UKH

John Naughton on great form on a very important topic.

https://www.theguardian.com/commentisfree/2023/dec/23/ai-chat-gpt-environme...

Why don't these journalists go to a bit more effort and discuss some of the solutions.

Analogue computers that specialise in a specific AI model for example. These can reduce energy requirements by 97% if brought to fruition.

And then there's our rapidly cleaner energy grid that's making great progress.

There are also data centers coupled to wind farms that spring into life when there's excess energy.

I was going to comment but the author last all credibility with their use of “exponentially”…

The article doesn’t mention heat recovery to provide district heating as is being done in Dublin I think, which looks a useful mitigation in colder countries.
I found the “125 round trips to New York” a bit misleading, as is discussed in the comments.

Not just district heating - https://www.bbc.co.uk/news/technology-64939558.amp

I found it quite shameful to work in a department with a gas boiler and a giant simulation cluster being cooled by air conditioners.  The nation is replete with failures to usefully capture waste heat. 

You would expect university management to actually implement modern technology that their employees are knowledgeable about?

I will accept Loughborough were reasonably forward looking using a CHP system to power the library and several halls while also using it for a coursework exercise, no idea if they have kept with the times as that was 9 years ago.

memristors (see https://www.frontiersin.org/articles/10.3389/fnano.2021.645995/full )

will give us logic where the junctions between elements will be analogue and varying as the values stored are increased or reduced as learning strengthens or weakens associations.

These could be closer to actual neural networks than the behemoths we need to model the same responses in our current steam generators.