The other way is not to spam someone with LLM-generated outputs, and actually review the package