It’s can’t be that hard to generate — it’s a linear character sequence just like text — and there’s gigabytes of public domain training data literally floating around us