Page 93 - CyberSecurityEssentialsEnglish
P. 93
TABLE 2-16: COMMON MISTAKES AND THE SUGGESTED CORRECTION
The mistake Why it weakens the result The correction
The system does not know the final Write what will happen after the
An undefined objective
function. output is used.
It fills the gaps with general Add only the information that
Missing context
assumptions. changes the decision.
Priorities are lost and conflicts Use headings and rank the
Much unordered detail
appear. constraints.
It mixes general knowledge with your Attach the source and define the
No source
specific material. policy on outside knowledge.
The word improve is repeated with Write an acceptance list that can
A vague quality standard
no diagnosis. be checked.
One unrepresentative Use varied examples and edge
It imitates a narrow case or a bias.
example cases.
Asking for lengthy The length of an explanation does Ask for brief justifications and an
thinking as a guarantee not prove the result. external test.
Uncalibrated numerical It suggests a probabilistic precision Ask for assumptions, evidence, and
confidence that does not exist. weak points.
Treating delimiters as Content may inject instructions Restrict permissions and inspect
security despite the organization. input and output.
Patching inside a lost Create a new brief and start from a
Conflicting instructions accumulate.
conversation clean version.
Copying a famous
Extract the principle and test it on
prompt without It may not suit the model or the task.
your own cases.
understanding it
Trying to solve It does not address missing Change the source or the model, or
everything by prompting knowledge or a missing tool. build a workflow.
● Remember: separating the prompt from the workflow
A prompt describes a step or an interaction. When a task becomes repeated and multi-
step and requires approvals, state, and error handling, we move from prompt
engineering to workflow design, which is the subject of Chapter Five.
2.13 A Prompt Library: From Personal Text to a Testable Asset
◆ Key idea: keep what worked, together with the conditions of its success.
A prompt library is not a file full of ready-made phrases. It is a small set of templates tied
to real tasks, with the inputs they require, evaluation cases, limits of use, and the date of
the last test.
92

