You're working on a text modeling problem and need to choose how to represent language for a downstream NLP task. Two common options are TF-IDF features and word embeddings, and the trade-offs affect both model quality and implementation.
How would you explain the difference between TF-IDF and word embeddings?