we just got a CAPTCHA which was quite worrying in that it presented photorealistic images of bridges, and also photorealistic images which had all the elements of a bridge but lacked spatial coherence

we marked the incoherent ones as not-bridges and it let us in. was that the intent, are we training it on spatial coherence now? or does it just accept that silly humans are wrong sometimes?

it's not going to be much longer before we are personally unable to pass these. it's really hard to visually parse the fucking things, and the fact some of them are intentionally nonsensical while others are only accidentally so is not helping.