Slightly related, on the cooperation between large models and small models (traditional ML) : https://arxiv.org/abs/2409.06857