Into the future with artificial intelligence
Artificial intelligence has attracted a lot of attention recently because it opens up the possibilities of automating manual tasks. But what advantages and potentials does AI actually open up for technical communication?
ChatGPT and Large Language Models
At the end of 2022, OpenAI released the groundbreaking chatbot ChatGPT. With its ability to generate human-like text and answer complex questions, this chatbot raises great expectations for the intelligent delivery of the right information. However, the innovative power of this technological advance also raises the question of whether we need to ditch tried-and-tested approaches and methods in technical communication – from metadata to semantic information models to knowledge graphs.
ChatGPT is based on a generalised pre-trained transformer and gives amazing answers in real time (in a positive sense) to a wide variety of user prompts. This was made possible by Large Language Models (LLM), which have learned to process and analyse language based on huge amounts of data. Among other things, words are correlated using statistical calculations and represented as vectors (embeddings). The calculated proximity of two words correlates with semantic proximity, without the language models having an understanding of language or the world in the true sense.
We will show you the way through the jungle of the latest technologies!
Feel free to contact us.
Applications in Technical Communication
In our opinion, however, it is not advisable to reliably provide customers with precise and often safety-critical technical information, such as repair instructions, on the basis of a language model alone. But if there is hype about these technologies, where do we see possible applications and added value? We have identified the following scenarios for you and tested them in practice.
Information Classification
A “classic” task of technical communication is the provision of information (documents and topics). Unfortunately, these usually have little or no metadata, especially if they are legacy data, so that a targeted search in the delivery systems (e.g. via facets) is not possible. Large Language Models offer a very good possibility to analyse and classify information stocks. The classification results become really powerful when LLMs are supplemented with factual domain knowledge, e.g. via knowledge graphs, and/or trained for a special task (keyword: fine tuning with ground truth data).
Information extraction
Knowledge often lies dormant in various documents within a company. Large Language Models have proven to be very useful in extracting knowledge from documents, such as technical features of products. Why? Because, based on the prompt, they are able to identify and extract exactly the information that is of concern. The extracted information can in turn be fed into a knowledge graph, for example. The filling of a knowledge base can thus be automated – for both structured and unstructured information. Again, these services improve when the tacit knowledge of the LLMs is concretised by explicit, structured knowledge, e.g. from a knowledge graph.
Dialogue-guided search and troubleshooting
In various projects, we use knowledge graphs to connect verified information from different information silos. We make this information available to the users via an application. We will stick to this approach in the future. Because only the content released by the relevant departments is reliable. At the same time, it is not always easy to find the information you really need. We see great potential in LLMs to guide users to the right information in question-answer scenarios without generating the answer ourselves. This is of great relevance, for example, in so-called troubleshooting or for diagnostic activities.
Content creation
This approach is still in its infancy: based on existing content, e.g. in a private instance, ChatGPT can create new content. Against the background of product liability and safety, this approach sounds daring. But how about ChatGPT supporting technical editors in the future with a draft, say of a new topic to be created? The task of the technical editors is then to prepare, verify and approve this draft. This ensures that only quality-assured content is made available. Content creation time can be significantly reduced by using ChatGPT.
We show you the way
We keep a cool head in the face of all this hype. We bring clarity into the technology jungle and show you the way. In doing so, we stick to the tried and tested and add new technologies such as LLMs and ChatGPT where it really makes sense. So that you are able to optimally support your processes and provide your customers with the right information according to their needs.
Our approach is built on different pillars. Depending on where you are; depending on which strategy you are pursuing. Metadata, semantic information models and knowledge graphs are not old news. They help us and you to provide information intelligently. And, by the way, they also help language models to become even better – because the rule is: the better the input, the better the output.
Contact us
Prof. Dr. Martin Ley
Partner and Senior Consultant
- martin.ley@pantopix.com
