Comment by nzeid

I see comments that this overlap between Qwen and GPT is due to rogue training or post hoc training. Did it occur to anyone that maybe the two sets of models were trained directly on the same solutions to the researchers' benchmark?