Is GPT-4 Getting Any Better At Quantitative Image Analysis?

Author(s): John Loewen, PhD

Originally published on Towards AI.

Can it do decent quantitative analysis from a data visualization?

This member-only story is on us. Upgrade to access all of Medium.

For me, one of the most useful GPT-4 tools is the ability to analyze and interpret image data.

But how good it this tool now with charting data and with map images?

I wrote an article on this last year:

Prompting GPT-4 for CSV data from line and bar chart images

pub.towardsai.net

With GPT-4o, there have been some pretty big changes — the big question for me is has quantitative analysis of chart and map data improved?

Armed with our GPT-4o prompt, let’s see if GPT-4o has improved in its ability to provide quantitative analysis from chart and map images.

For the first attempt, let’s take a look at a line chart.

The original chart representation, created using Python (from a UN data set on Global Happiness), looks like so:

Top 5 countries for “Global Happiness” (2015–2021)

Let’s save the file as a .PNG image, upload it to GPT-4 and ask it to give us its interpretation of the image.

Response from GPT-4: This line chart is titled “Happiness Scores of Top 5 Countries (2015–2021)” and plots the happiness scores of Finland, Denmark, Switzerland, Iceland, and Norway over a period from 2015 to 2021.

GPT-4o also provides overview of the range,… 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