AI models are trained on data, and better data helps make better models. Likewise, it’s hard to develop worthwhile models with bad data. So what happens when it grows more difficult for AI-based bots to scrape the web and more of the available data is generated via AI in the first place? For NYT’s the Upshot, Aatish Bhatia imagines a feedback loop of numbers and letters that grow fuzzier with each generation.

Natural diamonds require a lot of pressure and time, and then someone has to mine for them. Lab-grown diamonds can be produced to be nearly indistinguishable from the natural ones, minus the time and mining. For Works in Progress, Javid Lakha describes the process and the growing cost gap between natural and the cheaper lab-grown.

As the value of diamonds inevitably falls, do engagement rings turn towards something more rare?

Default visualization settings are useful when it’s just you analyzing data. However, visual focus is better when it’s time to present.

You know the classic game Pong with the paddles and ball that moves across the screen? Imagine the ball and paddles synchronized to music. Victor Tao approached the challenge as an optimization problem to figure out where the paddle and balls should go, based on the beats of a song:

Fortunately there is a mature field dedicated to optimizing an objective (screen utilization) with respect to variables (the locations of bounces) in the presence of constraints on those variables (physics and the beats of the song). If we write our requirements as a constrained optimization problem, we can use an off-the-shelf solver to compute optimal paddle positions instead of designing an algorithm ourselves.

The result is Song Pong, and the Python code is on GitHub. [via Waxy]

The Economist examines car weight and fatalities in car crashes. In two-vehicle collisions, while heavier cars tend to mean fewer deaths for those driving them, the opposite is true for the other car involved.

The heaviest 1% of vehicles in our dataset—those weighing around 6,800lb—suffer 4.1 “own-car deaths” per 10,000 crashes, on average, compared with around 6.6 for cars in the middle of our sample weighing 3,500lb, and 15.8 for the lightest 1% of vehicles weighing just 2,300lb. But heavy cars are also far more dangerous to other drivers. The heaviest vehicles in our data were responsible for 37 “partner-car deaths” per 10,000 crashes, on average, compared with 5.7 for median-weight cars and 2.6 for the lightest cars.

There is no shortage of color-picking tools, but it seems there can never be enough. Enter Huetone by Alexey Ardov. The tool is based on a Stripe article on accessibility and color design. Mainly, it shows different shades of selected hues.

It’s not the most usable of tools but maybe you’ll find it helpful.

K-pop grew mainstream-popular in countries outside South Korea over the past few years. This growth partially comes from efforts to internationalize the music genre with changes in language and group members from around the world. For Bloomberg, Sohee Kim, Jin Wu, and Jeremy C.F. Lin have the data rundown.

Plus points for the variable width bar chart and a star microphone for a cursor.

Kirk Goldsberry, with help from Andy Woodruff, combined two joys — basketball and maps. The result is ATLAS, which is a basketball that is also a globe. Genius.

The limited first run already sold out, but they’re taking pre-orders for a larger run now.

To show the counties with more or fewer jobs when comparing 2023 to 2019, Ben Casselman and Ella Koeze for the New York Times use a county map with up and down arrows. Green and up means a gain, whereas orange and down means a loss.

We’ve seen similar maps with arrows, but they’re usually angled or swooped. I guess I always assumed arrows going straight up and down would jumble together, but this seems to work.

Every month I collect tools and resources to help you make better charts. This is the good stuff for August 2024.

To show poverty in the Pacific Region between age groups, gender, and location, Kristin Baumann used a plant metaphor. Each element on the interactive represents an attribute. As a whole, the better the plant grows for a country, the lower the poverty.

Pretty.

The project was centered around a Pacific data challenge, but it’d interesting to see a whole bunch of plants for a lot more countries.

In a riff on Climate Stripes, which shows global temperature change as a color-coded barcode chart, Air Quality Stripes uses a similar encoding to show pollution concentration from 1850 through 2021.

In the latest of the genre I-followed-AI-advice-blindly-here-is-what-I-got, Ceylan Yeğinsu for the New York Times, tried AI trip planners for a Norway itinerary:

Overall, Mindtrip — with its polished, dynamic interface that allowed me to cross-check details with maps, links and reviews — was my favorite. While it gave some good recommendations, Mindtrip needed more prompting than Vacay, which offered a wider variety of suggestions in more detail. Unfortunately, Vacay doesn’t save chat history, which I discovered halfway into my planning after closing the website’s tab on my browser.

The biggest drawback was the absence of phone apps for Mindtrip and Vacay, which led me to rely on ChatGPT’s basic A.I. assistant when I needed on-the-spot guidance. Mindtrip, I’ve since learned, is planning to debut an app in September.

This seems like the natural next step from searching Tripadvisor for top reviewed things to do in a given place. So the experience seems as expected.

But I can’t wait until AI just takes the entire trip for you and then we won’t ever have to go anywhere. When is that going to happen? Amazing times ahead.

Annette Manes, a retired widow and single mother who saved by spending little, was scammed out of $1.4 million of her life savings. Bloomberg shows the large deposits and withdrawals through Manes’ JPMorgan checking account with a step chart. There’s also a Sankey diagram to show the splits and bank-specific timelines that make you wonder why the banks’ systems didn’t start alerts sooner.

The style of annotation and scrolling through time reminds me of the 2015 chart showing one person’s weight loss diary.

The Data GIF Maker is a fun tool from Google that lets you add movement to a handful of straightforward charts. Enter up to five values and select from four chart types. Get a downloadable GIF that you can stick in a presentation.

Here’s a small example for a stacked bar with two values:

The charts are simple and the animations are just bringing the data into view. I prefer to show charts outright, but if this is your thing, the tool makes it really easy.

Ben Grosser made a personal mini-site that indicates whether or not he’s on TikTok:

Stuck in the Scroll is a compulsive scrolling monitor that reveals, in real-time, whether I’m currently scrolling TikTok. Think of this work as a last-ditch effort, a sort of public confessional as therapeutic tool aimed at defusing the intense compulsion I feel every day to endlessly scroll the world’s most popular video app.

Combine this with counting the days you have left in life and you’ve got yourself a proper YOLO app.

You spent time and effort on a data project and it ends up in unexpected places. Do you just shrug and let it pass or do you try to get a handle on the work so that it’s not misconstrued out of context?

The climate is changing, and researchers believe that after some point, there will be no going back. The balance will be too out of order to fix. For the New York Times, Raymond Zhong and Mira Rojanasakul map the possible places affected, along with timelines for where we are and when the change might happen.

I appreciate the illustrative aesthetics. A sort of fragile uncertainty in our future.