No.

Modern LLMs are trained on a large percentage of synthetic data.

This sentiment is largely legacy (even though just a couple of years old).