[weinzierl]

> You're forgetting the main benefit for google, which is getting humans to train all their vision models for free.

Is this still true? I keep seeing the same type of images for years and there might be 7 or 8 different categories but that's it. To me reCaptcha looks like a service well in its maintenance phase. If it was actually in use for training purposes you might expect images to match a wider range of tasks.

[krut-patel]

I could swear I've seen challenges with night scenes (low light conditions) in the wild. Those were definitely not present earlier.

[airstrike]

I've lost track of how many times I've had to read house numbers from Google Street View...

[sli]

I haven't gotten one of those in years. These days it's just picking out buses, cars, traffic signals, and sometimes motorcycles. Maybe once in a while it'll ask for storefronts.

[eythian]

Most of mine lately have been traffic features also. This is a little tricky in some cases, e.g. with crossings, as it sometimes gives me things that I don't think are crossings but it insists I select, perhaps they are in the US, or the perspective is weird, or someone else has told it that a series of white squares is a crossing and it requires me to agree.

[sixothree]

Like trees, bridges, fire hydrants, cars, buses, house numbers, etc?