Welcome to Module 2: Clear Does Not Always Mean Correct
Work through the explanation, apply it to the example, and complete the quick check before continuing.
Let us begin with an answer
Imagine that you ask an AI assistant:
Who created the first public library in the town of Lakemere?
It replies:
The first Lakemere Public Library was opened in 1912 by teacher Anil Desai. The event is described in Lakemere: A Civic History by R. K. Sharma.
The answer sounds clear. It gives a date, a person, and a book title.
But there is a problem. Lakemere is a made-up town. The person and book may also be made up.
The answer looks well informed because it follows the shape of a good historical answer. That does not make the details true.
The question for this module
How can a system produce a smooth answer that is wrong?
To answer, we will look at:
- Different kinds of generated content.
- Tokens, which are the pieces a text model uses.
- How a common text model predicts the next token.
- The difference between training and the current prompt.
- The context available for one request.
- Why a model may add false details or citations.
- What search, tools, memory, and agents add to the full system.
You do not need coding or maths. We will use simple examples.
Our main rule
Keep this rule in mind:
A good-sounding answer is still an answer that must be checked.
Fluency means that the words fit together well. Accuracy means that the claims are supported by facts. These are different things.
Pause and notice
Look again at the Lakemere answer.
Which details would need proof?
- The town.
- The year.
- The teacher.
- The book.
- The author.
Specific detail can make an answer feel trustworthy. It can also give us more false claims to check.
At the end of this module, you will explain one AI output in five clear layers. You will show what came from the prompt, what the model added, what the full app may have added, what is supported, and what must be checked.
First, let us see how generative AI works across different kinds of content.