The new frontier in large language models is the ability to reason" their way through problems. New research from Apple says it's not quite what it's cracked up to be.