Would a LLM under human guidance turn out to be a good verifier ? i.e. if LLM knows the rules to verify or has enough data points (internet access, actual responses)