But the signals used to filter out human garbage are not the same the signals that would be needed to filter LLM garbage. LLMs generate texts that look high-qua

But the signals used to filter out human garbage are not the same the signals that would be needed to filter LLM garbage. LLMs generate texts that look high-quality at a glance, but might be factually inaccurate. For example, an LLM can generate a codebase that is well-formatted, contains docstrings, comments, maybe even tests; but it will use a non-existent library or be logically incorrect.