Use LLMs to share knowledge within your organisation

Since the rise of ChatGPT, there’s no getting around it: ‘Large Language Models’* (LLMs) are here to stay. Several technology companies have developed advanced LLMs. Well-known examples include OpenAI’s ChatGPT, Meta’s LLaMA and Google’s laMDA. You might be thinking: ‘All very interesting, but what can I do with them?’

In practice, finding a suitable business case for LLM solutions is a major challenge for many organisations and entrepreneurs. In this blog, we explain why you might want to use LLMs to share knowledge within your organisation. We look at the challenges of knowledge management and the potential role of LLMs.  

What is knowledge management?
Knowledge management means storing, organising, managing and sharing knowledge within an organisation. It contributes significantly to an organisation’s success: when knowledge is not easy to access, the consequences can be damaging, as valuable time is spent searching for information rather than on core activities. Successful knowledge management depends not only on people, but also on processes and technology. Knowledge management systems have been developed to streamline the processes involved. Yet many organisations still face challenges in this area. Some common ones are:  

1. Fragmented knowledge silos

Does this sound familiar? Different departments in your organisation each have their own tools and methods for sharing data (think SharePoint, Slack or Google Drive). Useful within teams, but frustrating for organisation-wide collaboration. Without clear rules for organising documents, finding information becomes like looking for a needle in a haystack.
 

1. Hard-to-access information

As your organisation grows and collects more information, it becomes increasingly difficult for employees to find what they need. This is particularly a problem if your knowledge system relies on folder structures, as in SharePoint.
 

1. Lack of adoption among employees

When knowledge systems are difficult to use, employees are slower to adopt them. This leads to the loss of valuable knowledge and limits the company’s ability to learn and adapt quickly.  

LLMs to the rescue (?)
LLMs are also known as generative models. They are designed to generate content based on the input they receive. Although traditional LLMs are powerful, they have limitations. They require a lot of training data and computing power, and they cannot answer questions that depend on context they have not been ‘pre-trained’ on. In practice, this means a general-purpose LLM such as ChatGPT is unlikely to answer specific questions about your company data correctly. After all, it is trained on internet data, so it only knows what is publicly available online. There is little point in asking ChatGPT about your revenue last year: it will simply refer you to your own finance department.  

Fortunately, there is always a solution in the world of AI. In this case, it is (don’t be alarmed) ‘Retrieval Augmented Generation’, or RAG. Let’s break that down: ‘Retrieval’ means ‘finding and bringing back’, ‘Augmented’ means ‘adding to’, and ‘Generation’ means ‘producing something new’. Retrieval models are not new. They are designed to retrieve relevant information from a given knowledge base. They often use semantic search techniques to find and rank the most relevant information for a query. What retrieval models cannot do on their own is generate creative or new content. RAG addresses this by combining the generative power of traditional LLMs with the specificity of retrieval models. For your organisation’s data, RAG could work as follows:  

1. Based on a question about your organisation, a retrieval model retrieves relevant context from your own knowledge base
   
2. An LLM answers the question using the retrieved context, including a reference to its source 

Using RAG could offer an effective solution to the challenges of knowledge management. Here’s why:  

• By connecting individual knowledge systems to the RAG model, the problem of fragmented knowledge silos disappears. Instead, there is now a single knowledge portal where employees can access organisational information. 

• The problem of hard-to-access information caused by deep and complex folder structures disappears with a simpler search method. Employees only need to enter a question or search query through an interactive interface. This makes it easier for everyone to find relevant information, even as the organisation grows and collects more data. It helps make information more accessible. 

• A more user-friendly way to interact with knowledge management systems helps reduce the lack of adoption among employees.  

Conclusion
It is an exciting time for organisations looking to benefit from the power of LLMs
The emergence of these new technologies, particularly Retrieval Augmented Generation, offers opportunities to address challenges in knowledge management.