: start yesterday. Rest assured, there is little risk that you will drown in the GenAI pool, and while the first strokes may feel a bit clumsy, you will quickly learn to swim smoothly.

First, the openness of the industry allows for start fast, fail fast. As you move on, your GenAI framework starts being built around start fast, succeed fast components.

Second, is your data. As we explained in our first post on GenAI, having the right data is critical for reaping the benefits of GenAI. Ideally, you will have started building a structure around your data a long time ago. If not, start now. This will allow you to attach your data to your GenAI framework irrespective of the state of GenAI technology.

Last, your GenAI framework and all your data only need to make sense for your use cases. Full-circle network automation is the ambition, but your network automation assistant needs to walk before it can run, so the starting point could be an API assistant, a troubleshooting assistant, a dashboard assistant, and other smaller- or larger-scope assistants. You can evolve your assistant step by step.

Time moves on and your data does, too

Just as time moves on, the industry has moved on from GPT3 to GPT4 and now GPT5. The result is a massive large language model, trained on the last 100 years of data for GPT3, 103 years for GPT4 and 105 years for GPT5. As a result, the past has been factored in, and indeed moves on year by year. Somehow, it seems acceptable to assume that the most recent data brings the most value from GPT5 for your decisions and objectives today, even though next year this data will be a year old. 

A technique called Retrieval Augmented Generation (RAG) allows you to augment that past data with your most recent data, which will be available in the prompt. However, the best results would come from training the language model on past, present and future data. 

This requires you to gather all the information and data that’s relevant to your use case and product as you go along—through an assistant that sees everything related to your intended outcome, hears everything as your teams collaborate on the common objective, and picks it all up as you go through your CI/CD pipelines, test results, meeting minutes and whiteboard diagrams. 

An approach like this would allow the language model to pick up on the intent and how it should work, as well as what is not intended, which tests have failed and why they have failed. As your product is delivered, it will include all the information captured in the language model related to the product’s intent. The language model does not need to be trained on the product after it has been released. Your community has talked about it, used it and verified it during its development.

Network automation is moving on fast

Nokia Network Services Platform (NSP) is the network automation platform of choice for the world’s most advanced communications service providers.

It configures and assures entire end-to-end IP networks. As a software-defined networking (SDN) controller, it incorporates all the IP network know-how and algorithms built up over the past 30 years and oversees all data coming from the nodes across the entire path of the network service. It has embedded its developer community into a GenAI-assisted NSP, incorporating and improving every minute on the good and the bad experiences of the developers.

Dare to swim in the ocean of opportunities GenAI and NSP can bring to you and your networks. You may not swim like a mermaid at first, but you will soon be able to deliver a healthy end-to-end network mind in a healthy NSP body.

Visit our IP networks and AI page to find out how we are integrating GenAI more extensively into NSP for automating network operations.