Wim🧮Apr 23

Altman says "The fair comparison is if you ask ChatGPT a question, how much energy does it take once a model is trained to answer that question, versus a human, and probably AI has already caught up on an energy efficiency basis, measured that way".

This is really devious: (1/3)
https://www.cnbc.com/2026/02/23/openai-altman-defends-ai-resource-usage-water-concerns-fake-humans-use-energy-summit.html

Show threadWim🧮
This is really devious: it gives the impression that what really matters is the energy efficiency per query, whereas of course energy efficiency has never led to reduced energy use. The whole history of the industrial revolution is one of improving energy efficiency. This is what caused global warming: it became cost effective to burn all that coal and oil.
(2/3)
Apr 23 at 5:56pmWeb
Show threadWim🧮Apr 23

By phrasing it as a ChatGPT-vs-human comparison he is aiming to distract us from the only real issue: the additional emissions caused by the "AI" hype he is so desperate to keep alive.

And of course it also suggests we could offset the energy use of his precious "AI" by having fewer humans. Altman is easily one of the most loathsome people on the planet.
(3/3)
#NoToAI

Show threadKarin DalzielApr 23
@wim_v12e Funny how when they want to downplay environmental impact they measure water/energy use against a query/less than a token of use, but when they are hyping the tech they talk about using up millions of tokens at once
Show threadWim🧮Apr 24
@nirak Yes, it is always enlightening to compare sustainability statements against the growth projections for investors.
Show threadRobotistryApr 23

@wim_v12e And it completely eliminates the cost of training from the discussion, even though there is no indication that they or their successors will stop *training new models* no matter how good the existing models are.

Until they provide a metric for "this is when we will stop training new models" I don't think "excluding the cost of training" is a viable basis.

Show threadWim🧮Apr 23
@Robotistry I agree, it is considerable, especially as the energy cost of training has increased a lot because of the "reasoning" models. But with agentic "AI", I think the cost of inference is much higher than the training cost. He avoided that as well by focusing on ChatGPT.