Data Visuals Gone Bad: Avoiding Common GPT-4 Prompting Pitfalls

Author(s): John Loewen, PhD

Originally published on Towards AI.

GPT-4 prompting tips for best-practices data visualizations
Dall-E 2 image: impressionist painting of a person sitting at a computer visualizing data

ChatGPT has greatly simplified the process for analyzing and visualizing datasets.

With the Python pandas and matplotlib libraries, GPT4 can visualize your charts, graphs, and maps on the fly without blinking an eye.

Boom, there they are.

But they don’t always look optimal — in fact, sometimes they look like shit.

And that can make you very frustrated — frustrated enough to start cursing out the tool that you are trying to use to help solve the problem.

So can we get this “lousy-intern” AI tool back on track, or better yet, just keep it on track?

Here are some common pitfalls and some tips on how to avoid them.

Recently, I was working on a development dataset, and I was prompting GPT-4 to get some ideas on what visuals might be useful.

Prompt to GPT-4: Using the PDF file provided, please come up with some interesting and informative data visualizations that showcase this data.

In its infinite wisdom, GPT-4 generated this shit-show:

GPT-4 shows a lack of common sense on data visualizations

Response from GPT-4: This visualization provides a clear comparison of literacy rates across different countries.

WTF?Really?

Quantitatively accurate, yes. Clear and/or useful? Not so much.

And another beauty… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI