An LLM is not a human and in particular doesn't perform at a consistent level of ability like a human would. So past performance is only a weak indicator of future performance, even without changes to the underlying model. It was always a slot machine.
An LLM is not a human and in particular doesn't perform at a consistent level of ability like a human would. So past performance is only a weak indicator of future performance, even without changes to the underlying model. It was always a slot machine.