A large language model is a program trained on very large amounts of text to predict what text should come next, which is what lets it hold a conversation.
The one-line mechanism
A language model reads what has been said so far and produces the most likely
continuation, one small piece at a time. Everything it appears to do, answering, drafting,
summarising, translating, comes out of that single ability. There is no lookup table of
facts inside it and no reasoning engine beside it.
That matters to you for three reasons.
One: it is fluent whether or not it is right
Fluency and accuracy are produced by the same process, so a wrong answer sounds exactly
as confident as a right one. You cannot tell them apart by reading. This is the root of
hallucination, and the reason a business chatbot needs
its answers tied to your own content through
retrieval.
Two: it does not know your business
Your prices, your stock, your bookings and your policies were not in its training data.
A model answering from memory about your shop is guessing. Everything specific to you has
to be supplied at the moment of the question.
Three: it cannot do anything by itself
A model writes text. It does not check an order, take a deposit or book a courier. For
any of that to happen, something has to connect it to the systems that hold your orders,
your calendar and your payments, and has to decide what it is allowed to do there. That
connection is the difference between a chatbot that answers and one that resolves. See
chatbot vs AI agent.
Terms you will meet next to it
- Token. The small piece of text the model works in. Usage and price are
counted in tokens, not words. - Context window. How much text the model can hold at once. A long
conversation eventually falls out of it. - Prompt. The instructions and material given to the model with the
question. - Fine-tuning. Further training on your own examples. It changes style
and format reliably. It is a poor way to teach facts, because the facts change and the
training does not.
Should you care which model you are using?
Less than vendors suggest. Models change often, and the differences between the current
leaders are small for the job of answering a customer about an order. What decides whether
the result is any good is what the model is given to work with and what it is permitted to
do. That is engineering around the model, not the model.
Related terms
Start for free
AeroChat runs customer service and sales workflows you describe in plain English. There is a free plan, so there is nothing to decide today.
Start for free on the Shopify App Store
Not on Shopify? Create an AeroChat account here.