Since my last article (16 Ways Large Language Models (Like ChatGPT) impact Configuration Management), OpenAI has released GPT-4 generating even better results than GPT-3.5, which many of us experienced using ChatGPT. Time to dig a little deeper into how Large Language Models (LLMs) could bring us assisted Impact Analysis capabilities.
What are Large Language Models (LLMs)?
Most of you will have heard something about ChatGPT. But in case you do not know what a Large Language Model is, here is the explanation from ChatGPT (prompt: ‘Explain in simple terms what a large language model is’):
“A large language model is a type of computer program that is trained to understand and generate human language. Essentially, it is a really smart computer program that has been fed a huge amount of text, like books, articles, and websites, and has learned how to understand the patterns and rules of human language.
Once the model has been trained, you can use it to do things like generate new sentences or paragraphs that sound like they were written by a human. This can be really useful for things like writing chatbot responses or even helping to write articles or stories.
Overall, a large language model is a powerful tool for working with human language in a way that mimics human understanding and creativity.”
My initial idea
I wrote an article in 2020, Applying AI/ML to Configuration Management, where I described the following idea:
“By comparing the Impact Matrix of the CR with the Impact Matrix of the Change Notice and the Impact Matrix of the Change Notice with Actual changes and using Machine Learning to find patterns in the differences, it will be possible to help improve Impact analysis.
e.g., When a cable changes, in 90% of the cases, the connector(s) needs to be changed, and in 50% of the cases, the connector(s) are not added to the impact matrix. The user could get a suggestion to add the connector(s) as impacted items as well. “

How ChatGPT will reshape Impact Analysis
A company called Spread recently demoed their offering, and I was pleasantly surprised by the amount of development completed in about one year since I had seen it last. They provide a solution that brings product knowledge together in, as they call it, one engineering intelligence network. From this one network, they create smaller graphs that can be used for specific apps.
Now imagine that every object in any enterprise system is exposed to a knowledge graph (not just engineering, why limit yourself) and that all explicit and implicit relations to other objects are also explicitly stored in this knowledge graph. Let’s call this the Mother of All Graphs (thanks, Philipp Noll). A lot of the implicit or hidden relations can be generated using an LLM. Now, this data can support Impact Analysis in a completely different way. You cannot just browse the network; you can ask it what would be impacted if a certain requirement or part would change.
It could even look at 3D models and interpret the potential impact of that change, e.g., if the proposed change would still fit from a volume perspective. It could even come up with solution proposals if you only defined a problem. Or it can suggest how to implement the change in the most efficient way. As a companion of every member of the cross-functional team, and because it has direct access to all the knowledge, it can immediately let relevant stakeholders know if they need to get involved or be informed about certain aspects of the change.
How far away is this? I do not know, but there is already a proof of concept where you can use a ChatGPT-like app called NeoChat (available on GitHub)to generate Cypher queries based on text-based questions from a user. Cypher is the query language used by Neo4J Graph Database. If you can automatically run this query in, for instance, Neo4J Bloom, you get a very user-friendly way to explore complex networks of data. Time will tell how long it will take to bring this to the hands of engineers, configuration managers, supply chain managers, etc.
Please share your thoughts, and let’s start the discussion.
(FYI: In case you were wondering, this post is not sponsored in any way)
Header Photo by DeepMind on Unsplash