[zamadatix]

Sprinkling in benchmark training isn't a loop, it's just plain cheating. Regardless, not all of these benchmarks are public and, even with mass collusion across the board, it wouldn't make sense only open weight LLMS have been improving.