Generalist embedding models are better at short-context clinical semantic search than specialized embedding models.