Re:The Collapse of GPT
by lelanthran on Saturday May 17, 2025 @12:48PM
(#44015513)
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.
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.