Researchers have only trained these LLMs on more data and have even less understanding of what these LLMs do internally since their architectures are with in a massive black-box with unexplainable numbers and operations going on.
That isn't helpful to researchers or even serious professionals in high risk industries. It makes LLMs less trustworthy for them are is incredibly unsuitable for their use-case in general.
