The fact that big bench includes kanji-ascii art classification task is pretty funny.

But i guess if you want to have over a hundred tasks in a benchmark you run out of ideas at some point