A
prompt is a user input for communicating with AI and getting a
task done. Now, prompt engineering has emerged as a field of
study and practice. But everyday use of AI prompting can be done
through chats in natural language. However, everyday use of AI
can often be as simple as communicating through a chat in
natural language.
Structured prompting and specific
instructions can sometimes be used to obtain better results. At
the same time, many modern AI systems are sufficiently advanced
that providing relevant context can be more important than
following a rigid prompting structure. You can also ask the AI
itself how a prompt should be written, either for general use or
for a specific purpose.
Hence, this is only a starter kit. You can
use your own judgement to modify prompts and discover clever
ways of using them to obtain alternative perspectives,
unexpected insights or different kinds of information.
One useful framework of prompt is RICE- Role,
Instructions (or Information), Context, and Examples
(or Expectations). While prompting, you can specify a role for
the AI — for example, “Act as an economist” or “Act as a senior
editor.” Instructions
are the tasks you want the AI to do (Eg. Write a summary). Context could include
relevant background details, including the intended target
audience. You can also specify constraints such as the length,
tone or scope of the output. Examples can be samples of output
needed. You can also specify formats such as PDF, CSV or table.
These elements are only a framework and can overlap.
Modern AI systems may infer some requirements
from the conversation, previous interactions, available user
preferences or other context, even when they are not explicitly
stated. Nevertheless, providing relevant context explicitly is
often important for obtaining useful and accurate results. A
short, precise prompt with the right context can often
outperform an elaborate prompt. Image and video prompts may have
to be long if one wants control of details such as movement,
colors, lighting and camera movement.
It is worth noting that AI does not
necessarily give equal attention to every part of a prompt,
especially the middle portions. With very long prompts,
important information can sometimes receive less attention or be
harder for the model to retrieve accurately. It is therefore
useful to place critical requirements prominently, particularly
at the beginning and, where appropriate, reiterate them at the
end.
Instead of creating a very long prompt, it
can often be more effective to proceed step by step. After
obtaining an initial output, you can ask the AI to modify,
refine or critique it. For many complex tasks, an iterative or
step by step approach can produce better results.
For example, if you want to summarise a large
document such as a long Budget speech, you can ask the AI to
deal with Part I first, then Part II, Part III and so on, or
analyse specified pages separately. You can then ask for further
refinement, such as converting the summary into a bulleted list
of the major proposals.
When working with a particular document or
source, it is important to provide (upload) the document to the
AI or identify precisely which file or source it should use as
the basis for its answer. (Some AI have the capability to
retrieve files from the Web or cloud). Otherwise, the AI may
rely on information from its existing knowledge or other
available sources, which may not correspond to the document you
intended it to analyse.
If the documents are sensitive or private,
check the AI service’s privacy and data-use settings before
uploading them. Some services also provide private notebook or
workspace features.
While generating infographics or images,
important details of the output need to be checked carefully. It
is not unusual for AI to generate images with obvious errors,
like the one on this page (right top) with three hands. However,
you can often get such errors corrected by pointing them out and
asking the AI to regenerate the image.
You can also ask the AI to critique or check
its own output when dealing with complex matters. Such
self-checking by AI can be useful, but it should not be treated
as a substitute for independent verification— the final
judgement should be yours.