AI Datacentres use about as much electricity as the NHS - go figure.

They should be built in remote places and made to generate their own electricity by renewables they build for their own use. They shouldn't be taking up a nation's limited supply of energy.

Better still they shouldn't exist at all for general use LLM's, just for scientific, technological and medical specialisms intended to advance quality of life and care of the planet.

I guess I'm an old fuddy duddy.
 
The power density of these AI machine has really shot up compared to data centres of old. It used to be we talked about maybe 240Vx 32Amps for a rack, so about 7.5Kw but now they are many GPUs stacked together they are now talking about 1000KW per rack. This is why they have moved to exotic cooling with water as the transfer and all the extra noise. These GPU based data centres are nothing like the data centres that are all over the country already they are completely different level of consumption. I wish it was for something that had more value.
 
Last edited:
They should be built in remote places and made to generate their own electricity by renewables they build for their own use. They shouldn't be taking up a nation's limited supply of energy.
exactly.

It’s the same thing for water use. Like they don’t need clean water. They could be integrated into greywater/industrial water systems and not “waste” any water. But no apparently profit and building them as fast as possible is all that matters so they’re plugging into the clean water grid and taking it away from communities who need it. The fact this is happening in the desert in the US is so ridiculous.
 
The strategy in Norway has always been to build any power hungry facility in small villages where the tens or hundreds of jobs would make up a large enough proportion of the job market to get the politicians to agree to basically anything.
 
I just don't see all the data centers being built. Selling matrix multiplication as something that will take everyone's job was a massive mistake by the tech oligarchs. People will be protesting these centers everywhere within a year, in fact it is already happening and this was before increasing energy prices.
 
The strategy in Norway has always been to build any power hungry facility in small villages where the tens or hundreds of jobs would make up a large enough proportion of the job market to get the politicians to agree to basically anything.
I wonder if a strategy in the future will be not to build anything in such countries and out-source it to countries where electricity is virtually free (or at least is pretended to be so). Maybe they are waiting for more stable governments, but I'm sure they can arrange that. Doesn't Palantir have a contract with the DOD?
 
Beyond the noise and environmental impact one of the big problems with such an enormous energy consuming centre is that it will result in yet more gas usage because we don't have enough wind/solar yet. This will push energy prices higher for everyone just based on how our energy market works. The same thing has been happening in the USA where more expensive and easy to access fossil fuels have been used to power them. The UK is already struggling with some of the highest electricity energy prices in the world and they have only recently started to come down due to a higher percentage of solar and wind, this would set that back quite a bit.

I don't want to be paying higher electricity costs just because a new data centre has been built, its not going to bring much in the way of jobs since its mostly just computers after its built.
 
It just speaks to the total insanity of Silicon Valley that they started pushing this as the climate crisis became an imminent threat.

LLMs are not worth any of the impact these data centres will have. Before they enshittified it Google search worked just fine, and didn't require mass plagerism...
 
I think the impact on peoples lifes and their jobs is being underappreciated, I think the livelihood of tremendously many people will vanish in a very short timeframe.
We can barely build things in the west.

I don't see how we are supposed to build data centers for trillions in an inflationary environment with a lack of resources due to closure of the strait within a very short timeframe.

Even if you could build these giga projects within years, you'd still have to actually solve hallucinations in LLM, which there is no evidence of yet.

Looking at the stock market it's hard to argue that AI impact is underappreciated. People seem to think we will have magical AI within years, given that semiconductor stocks now make up a ridiculous amount of the SP500. Maybe these speculators are right!! However I just don't see it.
 

Only scanned this just now but it seems relevant
 
How about technological advancements which would make it cheaper and less resource-hungry?

People have got a glimpse into the possibilities of AI and I think there's no turning back. Even if the companies working on AI chips don't deliver soon, I think those won't be the last attempts to make AI more viable.
 
I don't see how we are supposed to build data centers for trillions in an inflationary environment with a lack of resources due to closure of the strait within a very short timeframe.
I think they are already being built by private companies hoping for fortune.

Even if you could build these giga projects within years, you'd still have to actually solve hallucinations in LLM, which there is no evidence of yet.
That depends on the discipline. Hallucinations have reduced substantially for LLMs and for many jobs they will do are rather irrelevant as you can just have infinite agents with some form of automated proof verification. For example in mathematics or programming that proof verification can even be formal, for other hallucinations you can already get quite far via informal verifications. If the data is good, the problem decreases. The issues we usually discuss on this forum tend to be of a different order and I currently don't see it being of much use in that scenario, but that doesn't mean GP's and a whole lot of other people in the medical industry aren't at risk for loosing their jobs. It's a monkey with a type writer situation and as long as you have a million different monkeys checking that the end result is Shakespeare you're quite good, especially since we've always been pretending energy is essentially for free.

I used to be very sceptical until recently, especially of LLMs, and am aware of certain limitations, but the newest (non-free) versions of the leading LLMs are already surprisingly sophisticated, certainly sufficiently sophicasted for millions of people to loose their jobs, whether we like that or not. I've been suprised by how far you can already get with "matrix multiplication", when you have good data, despite it not being very sophisticated.

My partners company just made a new marketing video what would usually take months and tens of thousands to produce, within 3 minutes via agents. Maybe some of the images were slightly off and hallucinated or not very original but you can suspect that people in that industry will struggle as well.

@V.R.T. mentioned the problems about financial sustainability and I hope it turns out to be highly unsustainable especially when the margin of improvement decreases, but the problem is that even now there's already a massive dependency in certain sectors both in businesses and in other sectors such as higher education with seemingly no one guiding the ship to a worthwhile harbour.
 
The noise pollution by data centers is horrendous, too:

sidenote whilst I watch this before I forget. At 24min there are the results from the experiment and he rules out the lethargy because he couldn't be sure what it measured when it had been so different a number to tiredness. Lethargy gave 80% where tiredness 20% or something

I of course find this fascinating to see a group of not me/cfs (or other type of patient) being asked scores for various 'symptoms' and giving such different scores for tiredness and then lethargy given the term 'fatigue' that gets used ambiguously for us when we try and describe more specific things
 
They should be built in remote places and made to generate their own electricity by renewables they build for their own use. They shouldn't be taking up a nation's limited supply of energy.

Better still they shouldn't exist at all for general use LLM's, just for scientific, technological and medical specialisms intended to advance quality of life and care of the planet.

I guess I'm an old fuddy duddy.
Indeed. It puts a new spin on the stuff out there at the moment talking about AI having the potential to destroy humanity. through other means/issues.

sucking up resources needed for other things and pollution of different sorts (noise, air)

which the bitcoin things on the video also seem to do, which is another thing I really struggle to get the justification of happenning to allow planning of such big things to have happened when apparently it isn't an accepted currency so what is an application for planning based on? or do the electricity production related bits not go thru some sort of planning type license/sign off?
 
The strategy in Norway has always been to build any power hungry facility in small villages where the tens or hundreds of jobs would make up a large enough proportion of the job market to get the politicians to agree to basically anything.
That reminds me of what was said of the behavioural 'schools' in the netflix series 'The Program' - set up somewhere that needed employment, and hence any workers seeing inside have the added peer pressure that the whole town relied on it, either directly or indirectly, staying open to keep the town alive.

Which adds another issue regarding oversight and the like
 
Back
Top Bottom