|
Top-level
@tjr @timnitGebru Yep. To compute the CO2 per written page, they took the CO2 a person emits per year by living and took the fraction representing the time required to write a page. As if humans would exist for less time if they did not write text. In case, anybody wants to see for themselves: https://arxiv.org/abs/2303.06219 4 comments
@uncanny_static @tjr @timnitGebru lol for the human writer they plug Mark Twain's output into the peak mean US carbon footprint where they should have plugged in a typical American writer penning SEO-filler clickbait for exposure in a country where 40% of the CO2e is emitted by the top 10%. @buermann @uncanny_static @timnitGebru "The quality might suck, but there's more of it! A win for AI!"  @uncanny_static @tjr @timnitGebru I only skimmed the “paper” but they don’t seem to be accounting for the time it took to develop the LLM software; the time to build, provision, maintain, and decommission its hardware; the time to develop the content comprising its training sets (though, fairly, one should divide this by the number of LLMs trained upon it…) All that to say, it’s a dumbass paper intended to reach a dumbass conclusion to be cited popularly by dumbasses. |
Gregory's Server
@uncanny_static @tjr @timnitGebru There is a whole field of research (life cycle analysis) dedicated to assessing environmental impacts. None of the author seems to belong to that field. From what I read, this paper does not qualify as a proper lca.