That's a good overview of the energy demand, it doesn't account for all the resources needed of course. Now of course like most articles and studies on the topic, it's very inaccurate because of the opacity from the major providers in that space. The only thing we know is that the numbers here are likely conservative and the real impact higher. Mass use of those models inferences is already becoming a problem, and it's bound to get worse.

Even with conservative estimates some uses are very much energy hungry... Especially when they support a surveillance apparatus. Many reasons to not get there.

Are they really believing their own lies now? More likely they're trying to manipulate clueless lawmakers at this point. They can't afford to let the circus end.

I mostly agree with this piece. There's lots of room for optimization still so we might see a temporary drop in the energy consumption of those systems. That said, longer term energy consumption is indeed the main leverage to improve performance of those systems. It can only get us so far, so new techniques will be needed. Hence why my position is that we'll come back to symbolic approaches at some point, there's a clear challenge at interfacing both worlds.

This is not all bad news, there are a few things to rejoice about.

We always think about the energy consumption, but large data centers gobble billion liters of water too. This would need to be improved.

More signs of the current bubble being about to burst?

Nice paper which debunks the choice of the language as an important factor for energy efficiency. The previous papers had a too simple model, this one puts forth a more complete causal model. There are many factors at play regarding energy efficiency, the programming language itself is not really one of them.

If you run the number, we actually can't afford this kind of generative AI arm race. It's completely unsustainable both for training and during use...

Interesting analysis... I wonder if and how Jevons paradox will get in the way though.

This arm race should be stopped... This is becoming an ecological disaster, so much wasted energy.

More work about eco-design of software. This is definitely welcome. I found this work a bit weak on the state of the art and the interview parts (10 people in the same company). But the field is so nascent that it's to be expected I guess, PhD students have to do with what they have access to. Unsurprisingly this shows a great lack of proper tools to tackle the measurement problem. This thesis shows interesting prospects to reduce variations in measurements though, some of the proposed guidelines might help but cannot offset the hardware heterogeneity completely... The parts focusing on practical advices around Java use and deployment are interestingly easy to apply though. You need to take into account the context of your application to make the right choices of course.

Smaller models with smarter architectures and low-bit quantized models are two venues for more efficient use. I'm really curious how far they'll go. This article focuses on low-bit quantized models and the prospects are interesting.

This is one of the main problems with using those generative models as currently provided. It's time for the legislators to step up, we can't let a couple of players hoard energy and water for themselves.

Interesting paper evaluating a Life Cycle Assessment (LCA) method to estimate the power consumption and environmental impact of generative AI services. This is illustrated on a single service, hopefully we'll see more such assessments.

Very interesting paper about the energy footprint of the latest trend in generator models. The conclusion is fairly clear: we should think twice before using them.

Important and interesting study showing how the new generation of models are driving energy consumption way up. As a developer, do the responsible thing and use smaller, more specific models.

Good advices. Keep some slack, have an idea of your energy level, that's what impacts your capacity. Time is a given you can't do much with it.

Interesting position from AMD regarding the race on the next super computers. They're all being caught up by energy efficiency so it'll need to be addressed both at the processor architecture level but also at the software architecture level. How we design our computing tasks will matter more and more.

Looks like a young but interesting tool to assess the power consumption of a service. There's been quite some work in this domain on the client side, not so much on the server side. This is welcome.