Ah yes, another day, another paper on optimizing a fundamental AI capability. When will we stop reinventing the wheel and start solving actual problems?