I open PoweReader in the morning and there are over a hundred unread articles again. One of them is about a telescope in Chile and its sky survey data. The headline sounds interesting, but I'm not sure it's worth five minutes.

I tap Summarize, and a few seconds later a paragraph appears: the survey's first data release added about a billion objects to the catalogue, doubling it overnight. Below the summary there's a diagram, and one glance makes it clear: the hardest part was storing and calibrating the data, and the hardware that processed it fits in a garage. In under ten seconds I know what the article is about, and I mark it as read for now.

The diagram shows what the summary can't

The summary already gives you the key points in words. If the image just laid out the summary again, you'd be reading the same thing twice and getting nothing new.

So before drawing anything, AI works out what in the article is faster to see than to read, and picks one way to draw it:

Steps and their causes, drawn as a chain
Before and after, old and new, or two sides, placed side by side
The article's numbers, drawn to their real proportions
An object, drawn with the parts the article talks about labeled
Dated events, laid out on a timeline
Every item in a roundup, grouped into a list

One look tells you what happened

The headline on the image tells you the article's news, like "A billion new objects, and the hard part was the data". Every label is a specific number, name or date from the article, so you can tell what happened from the image alone, without going back to the summary.

Visual summary of the Chile telescope survey article

Decide faster what to read

When you're going through a lot of feeds, reading the summary and glancing at the diagram is usually enough to tell whether an article deserves a full read. Keep the ones that do for later, mark the rest as read, and your unread count stops piling up. The articles that really matter are still worth reading from start to finish.