> These usages of LLMs require us to trust the output of the LLM without having a human read and understand it. Would be nice if it was true. Whereas in fact a

> These usages of LLMs require us to trust the output of the LLM without having a human read and understand it. Would be nice if it was true. Whereas in fact anyone can put out parrot-generated code without trusting anything except that it saved time and effort.