How can a language both be human-unfriendly and also sanely auditable? The types of issues that require human intervention in LLM output are overwhelmingly things where the LLM depends on the human to detect things it cannot. Seems to break the loop if the human can’t understand well
No comments yet.