That's still very different to "drift error compounding". If you output a mere 5% drift error and then use that as input you only need a few cycles (single di

That's still very different to "drift error compounding". If you output a mere 5% drift error and then use that as input you only need a few cycles (single digits) before your output is more erroneous than correct. We are already partly into the second cycle. By the fifth the LLM would be mostly useless.