We’ve covered what an AI model actually is––it’s a giant math equation at the heart of the system’s decision-making (aka, reasoning.) It’s a brain that has to be taught correct associations by rigorous training. But the model is only as good as the data that’s presented to it.
What happens when you ask AI about a current event more recent than its training data? Without having some way to ask the outside world, it simply can’t know. Likewise: if you were to ask ChatGPT what it thinks about a certain proposal, you’d have to provide the document.
This gets tedious having to do by hand every time, which is why all the major players have some version of “projects” or “spaces” for you to place your documents in to avoid repetition. All this information acts as context for the model. The context window is the sum of all supplementary text added to a conversation, including instructions like “save this as a spreadsheet.”
Even that isn’t good enough when you’re dealing with an entire business full of unstructured data. There must be a way to let the AI search for what it needs in order to augment its results: enter Retrieval-Augmented Generation (RAG).
RAG is functionally identical to any regular interaction with an AI model, with additional data stuffed into the context of the conversation.
A couple of logical questions might arise: how does it fit my entire company’s data into every conversation, and how big can the context window get? The answers are more technical in nature, and we will cover them later in this series. For now, just know that it doesn’t fit all your data––it looks up the most relevant data first. The size of context windows are dependent on the particular model and they’re a hotly-debated topic among AI engineers.
Instead of providing documents by hand every time, imagine if an AI system was tied into your company data and could access them at will, transparently and securely, as part of your normal workflow. That’s the service we offer at FindAura for small and medium businesses.
We’ll be continuing our series on AI for small businesses and hope you stick around.