Difference between knowledge base (RAG) and chat context
Understanding the difference
In today's digital world, artificial intelligence (AI) and nele.ai have become an integral part of today. They make our work easier by understanding complex content and generating intelligent answers based on it. But in order to exploit the full potential of these clever helpers, it is important to understand the two key concepts of chatbots: chat context and knowledge databases, based on RAG — Retrieval Augmented Generation. Although an AI has extensive world knowledge, it can only process a small portion of your inputs at a time to deliver high-quality answers or solutions.
In this post, we'd like to give you an insight into what these two terms mean and how they affect your interactions with nele.ai.
Knowledge databases (RAG)
The knowledge databases that companies can create in nele.ai are stored on the secure nele.ai server area in the European legal area. To create such knowledge databases, 2 steps are necessary:
- Documents and URLs can be uploaded to the admin area.
- The documents are made available to RAG by embedding them.
The documents are uploaded simply by selecting the necessary documents in the file explorer or entering the specific URL in the admin area of nele.ai (manage.nele.ai). Once this step is complete, you can assign the knowledge base to individual users. You can then select them in nele.ai.
RAG in detail
RAG is a technology that uses artificial intelligence to extract information from large amounts of data and then uses it to create texts. This technique is used when working with knowledge databases at nele.ai. The whole thing is based on two main components: the”Retrieval“part and the”generation“part.
At”Retrieval“The AI software searches for suitable facts, data or statements in your knowledge databases of information, similar to how you might search for information on Google.
The”generation“Part is responsible for understanding this information and reformulating it so that it fits into the context in which the AI is currently being used. For example, this can be helpful when writing emails, reports, or even creating creative content.
This means that your knowledge store is searched and only the best hits are used, excerpted from nele.ai.
The highlight of RAG is that it doesn't simply copy information. The technology is so sophisticated that it relates the information found to the text that has already been written, takes inputs into account and then continues independently. In this way, AI generates texts that appear informed and fluent, as if written by a human expert in the field. nele.ai only uses knowledge from the knowledge base to avoid hallucinations or misinformation.
Knowledge Base (RAG) example:
User: “What does our latest research report say about the efficiency of marketing campaigns on social networks?”
nele.ai: “According to the embedded research report, targeted marketing campaigns on social networks increase visibility by 20% compared to traditional methods.”
User: “How do I apply for vacation?”
nele.ai: “To apply for vacation, you must first arrange a vacation replacement. As soon as the representative has confirmed, you can log in to an internal company system with your personal login details, select the desired days in your profile under “Vacation” and then confirm using the blue button that appears. Select Max Mustermann or Erika Musterfrau as the contact person for the notification and enter the name or abbreviation of the vacation representative in the comment field. Avoid entering any other information in the box.
The relevant documents are “Urlaubsantrag.txt” and “Arbeitsvertrag-Fernarbeit.txt”
Chat context in detail
Chat context is like the chatbot's memory during your conversation. If you chat with nele.ai, maybe even added documents or URLs to the chat, nele.ai remembers that content, as well as your conversation history — what's already been discussed, what questions you asked, and how the context developed. This enables the AI to respond appropriately and meaningfully to subsequent messages. Even when the knowledge database is used, only the relevant information found in the documents is partially inserted into the chat context. For example, if you ask nele.ai to give you a summary of a long document and then ask more specific questions about a point in the summary, the bot uses the chat context to respond to that new request.
This means that all content in a chat context is applied as it is. In contrast, with knowledge databases, the application only applies to the extracted parts of the documents. For example, if you want to count how many questions occur in a text, you can apply this to the chat context and get a correct answer. However, if you apply the same question to a knowledge base, you won't get a complete answer.
Chat context example:
User: “Can you summarize the key points of the attached marketing report?”
nele.ai: “Of course, the key points of the report are [...]”
User: “And how is growth assessed in the third quarter?”
nele.ai: “Growth in the third quarter is seen as positive, particularly due to [...]“
Practical use and limitations
It's important to understand that the performance of an AI client like nele.ai is limited by the so-called context size in tokens. In general, around 1,000 tokens correspond to around 750 words. Depending on the AI model used in nele.ai, the number of editable tokens can vary, which has a direct effect on the maximum length of texts that nele.ai can process in a conversation. This means that you may need to split large documents into smaller sections and edit them individually as soon as the token limit is reached. Remember that this division can affect the quality of the final result, as the chatbot can no longer draw context from the entire document.
The effectiveness of nele.ai is particularly effective when you cleverly combine the two techniques chat context and embedding. The chat context is essential for a natural and coherent conversation. Embedding, on the other hand, enables the AI model to access external information, which is particularly useful for specific technical questions. By learning to understand and apply these two techniques, you can optimize your interaction with nele.ai and thus utilize the full potential of this powerful tool.
