And that there's this slow realization that "we really do not understand what it really means to understand" is at the heart of current limitations of LLMs (or self driving cars) therefore a higher several trillion parameter count might not necessarily translate into more intelligent AI with a deeper understanding of things.
EDIT: Typos