>The scientific basis for this statement is unclear.

Is it? Actually this seems like one of the MOST clear aspects of LLM performance we can measure.