Eugene Mimlin/Getty Images
You know what’s fun for me? Charts and graphs.
I get irrationally happy when I get a cool graph to dig into. I love watching the news on election night, not for the vote count, but to see the awesome graphs. I’ll flip through all the channels all night to see all the ways each network finds to present numerical data.
Is that weird? I don’t think so.
Also, the moment I realized that ChatGPT Plus would revolutionize my business.
After all, ChatGPT excels at creating graphs and tables, and this ubiquitous generative AI chatbot can synthesize a ton of information into something graph-worthy, so what ChatGPT lacks in aesthetics it more than makes up for in informational value.
It should come as no surprise to anyone that the feature set of AI chatbots is constantly changing. As of this update (end of May 2024), OpenAI has released a Mac application and GPT-4o LLM available to both free and paid customers. The GPT-4o version that comes with the Plus version for an extra cost will have interactive charting capabilities and the ability to interact with the engine for longer per session.
But not by much: My free account doesn’t offer GPT-4o at all yet, it hasn’t been rolled out to all free accounts yet, and the paid ChatGPT Plus plan offers the interactive charts feature in Chrome and Safari, but not in the Mac app.
Also, ChatGPT vs ChatGPT Plus: Are the paid subscriptions still worth it?
This article was last updated when advanced data analysis features (including charts) were available only to Plus customers. Some of these features should also be available to free customers, but because my free account does not yet have those features, the remainder of this article will describe the charting features as if they were only available to Plus customers. If you are a free customer and have GPT-4o, feel free to try out some of the prompts. These features may be useful to you, and will undoubtedly be useful over time.
Advanced data analysis produces some relatively ugly graphs. But these are nice graphs. First, let me explain where ChatGPT gets its data from, and then I’ll create some tables.
How to create graphs and tables using ChatGPT
Earlier we discussed which charting tools are available in which version of ChatGPT. But there’s more to it than just charting tools. To use ChatGPT productively, you need to understand what the different editions can do.
ChatGPT Free Version: This version has so far used the GPT-3.5 Large Language Model (LLM), which is not as capable as the GPT-4 version. As of May 2024, the GPT-4o LLM is also available to some free users and is being gradually rolled out.
ChatGPT Plus: ChatGPT Plus is the commercial, full-powered version of OpenAI’s ChatGPT. Currently, ChatGPT Plus offers three main selection options per session: GPT-3.5, GPT-4, and GPT-4o. It previously offered a plugin, but this has been replaced by Custom GPT.
GPT-4 and GPT-4o versions now include DALL-E 3, Bing web access, and advanced data analysis. Some users have reported issues using Bing for web access. Most of what we do uses an advanced data analysis component. Even without Bing web access, GPT-4 and 4o report that their training data now includes information up to December 2023.
Also, what does GPT stand for? Understanding GPT 3.5, GPT 4, GPT-4o, and more
For the majority of this article, we use the GPT-4 optional Advanced Data Analysis component. This tool imports data tables in a variety of file formats. There is no specified size limit for the imported data, but it can handle fairly large files, but it will fail once the file exceeds an undefined level of complexity.
We will keep you updated with more information as changes are made to ChatGPT Plus, but for now, let’s take a look at how to create cool graphs.
ChatGPT Enterprise: The Enterprise edition also offers advanced data analysis and plugins. Files uploaded to Enterprise are kept confidential. Enterprise also allows for larger files and larger responses. Pricing is TBA.
Let’s start with an example. The following demonstration covers the top five cities in terms of population.
List the top 5 cities in the world by population. Include countries.
When I asked the free version of ChatGPT this question, I got the following response:
Screenshot by David Gewirtz/ZDNET
Converting that data into a table is easy: just tell ChatGPT that you want a table.
Make a table of the top 5 cities in the world by population, including countries.
Screenshot by David Gewirtz/ZDNET
You can manipulate and customize the table by giving ChatGPT more detailed instructions. Again, using the free version, I add a population count field. Of course, that data is out of date, but it still displays.
Create a table of the top 5 cities in the world by population. Include country and population fields.
Screenshot by David Gewirtz/ZDNET
You can also specify details for your table, such as field order and units. Here, we’ve moved the countries first and compressed the population counts.
Create a table of the top 5 cities in the world by population. Include country and population fields. Display the fields in this order: rank, country, city, population. Display population in millions (1 decimal place), i.e. 37,833,000 will be displayed as 37.8M.
Note that I gave the AI an example of how I want the numbers to appear.
Screenshot by David Gewirtz/ZDNET
That’s it for the free version. From now on, I’ll switch to the ChatGPT Plus version, which costs $20 per month.
Enabled with advanced data analytics, ChatGPT Plus allows you to create line charts, bar charts, histograms, pie charts, scatter plots, heatmaps, box plots, area charts, bubble charts, Gantt charts, Pareto charts, network diagrams, Sankey diagrams, choropleth maps, radar charts, word clouds, treemaps, and 3D charts.
In this example, we’ll create a simple bar chart.
Create a bar chart of the top 5 cities in the world by population
It’s a chatty little tool.
