I would love to see real LLM performance benchmarks for these machines. Apple statement regarding LLM performance seem little vague.