Interesting example is the "black sheep problem" - language models don't "understand" that sheep are overwhelmingly white. Why? Because we tend not to specify t

Interesting example is the "black sheep problem" - language models don't "understand" that sheep are overwhelmingly white. Why? Because we tend not to specify the colour of a sheep in writing if it's white, and because of the associated idiom about misfits ("black sheep" outnumbers "white sheep" 25:1 in written English).