QUANTIC
LEARN TO CODE WITH CHATGPT
PROMPTS FOR PROGRAMMING IN CHATGPT
Prompt Template
When building a prompt for ChatGPT to generate code, make sure
you consider including the following elements:
• What specialized knowledge currently exists? Is there
existing code to start from? Do you already have the steps
needed for a particular calculation or procedure?
• What are the inputs to the program? If the user provides
inputs, what are they and what format will they be in? If
there are data files, what formats will they be in? If the
program is to retrieve information from the internet, what
Write a Python program to
site(s) should it use? convert pounds to long tons.
• What should the program do? How should it handle errors There are 2,240 pounds per
long ton. The user will enter
that occur due to bad inputs or faulty code? pounds into a graphical user
interface, and long tons should
• What should the output look like? Should it be graphical, appear to two decimal places of
text-based, or a combination? If there are files to be output, precision. Explain the code you
what format should they be in? write.
• What constraints should the program follow? Should it have
a graphical user interface (GUI)? Should it be accessible as
a web-based application or a desktop app? What
programming language should it be in?
• What other instructions should ChatGPT follow? Should it
explain the code it generates? Should it provide alternative
approaches?
Some prompts might not require all of the elements, and there's
no requirement to put the elements in any particular order.
©2023 Pedago, LLC. All rights reserved.
LEARN TO CODE WITH CHATGPT QUANTIC
General Prompting Strategies
Specific and detailed prompts are more likely to deliver correct
code, while general and open prompts generate capabilities and
features you might not have thought of.
Use ChatGPT's Custom Instructions to give it information that Custom instructions only apply
applies to all of your prompts (e.g., "Write all code in Python," to new conversations. Your
"Comment all code," etc.) existing conversations will not
change when you create or
Use delimiters to help ChatGPT recognize separate elements of change custom instructions.
your prompt. For example, if you include code in a prompt, put a
line of three asterisks before and after it, and tell ChatGPT that's
what you've done.
Reference existing apps that do something similar to what you
want to help ChatGPT understand the task.
Provide sample inputs to help ChatGPT understand their format
and content.
Use programming terminology to get ChatGPT to write code
using specific constructs. For example, if you want your code to be
in a reusable function, tell ChatGPT to "write a function that..."
instead of "write a program that..."
Describe the user to help ChatGPT better understand the
problem. For example, "Write a program for elementary school
students that tests them on simple addition problems" helps
ChatGPT build a user interface suitable for children.
Specify how the code should handle errors. This includes
erroneous inputs, unavailable resources (e.g., a file is missing or
the network is down), or runtime errors. For example, while you're
in the debugging process you might consider including "Let all
unanticipated runtime errors generate a traceback."
©2023 Pedago, LLC. All rights reserved.
LEARN TO CODE WITH CHATGPT QUANTIC
Strategies for Complex Tasks
Use the least to most strategy. Start with the basic core of the
application, then once it's working prompt ChatGPT to modify the
code to add a new feature or capability. Get that working, then add
the next feature or capability. Repeat this process until you have
the app you want. This strategy gives ChatGPT smaller problems
to work on, which helps it succeed.
Use question refinement to understand what ChatGPT needs in a
prompt. This helps you understand what ChatGPT considers
critical information for the task at hand.
Use the generated knowledge strategy. Prompt ChatGPT to
explain its approach to the problem before generating any code.
Review that approach, and prompt ChatGPT to make changes
where you see flaws in its logic. Then prompt ChatGPT to follow its
approach, either all at once or in small steps as suggested by the
least to most strategy.
Give ChatGPT the steps it needs to follow to complete the task.
This helps ChatGPT by eliminating one potential source of error
(i.e., a flawed approach) and gives you more control over the
operation of the resulting program.
ChatGPT has a limit to the amount of context it can maintain in a As of September 2023
conversation, and once it hits that limit it starts to "forget" ChatGPT's underlying model
information early in the conversation. In long conversations, you has a token limit of 4,096. On
average, every two words of
may need to periodically reset the context by summarizing where conversation take up three
you are and including the most current version of the code. tokens. Both prompts and
responses count against the
token limit.
©2023 Pedago, LLC. All rights reserved.
LEARN TO CODE WITH CHATGPT QUANTIC
Strategies for Debugging
Regenerate the prompt to allow ChatGPT's randomness injection
to work in your favor by getting a different and hopefully correct
response. Use this strategy for simple prompts only. If ChatGPT
has already generated three or more incorrect responses you'll
need to take another approach.
Paste the traceback you get from Python in the event of a runtime
error. Since the traceback provides the specific error and where in
the code it occurred, ChatGPT has lots of information to put the
error in context. This strategy should always be your first choice
when you have a traceback.
Explain the problem your code is having in plain language.
Include the inputs that are causing the problem and what the
expected performance is. If you can anticipate ChatGPT giving you
an incorrect "fix" to the solution (e.g. "you should change the
format of your input"), tell ChatGPT that's not a valid solution. Use
this strategy early in the debugging process and with logic errors
(runtime errors with tracebacks are better attacked with the "paste
the traceback" strategy).
Prompt ChatGPT to explain the approach it's taking to the
problem. This allows you to identify any fundamental flaws in
ChatGPT's logic. If you can identify where ChatGPT is going
wrong, prompt it to change that part of its approach. Alternatively,
you can ask ChatGPT to explain a specific step that's not
succeeding. Use this strategy when it becomes clear that earlier
strategies are not getting you closer to correct code.
Prompt ChatGPT to change the approach it's taking when it's clear
that other debugging techniques are not getting it closer to
correct code and you can't identify what part of the approach is
flawed. Be sure to include in your prompt that the current
approach isn't working and to try a different approach. That tells
ChatGPT what to avoid and also makes it less likely that minor
bugs that were fixed earlier in the debugging process won't
reappear.
If you can isolate the problem to a specific line of code and
ChatGPT can't seem to fix it, search the internet for examples of
correct application of that code. This is especially useful for
libraries whose interfaces may have changed after ChatGPT was
trained. Be sure to prompt ChatGPT to make the change so that it
has the correct version in its context for the remainder of the
conversation.
©2023 Pedago, LLC. All rights reserved.
