An important thread about misrepresentation of CO2e numbers for different ML models. I believe it's good to use actual measurements of this data, rather than estimates that are flawed in a variety of ways.

→ View original post on X — @jeffdean