AI language will converge towards maximum entropy, like compression, due to cost pressure to conserve tokens. LLMs are not limited by the human 5+-2 working memory