Is ChatGPT Good At Data Prep?

Don't forget to click Subscribe if you want more weekly data-related updates!
Will Sutton wants to know if his AI co-pilot can help him prep data. To test this, he gives it 3 Preppin Data challenges. See how Will works with ChatGPT (using Advanced Data Analysis plugin) to read csv files and output cleaned up and summarised data...and learn where ChatGPT fails and you need to triple check it's answers.
📌Timestamp
00:00 Intro
00:13 Preppin Data challenge 1
03:49 Preppin Data challenge 2
07:40 Preppin Data challenge 3
12:10 ChatGPT to format data for a network diagram
11:10 Takeaways: What words
11:56 Takeaways: What to watch out for
🔗Links
- Will's ChatGPT for Tableau GitHub repository: github.com/wjsutton/ChatGPT_f...
- Do a Preppin Data challenge yourself: www.preppindata.com/
- Will uses Visual Studio (VS Code) for all his coding: code.visualstudio.com/
- Check out Will's Tableau Public Profile: public.tableau.com/app/profil...
- Connect with Will on LinkedIn: / will-sutton-14711627
- Follow Will on twitter: / 2fwjsutton12
- Subscribe to Will's UA-cam channel: / @wjsutton
🎧Want to dive deeper into AI+Data? Check out our AI + Data Viz playlist
▶️ • AI + Data Viz
----------------------------------------------------------------------------------------------------
Want more information about us? Head over to theinformationlab.co.uk
Want to learn more about the Data School? Head over to www.thedataschool.co.uk
----------------------------------------------------------------------------------------------------

[@timfngwena]

I wonder if the issue here is that chatGPT requires good prompts to guide it through the steps rather than a start and end. Would be interesting if you asked it to do the analysis that required the transformation and how that performed.

[@theinformationlab]

Good idea. Definitely something we should try. Also absolutely love all your videos and exploration using chatGPT for both analysis and building your own personal gpt.

[@carlkim2577]

Nice demo. I learned a lot. However, you are expecting too much from a language model. Until we get to AGI, we can't give input and output only. You have to lead it through the intermediate steps. In fact, in AI it's called Tree of Thought processing. When you ask the AI to take it step by step and explain each step, the results are qualitatively better. I mean that this process is built into the model. In your demo, you should manually lead the AI through the process and it will give you what you want. Still, this is even better than having a data engineer work with you. It listens to you and no whining about being hung over and a break up with a gf. Just gets the job done.

[@theinformationlab]

Thank you so much for your suggestion and help with how you can get current LLMs to work better by taking a step-by-step approach.