Photo by BoliviaInteligente on Unsplash
I provided the same JSON file to the Data Analytics ChatGPT plugin with a slightly different prompt. I did not use the word words/topics to avoid the possibility of getting a one-word topic response, a challenge I encountered on my first try.
In this endeavor, I would also recommend my approach to ChatGPT, rather than always saying “yes” and leaving it to ChatGPT.
Give us your ideas on:
– What are the top 10 topics based on this conversation?
– How many times the topic was mentioned
– Label each conversation as an action: e.g. ask for concept, summary, idea, etc. This is just an example, the labels are not limited to these
result
After the first prompt,
Data Analysis The ChatGPT plugin gave us the answers we expected. This accuracy, achieved without any fine-tuning, indicates a significant improvement in the plugin’s performance.
So, adding more content to the list of pros.
Strong Points:
Good and fast engineering can lead to good results
Here’s a breakdown of the results…
The first part of the first response demonstrates the ability to self-correct without me explicitly saying “fix it.”
Response Part 1, Image by Livia Ellen
The image below shows the second part of the response, answering objectives #1 and #2. It is noteworthy that in this attempt, the topics identified by the plugin are more granular and detailed compared to the one-word topics generated in the first attempt.
Response Part 2, Image by Livia Ellen
The final part, the Data Analysis ChatGPT plugin, addressed goal 3 nicely (labeling actions in each conversation and counting the number of times they occur).
Response Part 3, Image by Livia Ellen
Data Analysis I am intrigued by the response from the ChatGPT plugin. So far, its performance is a data analyst’s dream.
Let’s move forward!
Give us your feedback…
I intend to ask ChatGPT to fix the classification of the “General/Other” category to label it correctly.
Feedback Response, Image by Livia Ellen
The data analysis plugin was able to fine-tune the classification function through this experimentation.
Performance is good so far, so let’s export it to a Jupyter Notebook.
In this step, we test the plugin’s ability to seamlessly integrate its output into a Jupyter Notebook format for further exploration. As a data analyst, I want to test it, so I need some code.
Data Analysis Plugin Challenge, Image by Livia Ellen
It looks like a series of errors have occurred, and it seems like the notebook generation process is failing due to a token limit issue.
I told ChatGPT to force stop the process.
Cons:
The plugin will keep running the process until it is complete unless you force it to stop, and it seems to lack any built-in break logic or mechanism to automatically stop it when a critical issue occurs.
Following this experience, I provided feedback to ChatGPT highlighting the need for this issue to be addressed and addressed.
First notebook output, image courtesy of Livia Ellen
The Jupyter Notebook was generated successfully, but when I downloaded it, there was a middle section of code missing that I needed to take care of.
Improved notebooks, image courtesy of Livia Ellen
Plugin Credits:
Data Analysis ChatGPT plugin admits its mistake. Morally good GPT! — Highlighted in blue box in image
The generated Jupyter Notebook still has some code missing. So I gave a pep talk to ChatGPT to motivate and improve the plugin towards a more effective solution. Data interns need this kind of encouragement too 😉 Hahaha…
and…
Oops, another error!
Honestly, I’m tired of debugging and asking questions…
I realized there was still a learning curve with this plugin. The most annoying thing was not knowing best practices and not knowing when to stop working, so I had to ask questions and send reminders. The cycle of debugging and providing reminders was exhausting.
