Hacker News Reader

Yeah, I just tried it with the inverted question mark (¿) and interrobang (‽) -- both attempts had the correct glyph for the first result.

Maybe we're training it?

Probably not training "matching", given that there is no place where you indicate what the correct thing was.

It could be gathering data about how people draw shapes in general, but it's not immediately obvious to me how much that can help.

> given that there is no place where you indicate what the correct thing was.

Actually there is a "Good" and "Bad" voting feature for each result.

Retracted then. I hadn't noticed that.

I've been completely unable to get it to recognize an inverted interrobang (⸘) with no luck, though. Still, very very cool.

EDIT: Tried one more time and got it as the 9th suggestion.