The fundamental problem is that there’s money to be made by consuming more and more “sustainable” resources. The real solution is to reduce consumption on a global scale.
And how do you intend to “reduce consumption”, may I inquire?
Not OC, but some ways to “reduce consumption” are reducing our usage of inefficient technology by replacing it with more energy/resource efficient means.
Examples include replacing individual automobiles with mass transit, building more dense cities to reduce consumption of construction materials/ vehicle miles, and not training massively large language models in facilities that consume more energy than an entire small country.
In real world application, increased efficiency doesn’t decrease energy usage nor decrease labor required to live. Tech has gotten more efficient since the industrial revolution, but demand for technology has increased exponentially, energy use is astronomical, and workers still work more hours.
… gotta admit this is quite a bit more sound than I anticipated
As for LLMs, people don’t really like when others say they can’t explore the applications of tech, even if it’s unsustainable, so there’f bacaklash ofc
This may be true of chopping down forests or mining coal. But we can use nuclear power. And the earth has plenty of water – does chatgpt need clean drinking water specifically?
Datacenters moved to using evaporative cooling to save power. Which it does, but at the cost of water usage.
Using salt water, or anything significantly contaminated like grey water, would mean sediment gets left behind that has to be cleaned up at greater cost. So yes, they generally do compete with drinking water sources.
There’s no way nuclear gets built out in less than 10 years.
A simple trick would be to stop the wars
I wonder how much energy google wastes on its AI service in the regular search just to give me a worse answer than the top results I was actually looking for.
Purchase more carbon credits
Indulgences never really went away did they
Unlike purchasing things for imaginary gods, carbon credits could work in theory. At least well enough to be part of the solution. That is, if they were properly regulated around strategies that actually absorb carbon and everyone is forced to be honest and transparent.
Which none of them do, of course.
:o
I mean the ibm 1401 used ~13,000 Watts so power hungry is just the way it starts.
AI uses water?
Cooling uses freshwater and often drinking water
That’s crazy, they should use heat pumps and maybe underground refrigerant loops instead.
But that costs money
Also the internet uses massive amounts of water and energy
Well obviously the little people who live in the computer get thirsty because it’s so hot in there
Oompa-Loompas need their pools
All data centers use lots of water for cooling.
The water is mostly returned to the watershed, from what I understand. Except the bits that evaporate
They primarily use evaporative cooling. Way less energy use, but no, it doesn’t get returned.
Transition to paperless office
The problem is there is people who say bullshit like this unironically.
Seriously. It isn’t helpful towards the environment if we are using so many resources to mine chips and metals and then push it along the internet to then be trained on said AI bots. Would be more sustainable using paper and planting trees smh
And it doesn’t even make sense
99% of a modern office’s correspondence already goes on online, and only the most important stuff gets backed up on paper copies, often because of regulations that are there for a reason
Ehhh. I get that exploitative techbros and cryptobros have confused the issue by latching onto the AI bubble.
But at the same time generalized artificial intelligence is very likely possible and will be an absolute game-changer if and when it happens. It’s easily of similar value to fusion technology.
And it is already bringing truly impressive results into reality - protein folding and diagnostic medicine come to mind.
But at the same time generalized artificial intelligence is very likely possible and will be an absolute game-changer if and when it happens. It’s easily of similar value to fusion technology.
The “AI” we have now is basically advanced Autocomplete.
In the same way that computers are basically advanced abaci.
Don’t confuse a simplification made to demonstrate the basic functioning to a layman with how things actually work.
LLM’s are neural networks, which are based on a model of brain function. There’s little reason to believe that we cannot eventually reach similar levels of effectiveness as human brains.
Hell - reaching the levels of pigeon brains would already be absurdly useful.
While I agree that LLMs can achieve human-tier efficiency at most tasks eventually (some architectural changes will be necessary, but the core approach seems sound), it’s wrong to say it’s modeled after the human brain. We have no idea how brains work as they’re super complex, we’re building artificial neural networks from the ground up. AI uses centuries’ worth of math, but with our current maths knowledge the code isn’t too complicated. Human brains aren’t like that, they can’t be summed up in a few lines of code because DNA is a huge mess that contains so much more than just “learning”, so many inactive or redundant bits and pieces. We’re building LLMs with knowledge of how languages work, not how brains work.
Transformers are not built with our knowledge of language. That’s a gross approximation – it would honestly be more accurate to say they’re modelled after the human brain than that they’re built with our understanding of language. A big problem is that the connection between AI and language is poorly understood – we can’t even understand what the word2vec axes are.
i’m not talking about knowing about how humans perceive/learn languages, i’m talking about language structure. Perhaps it’s wrong to call it “how languages work”
That’s what I meant, yes. They’re not built based on any linguistic field
different neural network types excel at different tasks - image recognition was invented way before LLMs, not only for lack of processing power, but also because the previous architectures didn’t work with languages. New architectures don’t appear out of thin air, they are created with a rough idea of what we could need to make the network do a certain task (e.g. NLP) better. Even tokenization isn’t blind codepoint separation but is based on an analysis of languages. But yes, natural languages aren’t “parsed” for neural networks, they don’t even have a formal grammar.
The problem is they’re already talking about needing trillions of dollars worth of hardware to make it happen. It’s absurd.
