Agree that better to compare a human's learning lifetime (e.g. childhood and schooling) to the training time of the model. We all know evolution is extremely slow. But humans learn (train) a tremendous amount in their own lifetime -- Mandarin, English, etc. are not encoded in our DNA but learned in two years. Our genome is fairly small relative to the numbers we are talking here, right?
I'm confused by the 3000MW figure. I go to top 500 and see ~30,000KW, i.e. 30MW???
Your comment is a reply to me, but this part is a criticism of OP. And I agree with the criticism: OP seems to be wrong about 3030 MW, unless I’m misunderstanding. (Or maybe it’s just a typo where Jesse typed “30” twice? Are the subsequent calculations consistent with 30 or 3030?)
This part seems to be implicitly an argument for some larger point, but I’m not following exactly what it is; can you say it explicitly?
Agree that better to compare a human's learning lifetime (e.g. childhood and schooling) to the training time of the model. We all know evolution is extremely slow. But humans learn (train) a tremendous amount in their own lifetime -- Mandarin, English, etc. are not encoded in our DNA but learned in two years. Our genome is fairly small relative to the numbers we are talking here, right?
I'm confused by the 3000MW figure. I go to top 500 and see ~30,000KW, i.e. 30MW???