AI already uses as much energy as a small country. It's only the beginning | Hacker News Reader

AI already uses as much energy as a small country. It's only the beginning

vox.com

10 pointscdme10 comments

I'm always suspicious about these pieces talking about how much energy crypto, or AI, or anything else use. How would you ever go about finding an accurate number for how much power is consumed for some specific use case of all computers in the world? It's mostly just guessing, and you can pump the numbers up however you want - include the energy used in the production of the servers, moving them around, recycling them after, the travel of the employees who work on it.

And then that's used to paint those technologies as morally wrong. You can do a similar article about anything. Why not video games? They use tons of power, and nobody actually needs them. Or accounting software, or art.

The fact is that energy expenditure and technological adoption will keep increasing. We should be focusing on generating the energy we use more sustainably, rather than trying to make people spend all day deciding which personal choice to make to minimise the amount of energy they use.

Perhaps this planet will eventually be just GPUs answering to each others' stupid questions and ASICs working at bitcoins.

A paper clip maximizer.

ChatGPT only required the electricity of 130 homes to train? This seems quite low given the benefits to society.

Those benefits have yet to materialize.

Not true. It’s one of the most groundbreaking changes to self-education on just about any well-known topic. It’s like having a tutor who is available to ask endless questions and provide a variety examples. When I’m learning something new to me, it’s the first place I turn. If I’m learning something critical, I do fact check ChatGPT, but it points me in the right direction.

I agree it’s often not useful but I’ve seen some gains from using it. The Model T also was not exceptional.

Reminds me of cryptocurrency "mining".

Energy crisis? What energy crisis?

That stuff people were warning about in the 1970s, it's all been fixed.

Environment is as clean as a whistle!

Maybe people should ask AI how to produce enough clean energy to keep it running?

Bad title, and incorrect characterization.

> electricity consumption associated with data centers, cryptocurrency, and artificial intelligence.

> The IEA estimates that, added together, this usage represented almost 2 percent of global energy demand in 2022

First, Bitcoin by itself uses about 0.5% of work electricity. This is purely wasted on proof-of-work, which could be proof-of-stake instead.

Second, grouping all datacenters with AI models is disingenuous. I don't think anyone argues training LLMs is very costly in energy, but it's also a fraction of the energy costs of "literally all compute". It's just the hot thing to write clickbait on.

that and there are a subset of mining sites that reduce emissions and sell energy to the grid

and another subset that only reduce emissions and are using energy that is still isolated from the grid, has always been there otherwise unused, and is still too expensive to transport to other use cases - except maybe AI training now

data centers never showed up because they needed fast internet. bitcoin mining needs minimal over the last decade and a half, and now training a model only needs minimal too.

Reply on news.ycombinator.com