A machine learning engineer proposes a new pre-training strategy for a large language model. They argue, 'Since the ability to learn from examples provided in a prompt is so powerful, we should explicitly include thousands of task-formatted examples (e.g., 'Translate this sentence: [sentence] -> [translation]') directly into our massive pre-training dataset. This will directly train the model to perform this skill.' Evaluate the engineer's proposal. In your evaluation, explain why this approach might not be the most effective way to develop this capability in a large model.