Screenshot by David Gewirtz/ZDNET
Those with a keen eye may have noticed a population discrepancy between the previous table and the results here. Notice that there are green icons in the table and purple icons in this graph. We jumped from GPT-3.5 (the free version of ChatGPT) to GPT-4 (ChatGPT Plus). It is interesting to see that the data for the different LLMs differ slightly. This difference is part of why you need to be careful when using AI, so double-check your work. In our case, we are just showing you the chart, but this is a concrete example where confidently presented data can be wrong or inconsistent.
One great feature of Advanced Data Analytics is the ability to upload a dataset. In this example, I downloaded the “Popular Baby Names” dataset from Data.gov. It’s a comma-separated file of baby names in New York City from 2011 to 2014. It’s 10 years old, but fun to play with.
The datasets I chose for this article are easily available from government sites, so you can replicate this experiment yourself. There are lots of great datasets on Data.gov, but I found many of them to be too large for ChatGPT to use.
Also, how to create an app using ChatGPT
When I downloaded this, I noticed that it also included information about ethnicity, so I could run a variety of graphs from the same dataset.
Click the small upload button and specify the data file you want to import.
Screenshot by David Gewirtz/ZDNET
To learn more about the file’s format, we requested to see the first five lines of the file.
I was curious to see how the dataset was distributed by gender name. The first prompt is:
Create a pie chart to show the gender distribution across the entire data set
And here is the result:
Screenshot by David Gewirtz/ZDNET
Unfortunately, the dark green color makes the numbers hard to read. Fortunately, you can tell Advanced Data Analytics to use different colors. I was careful to choose colors that didn’t reinforce gender stereotypes.
Create a pie chart that shows gender as a percentage of the entire dataset, using light green for males and medium yellow for females.
Screenshot by David Gewirtz/ZDNET
As we saw earlier, the data collected includes ethnicity. Here’s how to see the distribution of different ethnicities recorded in New York in the early 2010s:
Use a pie chart to show the distribution of ethnicities in a dataset, using only light colors.
Here are the results: Did you notice anything?
Screenshot by David Gewirtz/ZDNET
Apparently, New York did not properly normalize its data: it used “White non-Hispanic” and “White non-Hispanic” together, “Black non-Hispanic” and “Black non-Hispanic” together, and “Asian and Pacific Islander” and “Asian and Pacific Islander” together, resulting in an inaccurate representation of the data.
One advantage of ChatGPT is that it remembers your instructions throughout the session, so I was able to give instructions like this:
For all the following requests: Group “WHITE NON HISPANIC” with “WHITE NON HISP” Group “BLACK NON HISPANIC” with “BLACK NON HISP” Group “ASIAN AND PACIFIC ISLANDER” with “ASIAN AND PACI” When displaying ethnicities, use the longer of the two ethnic names.
And I replied:
Screenshot by David Gewirtz/ZDNET
Let’s try the graph again using the same prompts.
Use a pie chart to show the distribution of ethnicities in a dataset, using only light colors.
That’s good:
Screenshot by David Gewirtz/ZDNET
You should look carefully when viewing the results: for example, in a request for top baby names, the AI distinguished between “Madison” and “MADISON” as two different names.
Screenshot by David Gewirtz/ZDNET
In all requests below, baby names are case insensitive.
Finally, let’s try to create a complex chart from a single prompt. The prompt is:
For each ethnicity, display two pie charts, one for each gender. Each pie chart lists the top five baby names for that gender and ethnicity. Use only light colors.
In the end, the chart produced text that was too small to read, so to get a more usable chart, we export it again, specifying both the file format and the file width.
Export this graph as a 3000 pixel wide JPG file.
Screenshot by David Gewirtz/ZDNET
And here is the result:
Screenshot by David Gewirtz/ZDNET
Notice how Sofia and Sophia appear as two different names, even though they are both very popular. But that’s what makes the chart so interesting.
FAQ
How much does it cost to use Advanced Data Analytics?
Advanced Data Analytics comes with ChatGPT Plus. Some features are available in the free version of ChatGPT, GPT-4o. ChatGPT Plus costs $20 per month. Advanced Data Analytics is also included in the Enterprise edition, but pricing has not yet been announced.
Will the data uploaded to ChatGPT for charting be kept private or is there a risk of data leakage?
Assume there will always be a privacy risk.
I asked ChatGPT this question and received the following response:
Data privacy is a top priority for ChatGPT. Uploaded data is used only for the purpose of the user’s current session and will not be stored long term or used for any other purposes. However, for sensitive data, users should always exercise caution and consider using the enterprise version of ChatGPT with enhanced data confidentiality.
Plus: Generative AI brings new risks to everyone. Here’s how to stay safe
My recommendation: don’t trust ChatGPT or any generative AI tools, and while the enterprise version should have stronger privacy controls, I would recommend only uploading data you don’t mind being made public.
Can ChatGPT’s advanced data analytics handle real-time data, or is it more suitable for static datasets?
Yes, but there are some practical limitations. First, Plus accounts limit the number of requests you can make in a given period of time. Second, you need to upload each file individually. You can also use the licensed ChatGPT API for real-time analytics, but for the chatbot itself, you will be parsing stored data.
You can follow my daily project updates on social media: subscribe to my weekly update newsletter on Substack and follow me on Twitter. David GewirtzFacebook (Facebook.com/DavidGewirtz), Instagram (Instagram.com/DavidGewirtz) and YouTube (YouTube.com/DavidGewirtzTV).
