AI agents can make data visualizations now. Ask Manus, Claude, or ChatGPT to chart a dataset and you'll get something back in seconds. The output is almost always... fine. Functional. Generic. The kind of chart that communicates data but doesn't communicate insight. In this post, I experimented with how to make AI agents visualize data better.
In November 2025, AI coding tools went from “halting and clumsy” to surprisingly capable. Suddenly they could produce whole, working apps from minimal instruction, in ways they simply couldn't before.
This experiment makes that shift visible. We gave 22 AI models the same exact prompt to build a working analog clock from scratch, then ran each model five times independently.
The native GitHub IPython Notebook viewer also doesn't support within-notebook linking, which can be fairly annoying when you spend an hour putting together a nice table of contents system.
I originally didn't smooth the data, but it looked way too jagged and crappy for many of the names. Ultimately, I erred on the side of aesthetics and sacrificed the ability to look up exact numbers in the graph. Rest assured, though, if you want the exact numbers you can download the underlying data of every graph with one of the links under the graph.
Thank you for the kind words. :-) Although that app you linked is a bit different - it's showing when names were popular over time. This app goes one step further and calculates how many of those people are still alive and makes a best guess at how old someone is based solely on their first name.
An R^2 of 0.33 is a fairly decent correlation for a study involving human subjects. It's obviously not perfect, but the authors are not claiming that it's perfect.
Some of the trends in Part 1 are not statistically significant (e.g., "never attend church" vs. "sometimes attend church"), but I presented them regardless because they have a pretty clear trend. I believe p-values are only presented in comparison to the reference point.
That's a great idea! I've heard of researchers doing this kind of analysis to detect "hidden gems," i.e., papers that influenced papers that ended up becoming the big cited paper. The tough part, of course, is that most journals are keen on keeping all of their data behind company walls. Perhaps PLoS would be an exception, though...
