Beyond Transformers for Function Learning