How good is a 70B parameter open-source AI model? Testing WizardLM ...