How glassy are neural networks?