Where We Have Arrived in Proving the Emergence of Sparse Symbolic Concepts in AI Models