Very Large Language Models and How to Evaluate Them