Model, service and feature: three different layers

Goal: tell three things apart

You may have noticed that the same question produces different answers in different AI services, or even in different modes of the same service. In this guide, you will try this yourself with two examples and learn which of three layers usually explains the difference.

Make the difference visible in four steps

1. 1

   Choose one request

   Use either example below or write your own equally simple question.

2. 2

   Ask it twice

   Send exactly the same request to two AI services, or to the same service in two modes, such as with web search switched on and off.

3. 3

   Record what differed

   Write down a concrete difference: content, freshness, tone or additional functions offered.

4. 4

   Identify the layer

   Guess whether the model, service or individual feature explains the difference. Check your guess using the follow-up question below.

Try this

Briefly tell me what has happened in the news in Finland today.

Summarise your previous answer in three bullet points and add one source link to each point so I can check the claim.

I just asked the same question in another AI service, or in a different mode of the same service. Whether the answers were almost identical or clearly different, explain which of the three layers, model, service or an individual feature such as web search, most likely explains what I observed. Also tell me where I can check this in the service's own information.

The same exercise with a different request

Write a four-line poem about the first cup of coffee in the morning.

I also tried the same poetry request in another AI service, or in another mode of the same service, but I am not pasting either poem here. Still, explain which aspects, such as word choices, length, rhymes or mood, are generally influenced by the model, which by service defaults, and which by extra functions provided by an individual feature. Finish by explaining which of these three I should suspect first if the poem surprises me.

Now name the three layers

A language model is a technical model that generates responses based on text or other material. A service makes the model available and determines matters such as accounts, storage and terms of use. A feature is one function within a service, such as summarisation or organising images. The same model can be used in several services, and a service can change its model.

Neither difference means that one service is generally better. The difference only tells you which layer to examine when something does not work as expected.

Imagine an AI feature that reads my incoming emails and suggests a ready-made reply draft directly in my email application. Explain separately what the language model is, what service provides it and what the feature itself is. Also describe its likely purpose and the kinds of information it probably processes. Mark anything you cannot know from this description, and explain where I would check it in the service I actually use.

Once you know the basics

Next, you can explore when a service uses external search, files or other tools. Keep the same principle: define the task, check the answer and verify important facts against a source.