As always, I find it hilarious that some people believe that these companies will train their flagship model on uncurated data, and that text generated by a Markov chain will not be filtered out.