i think it's auto-generating a problem based on geotagged pictures. the indoor one isn't supposed to be a good clue, it's just a photo that happened to be in the dataset.
Could maybe pass the pictures through an object detector. Too many faces and it skips the picture. It may already be doing it to remove pictures with words as one "clue" containing a sign would spoil it. Though I'd rather it keep the picture but blur the sign in that case.
they could, but i kinda like the wildness of it. not knowing if you're going to get a completely useless clue (or one that totally gives away the answer) is part of the charm,.
