Post

This is great - it's about time someone updated the discourse on LLM energy usage to reflect that coding agents use massively more prompts than occasional questions to ChatGPT Simon estimates that a day of coding agent usage comes out close to the energy needed to run a dishwasher

that is: - not insubstantial - really surprisingly not that much either? good update, ty for sharing!

Yeah I thought it would be way more.

Based on estimated costs per query for some of these companies to even break even, I have to admit I expected it to be several orders of magnitude larger. (I guess the costs are mostly in training and hardware?)

There is no way that any of them are going to "win" the market unless the rest of them all collapse under the weight of their VC debts, because there's several times the amount of capital investment as seems plausibly likely to be eventual market for them.

Incidentally, I do know (from talking to people who provisioned machines for them) that the large models really are several orders of magnitude more expensive to run (and slower, especially in the case of generative ones) than older classifiers. I probably had guessed wrong how expensive those were.

Oh yeah, that doesn't surprise me since they are. Uh. Larger.