Progress compared to SLMs and the early days of deep learning is real.

However, I know of no theoretical limits on scaling laws other than compute and data.