Language model – what does it mean?

How it works

The assistant consists of two parts. The language model understands the question and formulates the answer; it provides general language skills. The company’s knowledge comes from the knowledge base: the price list, the products and the documents.

On its own, the model does not know your prices or your stock. The assistant writes its answer from parts of the knowledge base, and the source is shown next to it. Where the model runs also matters, because the conversations reach it too.

An example from a company

The same model can run the assistant of a winery and of a spare parts shop. The difference is the uploaded content.

The winery’s assistant answers from its own product descriptions and price list, the spare parts shop’s from its own stock and compatibility list. Customers can ask questions the same way in both places, yet the content of the answers is different.

What this means for business owners

The name of the model matters less than where it runs and how it is restricted to the company’s content. Ask where the data stays, and what prevents the assistant from answering from outside the knowledge base.

If the assistant is not tied to the company’s content, it may give a general answer that does not apply to your company. Also ask whether the provider changes models, and whether it notifies you when it does.

How we do it

The assistant answers from the company’s own knowledge base. The data stays in the EU, including model calls, and the retention period is 90 days. If there is no content, the assistant does not make up an answer: it asks for contact details and hands the question off to a colleague.

It answers in the customer’s language, even if the knowledge base is in another language.