Top 10 Rookie ChatGPT Mistakes I Used To Make
Here are my top 10 rookie mistakes I made when starting out with ChatGPT. Thanks to online trainings, including “ChatGPT and Data Analytics” by Maven Analytics, I learned the error of my ways. Don’t be like me. Learn these rookie mistakes early, so you can avoid them!
Mistake 1: Making your prompt too broad.
Be as specific as possible about the output you want, including variable names, data types, cell ranges, and any other context that would be helpful.
Example of a “bad” prompt: “Can you make a mock sample spreadsheet with dummy data?”
Example of a “good” prompt: “Can you make a mock data sample Excel spreadsheet with columns for student name, employer, and salary? List 10 rows of mock data. The employers should be Bain, Boston Consulting Group, McKinsey, Amazon, and Google. Salaries should be in the range of $100 to $100,000.”
2 important notes about this dialog with ChatGPT:
- I meant to say “Make a mock data sample spreadsheet with COLUMNS for student name, employer, and salary.” ChatGPT still understood what I meant.
- Some of the salaries ChatGPT listed in the mock dataset are ABOVE the range I specified of $100 -$100,000. If this happens, you can prompt ChatGPT to correct its output with a prompt like: “Some of those students have salaries above the range I specified. Can you try the prompt again and make sure no salaries are above $100,000?”
Mistake 2: Not double-checking the accuracy of ChatGPT output, whether it is code, a math equation, or a fact.
· More complex queries (including SQL queries) have a higher likelihood of error.
· You need to have the analytics knowledge and the foundational knowledge of the tool, whether it’s Python, Excel, Google Sheets, SQL, etc.
· If it’s code, copy & paste the code into the software to make sure it runs without errors.
· If you are not very familiar with the software, you can double-check to see if Bard, Bing AI, or another AI tool gives you a different output.
Mistake 3: Copying & pasting text output from ChatGPT without editing it to make it sound authentically like your voice.
I’ve heard from many recruiters that they can tell whether a cover letter has been crafted using ChatGPT. I think it’s still ok to use ChatGPT to help with marketing, communications, or other text output, but you must make sure it sounds like a human, not a robot!
Mistake 4: Not understanding how large language models (LLMs) work, as well as their limitations.
To understand how to work with LLMs, you need to know what they are and what they can and cannot give you.
LLMs like ChatGPT are generative models focused on producing text prompts.
For example, if you ask ChatGPT to fill in the blank: “The capital of Virginia is ___”, the model doesn’t know that “Richmond” is the correct response. It suggests the answer it thinks is most probable in your given context.
ChatGPT compared “Richmond” with thousands of other words and determined that it had the strongest relationship with words like “capital” and “Virginia.”
As part of understanding how LLMs work, you must understand that they have certain pitfalls:
· LLMs can “hallucinate”, presenting wholly invented language as if it were facts.
· LLMs may provide incorrect or inefficient solutions.
· LLMs don’t have common sense or human judgment. They may miss context that seems obvious to you.
· LLMs often provide broad text output and don’t know specifics.
When using an LLM like ChatGPT, remember that YOU are responsible for verifying any outputs. Test out any code or analyses that ChatGPT provides to make sure it works AND that it makes sense.
Mistake 5: Not supplementing ChatGPT knowledge with Bard, Bing AI, Grammarly, or any of the other AI tools available.
If nothing else, at the time of this writing, ChatGPT doesn’t have information after Sept 2021, so Bing AI or Bard would be a better source for accurate up-to-date information. Bing AI and Bard also provide citations so you can verify the accuracy of their sources.
For example, I asked Bard to give me a list of 5 yoga poses to get rid of gas (don’t ask!)
Bard gave me a list of 5 poses and links to articles so I can easily click to read more about each pose.
Mistake 6: Not assigning ChatGPT a persona.
You will get better and more detailed results if you ask ChatGPT to behave like a specific persona because it will have a better understanding of what you expect from the output.
e.g. “I’m a new Business Systems Analyst with limited SQL experience. You’re an experienced SQL user who is helping me comment on code…”
“I want you to act like an interviewer at Google who is interviewing me for a Data Analyst position…”
Mistake 7: Not realizing that ChatGPT can have back-and-forth conversations, so you can build on a prompt.
You don’t have to get it right the first time! Modify your prompt and use these phrases in your conversations with ChatGPT:
“Can you explain XYZ to me like I’m a child?”
“Can you say the above in a more concise format?”
“Can you edit the above text to be between 200–300 words?”
“I’m not sure that’s correct. Are you sure that makes sense to you?”
“Is there a simpler Excel formula I could use as a beginner user?”
Mistake 8: Not cleaning up your ChatGPT conversations in the menu on the left side of the screen.
Delete old conversations you don’t need anymore and change the titles of helpful conversations that you know you’ll refer to again and again. For example, if you just know that you’ll forget that Excel code that you used for that project!
Mistake 9: Mixing all your topics in a single chat session.
Since ChatGPT references the rest of your chat session, changing topics can mess up your results. Each time you start a new topic, create a New Chat. This also makes it easier to refer to specific chats later if the same questions arise again.
Mistake 10: Not learning more!
Being a data analyst is signing up to be an eternal student. There are a ton of resources, including:
· Coursera course on Prompt Engineering with Jules White
· Maven Analytics course on ChatGPT and Data Analytics
· Coursera course: DeepLearning.AI courses with Andrew Ng
—
I hope you enjoyed my top 10 rookie mistakes that I used to make in ChatGPT. Don’t repeat my mistakes — make your ChatGPT prompts specific, know GPT’s limitations, and always keep learning!
