DEV Community

Cover image for You Are Using ChatGPT Incorrectly!
Emma Ham
Emma Ham

Posted on

You Are Using ChatGPT Incorrectly!

What an accusation! Boo!
I know it might seem a bit strong for a blog post title, but let me explain.

In my defenseπŸ˜…

I've encountered this question at one of the recent tech conferences I attended, where an audience member asked the panel speaker,

"Will ChatGPT replace our jobs?"

As common as this question has become in the era of AI technology, I was quite impressed with the speaker's response that day.
The individual stated,

"No, we won't necessarily lose our jobs to ChatGPT, but we might lose them to someone who knows how to use ChatGPT better."

Ding! This made me question whether I've been utilising ChatGPT to its full potential effectively.

So today, I'm here to admit that I've been underutilising ChatGPT and to introduce some ways to use it more effectively, which I'll share with you.

In this post, we will delve into how to make the best use of ChatGPT and other AI-powered generative models as engineers in our daily lives.
To achieve this, we'll explore some of the most popular and useful prompt patterns that developers can employ.

Stay tuned!
Some of these may be techniques you're already using without realising, or ones you can take advantage of in the future.

But first,
let's take a moment to consider an interesting research finding from Purdue University.

An interesting research

Research

According to a recent study conducted by Purdue University, when ChatGPT was presented with around 517 programming questions from Stack Overflow, it provided answers with inaccuracies or non-functional code 52% of the time. The study also revealed that ChatGPT's performance in terms of correctness, comprehensiveness, consistency, and usefulness lags behind that of Stack Overflow.

hollucinate

Additionally, many of us have likely experienced instances of AI "hallucinating," where it produces convincing but incorrect answers.
This leads us to believe that we have the power to validate AI-provided answers.
To do so effectively,
We need to know how to ask the right questions and frame them properly.

After reading this article, I wasn't surprised at all because my ChatGPT has been apologising a lot for providing incorrect answers when I've called it out, which can be quite amusing.πŸ˜‚

ChatGPT apology screenshot


4 Prompt Patterns for today

Setting jokes aside, let's now dive into four prompt patterns that can help make ChatGPT's responses more meaningful, accurate, and aligned with the intent of our questions. Let's start with some standard patterns.

πŸ‘‰ Pattern 1

Persona

First up is the "Persona Pattern". Using this pattern, we can guide ChatGPT to produce responses tailored to a specific viewpoint or perspective. This is particularly useful for code reviews, architectural advice, tech discussions, and ideas commonly required in our professional field.


πŸ‘‰ Pattern 2

2Recipe

Next is the "Recipe Pattern". As the name suggests, this pattern provides step-by-step guidelines based on the question, making it ideal for situations where you have a general idea of what you want to achieve but need detailed instructions.

For instance, a question like, "Can you provide me with a step-by-step guide for releasing a mobile app to production, including the necessary steps for submitting it to the store for review?" would be an excellent use case for this pattern.

Example answer


πŸ‘‰ Pattern 3

3

The "Reflection Pattern" is the third pattern, and I consider it one of the most important patterns for daily use as a developer. It commands the AI to provide answers with clear reasoning, making it invaluable for spotting instances of AI "hallucination" and validating responses.

This pattern becomes essential as we increasingly rely on AI in our daily work, potentially leading us to lose our grasp of the truth and its meaning.

So highly recommend,
Use this pattern and then start asking questions!


πŸ‘‰ Pattern 4

4

Lastly, my all time favourite, we have the "Flip Interaction" Pattern.

This pattern is all about letting AI take control of the conversation, a creative approach when you know what you want to achieve but aren't sure what information ChatGPT needs from you.
In this scenario, the AI inverts roles with you, asking questions for you to answer. It's an exciting way to explore a topic.

With this pattern, you won't need to worry about asking the right questions; ChatGPT will guide you by asking the questions it needs answers to in order to address your original query correctly.


πŸš€πŸš€ That wraps up today!πŸš€πŸš€

Last

In today's developer landscape, it's increasingly likely that we'll turn to AI tools like ChatGPT or Bard in our daily work. AI technologies have made tremendous strides and have a significant impact on both our personal and professional lives. We are in a world where it's crucial to equip ourselves with the right skills and a deep understanding of what we have at our disposal to make the most of it.

I hope this post helps you feel more confident in using ChatGPT effectively in the future, and I'm open to creating more AI-related posts if there's interest. Thank you!

Top comments (30)

Collapse
 
shawn2208 profile image
Shawn2208

The recipe pattern I completely agree. Like I didn’t know where to start when creating a pseudo code to JavaScript interpreter and ChatGPT gave me specific steps and the pattern through out to follow. I also use it if I don’t understand a JS concept. Like closures! 😟 but chatGPT dumbed it down now I understand.

Great article by the way!

Collapse
 
pengeszikra profile image
Peter Vivo • Edited

You are absolutley right, the good task is cruical for talking with AI, also good testing other AI like LLama with same task to check the difference between version.

My usecase is interactive story telling which is also genereta images next to answer. It is also try to explore how can handle chatGPT4 the longer conversation, and which points made a mistakes.

Collapse
 
ham8821 profile image
Emma Ham • Edited

Very good insight. I have been meaning to look into LLama for a while actually, thanks for the comment, Peter!

Collapse
 
michaeltharrington profile image
Michael Tharrington

Really awesome and well-written post, Emma!

