Lucas Sato comments on METR’s Evaluation of GPT-5