cianc.bsky.social
@cianc.bsky.social
A critique of the recent Google paper on the environmental impacts of AI inference. There has been a lot of fair criticism, but some is about Google (and other AI companies) rather than this paper, and there is genuinely good stuff here too.
notes.engineeringforclimate.com/posts/measur...
notes.engineeringforclimate.com/posts/measur...
Paper: Measuring the Environmental Impact of Delivering AI at Google Scale
arxiv.org/abs/2508.15734 Abstract: The transformative power of AI is undeniable - but as user adoption accelerates, so does the need to understand and mitigate the environmental impact of AI serving.
notes.engineeringforclimate.com
August 25, 2025 at 7:18 AM
A critique of the recent Google paper on the environmental impacts of AI inference. There has been a lot of fair criticism, but some is about Google (and other AI companies) rather than this paper, and there is genuinely good stuff here too.
notes.engineeringforclimate.com/posts/measur...
notes.engineeringforclimate.com/posts/measur...
I wrote about a new paper that claims to be the first full lifecycle assessment of AI hardware notes.engineeringforclimate.com/An_Introduct...
tl;dr: 3x better efficiency over 4 years with operational emissions much larger than embodied. I have some small quibbles with the amortisation assumptions.
tl;dr: 3x better efficiency over 4 years with operational emissions much larger than embodied. I have some small quibbles with the amortisation assumptions.
Paper: An Introduction to Life-Cycle Emissions of AIHardware
www.computer.org/csdl/magazine/mi/5555/01/11097303/28IR6YdAq9W (with a somewhat extended version here that goes deeper into the methodology) Abstract: Specialized hardware accelerators aid the rapid a...
notes.engineeringforclimate.com
August 18, 2025 at 3:35 PM
I wrote about a new paper that claims to be the first full lifecycle assessment of AI hardware notes.engineeringforclimate.com/An_Introduct...
tl;dr: 3x better efficiency over 4 years with operational emissions much larger than embodied. I have some small quibbles with the amortisation assumptions.
tl;dr: 3x better efficiency over 4 years with operational emissions much larger than embodied. I have some small quibbles with the amortisation assumptions.