I think you've done a great job at explaining practical ways to approach using ChatGPT as a dev... I like the different patterns you've highlighted and your emphasis on trying to get accurate answers vs. hallucinations.

You're right that this tool is very impactful and as polarizing as it may be, it's a good idea to understand how it works.

Collapse
 
ham8821 profile image
Emma Ham

Thank you MichaelπŸ™ƒ

Collapse
 
lipelourenco profile image
Felipe Lourenco

Cool! Another thing that might help is to write "Take a deep breath before answering". Strangely enough, people seem to be getting better results with that. Just empirical confirmation for the moment, as we donΒ΄t have any data to support that claim, but itΒ΄s curious.

Collapse
 
ham8821 profile image
Emma Ham

That is such an interesting information! I actually found this article to support this which indicates that using a prompt like "Take a deep breath before answering" does almost always bring out better answers along with prompt like "let's think step by step".

They explain why this might be happening in this article fyi!

arstechnica.com/information-techno....

Collapse
 
lipelourenco profile image
Felipe Lourenco

Thank you!

Collapse
 
yaireo profile image
Yair Even Or

It's the stupidest question to ask if chatGPT will replace jobs. It's just one tiny insignificant AI bot. Like asking in 1996 if MS Paint will replace painters (artists).

It is a first step in a chain of software (AIs) and eventually a company & product will rise with an AI solution which will for certain replace any brain-related job.

Any work that is brain-based is the easiest to replace, unlike manual work which involves most of human senses, programming does not technically require any sense. not even vision, as a blind person can also code, although nothing that produces visual output probably.

ChatGPT will not replace any jobs, just like Bitcoin is just one coin out of many, with huge inherit problems and has not proven itself for over a decade.

There is word that Google Gemini will far superior from OpenAI's products and is a good candidate for replacing jobs.

Google itself will benefit from this very much as a ton of their capital goes into R&D teams and an AI that can replace most of its workers is a huge huge financial prize for them.

Collapse
 
ham8821 profile image
Emma Ham • Edited

Agreed. Will ChatGPT replace all jobs? Very unlikely. But will it replace some jobs? Absolutely. In fact, there are already documented cases in the state.

This brings to mind a few years ago when Amazon introduced cashier-less, walk-in, walk-out supermarkets in the state. People were concerned that it would take away jobs from those who made a living as cashiers. Did it replace all cashier jobs? No, but is it likely to replace more going forward? Probably.

Beyond AI, we're witnessing a linear progression in technology, sparing us from some manual tasks compared to 5 or 10 years ago. Even in the programming world, functions that once required manual implementation and duplication are now often integrated into languages or frameworks.

At this stage, AI stands as another transformative technology, steering us away from manual and repetitive hassles, serving as a valuable tool and assistant. While we wait to see what more it can bring into our world, Gemini is also an intriguing project to monitor in that regard.

On the flip side, we should also consider the potential jobs that AI might "create." As the ethical implications of AI take center stage, could this lead to new roles? Perhaps guiding those entering programming world post-AI arrival to obtain a deeper understanding of morals, conduct, and crucial concepts like copyright. As a developer, I'll be keeping a close eye on these developments!

Collapse
 
muhyilmaz profile image
MUHAMMED YILMAZ • Edited

`


No I am not.

`

Collapse
 
rolfstreefkerk profile image
Rolf Streefkerk

AI hallucinations will happen regardless of your "prompt patterns", it's a side effect of the limitations rather than "we didnt instruct it properly".

Collapse
 
ham8821 profile image
Emma Ham • Edited

That is pretty much what I intended to say in the article! If hallucinatoons are inevitably happening and not something we can control, we shall make the use of some of the helpful prompts to help ourselves to "spot" when it happens so that we know a bit more clearly of what we are provided with by AI:)

Collapse
 
rolfstreefkerk profile image
Rolf Streefkerk

yea I understand that's what you're trying to do, but from my experience it won't make a difference since it's a limitation of the technology it will always create erroneous data at this moment in time. The way to find out is validating it yourself by knowledge, perhaps asking ChatGPT to spot any issues or the old fashioned "google it"

Collapse
 
taqui_786 profile image
TAQUI ⭐

Nice Post πŸ‘

Collapse
 
ham8821 profile image
Emma Ham

Thank you!

Collapse
 
michaelsynan profile image
Michael Synan

Don't get upset with ChatGPT when it doesn't do what you want :)

dev.to/michaelsynan/getting-big-ma...

Collapse
 
sip profile image
Dom Sipowicz

I would add some more initial settings and GPTs. Here I show off my own personal ChatGPT setup:

Collapse
 
wisdomudo profile image
WISDOMUDO

Insightful. Thanks for sharing!

Collapse
 
best_codes profile image
Best Codes

Very enlightening! I use codeium, I don't have an OpenAI account, but I use it though my bots on Zapier and through Opera's in browser AI, Aria. This will be really useful for me, thanks!

Collapse
 
magnording profile image
Magnus Nording

Yes the Flip Interaction really worked for me, making the AI asking me to show certain methods and together we achieved our goal. Thanks, important article.

Collapse
 
ham8821 profile image
Emma Ham

Yes my favorite pattern to use as well it is such an engaging/interesting way to utlise ai capability!

Collapse
 
evillord666 profile image
Ushakov Michael

Correct way of using ChatGPT is to throw it to a trashbean

Collapse
 
virtualmachine profile image
ByteCodeProcessor

Good article. This shows chatGPT is not a general and/or democratic tool.