Can Language Models Teach Weaker Agents? Teacher Explanations Improve Students via Personalization