If a human makes errors and discovers it, the error gets fixed. With AI, it usually fails to fix the error totally on first try, even on nth tries. If on the off chance it does fix it, it generates other problems.

Humans do produce errors. But they produce far fewer than LLMs, and they can also learn. Once a person makes a given mistake and realizes what he did wrong, he will take steps to prevent that mistake in the future. LLMs do not do this and therefore are less trustworthy than humans, even if both are prone to error.