Scribd
Upload
12 views3 pages

Hints - and - Tips - On - Using - ChatGPT 2023-10-31 07 - 04 - 09

Uploaded by

23092323d
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
12 views3 pages

Hints - and - Tips - On - Using - ChatGPT 2023-10-31 07 - 04 - 09

Uploaded by

23092323d
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Hints and Tips on using GPT 3.

5 and GPT-4

What is Generative AI?


Generative AI is a type of AI that creates new content based on what it has learned from existing
content. The process of learning from existing content is referred as model training and results
in the creation of a statistical model. When a prompt is given, generative AI will use its statistical
model to predict what expected response might be and generate new content.

What is ChatGPT?
ChatGPT is a generative language model developed by Open AI and released to public trial in Nov
2022. ChatGPT learns about patterns in language through training data, when given some text, it
can predict what comes next. Use cases include summarization, question / answering, grammar
correction, translation and code generation. The outputs of ChatGPT have been calibrated using
massive data to train the algorithms including random elements.

What is a ChatGPT prompt?


A ChatGPT prompt is additional information or sample answer provided to ChatGPT model to
facilitate the model to do a similarity search on its language model and generate appropriate
response. The prompt may be contextual information or your expectation on model in term of
processing logic or output format.

Tips on using GPT 3.5 and GPT-4


▪ Good prompt input leads to good output!
Always provide a simple prompt, stating purpose, background, boundaries, examples, and
desired output format

▪ Ask follow-up questions to improve responses


Unless you start a new chat or clear all your previous chat conversation, you can drill down
to more targeted content you are looking for by inputting following up questions. GPT 3.5
or GPT-4 model will reference all previous chat conversation context before come back
with further completion response

▪ Start a new conversation if you intend to switch topic or context of conversation


Start a new chat before continuing with a new topic not only optimizes your tokens
consumption, it also allows the model to focus on your new ask and provide a more
comprehensive response

▪ Always verify the accuracy of responses received by cross-checking with other sources
Limitations of the model output depend on timeliness and accuracy of training data used,
labelling error infused during algorithms tuning as well as extension of creativity adopted.
Best practice is to ask the model the reference source of information and cross check via
other channels

▪ Chat conversation history


Latest chats conversation will be saved in PolyU on premise database to allow staff /
student easy reference. You can use “delete chat history” function on menu to clear such
if no longer required. Chat history over past 24 hours will be cleared nightly. Staff /
student can export chat history to save a copy for retention if required

▪ Avoid input personal identifiable or confidential information


Always remove / replace identifiable personal information with make-up data before
input into chat conversation. Avoid sharing confidential information with GPT 3.5 or GPT-
4 to contain risks of exposing such to third parties

▪ Provide reference on information source if intend to include GPT 3.5 o GPT-4 output to
content of your deliverables for third party
When using content output from GPT 3.5 or GPT-4, always properly cite its source and
observe University policy on academic integrity

How to “chat” with GPT 3.5 or GPT-4


▪ Start the conversation with background, boundaries, and desired output

▪ Ask for definitions:


ChatGPT can be a helpful tool for looking up definitions of unfamiliar terms. Use the
phrase "What is" followed by the term you want to define. For example, "What is the
definition of cognitive dissonance?"

▪ Ask for examples:


Sometimes it can be helpful to have examples to illustrate a concept. Use the phrase "Can
you give me an example of" followed by description of concept you need an example. For
example, "Can you give me an example of a homogenous mixture?"
▪ Ask for explanations:
If you're struggling to understand a concept, use ChatGPT to ask for an explanation in
simpler terms. For example, "Can you explain the concept of endogenous growth theory
in simple terms?"

▪ Be clear and specific, set the tone and style, tailor the prompt’s format and tone to match
your goal and get the desired responses. Style may be academic, instructive, journalistic,
critical, creative, conversational, or professional.

▪ Define the role and expertise. Tell GPT 3.5 or GPT-4 model to assume the identity of a
profession (like a marketer, programmer, coach, professor, or teacher etc.). For example,
“Tell me about marketing.” is not specific enough, while “Being a marketer, explain five
important digital marketing strategies that a small business should implement to increase
online visibility and drive sales, in a professional tone.”

▪ Provide feedback:
If GPT 3.5 or GPT-4 provides an inaccurate or unhelpful response, provide feedback to
help improve its responses in the future. This will help ensure that GPT 3.5 or GPT-4 is a
helpful resource for your learning needs.

▪ Experiment with question phrasing. Rephrasing questions and testing multiple variations
will help you find the best possible prompt. For example, “How to improve student
engagement in class?” is not as effective as “What are the pain points encountered by
students during class and actions that can mitigate students’ pain points?”

Limitations and risks in using ChatGPT


• Large language model with long-form setting have answers rarely completely wrong. Mix
of correct and wrong / misleading information (reasoning errors) resulting from
hallucination problem is reduced as model maturity and tuning advanced but not yet fully
eliminated.

• Can produce output that represents social biases and worldviews derived from its training
data.

• Large language model uses massive training data from various sources, intellectual
property protection and creator rights remain unclear.

You might also like