How many TOS agreements do you suppose they violated while training their models?