Because this is supposed to be a scientific paper, it's supposed to be talking about quantifiable effects, not of some speculations. Especially in this case whe

Because this is supposed to be a scientific paper, it's supposed to be talking about quantifiable effects, not of some speculations. Especially in this case where the high energy and monetary cost are easy to estimate. OpenAI spent $3 million just on the compute for training. Even inference requires multiple TPUs each using hundreds of Watts.