You might’ve heard about large language models lately. They’re the “brains” behind recent chatbots that seem to know an awful lot. Aatish Bhatia, for NYT’s The Upshot, walks you through how such a model “learns” to write based on a relatively small body of text.

There’s a little bit of Choose Your Own Adventure mixed in, so you can select the type of text. Bhatia used nanoGPT, an open source library, to model Jane Austen, Shakespeare, and Star Trek: The Next Generation.

Split it up into small pieces and then aim for depth.

Quantum computing promises a higher level of processing power over current computers, but it comes with its own challenges with implementation. When it comes to fruition, it’ll also break current privacy measures like your banking password. Financial Times explains how this works with a fine set of animations and illustrations.

Mira Rojanasakul and Max Bearak for The New York Times highlight the rise of pumped storage around the world. Instead of a big dam to store water, two reservoirs are used. One is high. The other is low. When there is excess power available, water is pumped to the higher one, and when more is needed, water is released down to the lower one.

Stephen Curry of the Golden State Warriors scored 50 points in game 7 against the Sacramento Kings. He made 20 shots. Instead of the standard shot chart with dots on a simplified court, Todd Whitehead overlaid video frames of every made shot for a single image.

I assume all shot charts will take this format from now on. Thanks, ESPN.

Bed Bath & Beyond filed for bankruptcy, so Alex Leeds Matthews and Amy O’Kruk for CNN took the chance to look at the product offerings for the store. It is mostly beyond at this point.

I enjoyed the goofy premise.

Ice is melting at the poles, which makes it seem like a localized problem. However, as NPR shows and describes in a visual series, the large amounts of fresh water melting into the ocean mixes in with currents and changes temperatures, chemistry, and water levels globally.

Here’s the good stuff for April.

For Bloomberg, Zahra Hirji and Denise Lu on the electrification of the national school bus fleet:

Most school buses today run on diesel. The climate footprint of a diesel school bus is about 3.3 pounds of carbon dioxide equivalent (CO2e) per mile, more than double the per-mile footprint (roughly 1.5 pounds of CO2e) for a bus powered on the average US electric grid, according to Argonne National Laboratory. If a large share of the American school bus fleet — the largest mass transportation system in the country — electrifies, that would translate to a significant emissions cut.

They used a LEGO school bus to show how a diesel school bus is retrofitted as an electric one, which is a plus in my book.

Saving for retirement is a slow process with an end goal that can seem far away, especially if you’re young. So we put it off until later. Then time sneaks up on us, for many it seems, and we’re stuck having to work longer and rely heavily on Social Security. Over at USAFacts, we looked at how much Americans save and show how your savings compare.

Disney began more layoffs, and data-centric FiveThirtyEight, which is owned by Disney, was part of the round. Nate Silver, founder and editor-in-chief of the site, also announced he is likely to be leaving soon:

Disney layoffs have substantially impacted FiveThirtyEight. I am sad and disappointed to a degree that’s kind of hard to express right now. We’ve been at Disney almost 10 years. My contract is up soon and I expect that I’ll be leaving at the end of it.

I had been worried about an outcome like this and so have had some great initial conversations about opportunities elsewhere. Don’t hesitate to get in touch. I am so proud of the work of FiveThirtyEight staff. It has never been easy. I’m so sorry to the people impacted by this.

Feels like a shift in data and stories.

Daniel Wolfe for The Washington Post looked at the similar word choices across cannabis business names:

To check if companies are distinguishing themselves, we analyzed every dispensary listing from WeedMaps, a map directory for local cannabis distributors. Here’s what patterns emerged when we examine the company’s name through a language model.

The premise is that businesses should aim for brand differentiation, and if all the dispensaries have similar names, it’s tough for any one to stand out.

I guess that’s true, but all I could think about was all the Chinese restaurants that I’ve been to, which also have similar names, even in the same city. People definitely are still able to pick out the good places.

Based on the United Nations’ world population report, it is estimated that India’s population will increase past China’s some time this year. For The New York Times, Alex Travelli and Weiyi Cai have charts to show how and why.

You know what they say about assumptions.

Neal Agarwal is up to his wonderful ridiculousness again. Imagining an elevator that goes up to space, a long scroll through the skies gives you a sense of elevation up until you leave Earth. See how high birds fly, where the wild yak resides, and who was the first person to break the sound barrier in a free fall.

Maybe you’ve wished you could quickly grab the data on a webpage and instantly have it in a structured format. But you don’t know how to program or you do, but don’t want to go through the trouble of writing another one-off script. Samantha Sunne provides a short guide for scraping without programming, mainly with Google Sheets.

When you have a lot of noisy data, it can be helpful to detect signals in some sort of quantitative and automated way rather than just eyeballing it. Signal detection theory can provide some of the pieces to this puzzle. Adam Krawitz has a detailed interactive guide:

This site approaches SDT from multiple complementary points of view. First, we use SDT to fit your empirical data from an example task, and consider how well the theory accounts for the data. Second, we use SDT to make predictions by running simulations with an SDT-based model performing the task and generating synthetic data. And third, we explore the space of possible results that can be generated by SDT, providing an existence proof for its capabilities.

A straightforward lineup of animals that fly provide a sense of scale, from tiny to very big. I feel like some everyday objects like a car or a helicopter would’ve really driven the point, but it’s neat to see.