Page 36 - CyberSecurityEssentialsEnglish
P. 36
Video: What Is Machine Learning?
A simple visual explanation of learning from
data and the difference between a learned
model and manually written rules.
Duration: about 8 minutes | Level: Beginner
Learning objective: Explain machine
learning through a clear example and
distinguish training from inference.
Source: IBM Technology [14].
https://youtu.be/9gGnTQTYNaE
Verified: June 6, 2026
● Remember: Section summary
• Key idea: Measure the model’s ability to handle new examples, not its ability to
memorize training examples.
• Practical decision: Prepare a local test set that represents the task and monitor changes
in the real environment after deployment.
• Mistake to avoid: Using the formal test set for tuning, or assuming past performance
will continue with a different audience or dataset.
1.5 How Do Language Models Generate Answers?
◆ Key idea: A language model predicts the next unit of text.
The model divides text into small units and predicts the next unit from the words and
instructions available in the context. It repeats this process to form a sentence and then a
paragraph. This mechanism can produce coherent language, but a statement is not true
merely because it sounds convincing [11].
The systems behind today’s conversational tools are usually called large language models,
abbreviated LLMs. They are described as large because they are trained on very large
amounts of text and hold a very large number of internal values. Size explains why one model
can handle many different language tasks. It does not make the model’s statements true.
Readers who want the technical origin of this family of models will find it in the architecture
paper listed in the references [8].
Section 1.4 noted that a model may be trained to predict a hidden part of a text or to predict
the next unit. These are two different training styles. Models used for open-ended writing
and conversation are usually trained to predict what comes next, which is the process
described in this section.
35
