> For the human writing process, we looked at humans’ total annual carbon footprints, and then took a subset of that annual footprint based on how much time they spent writing.
So, they aren't asking "does a human painter use more electricity to do a painting than an AI?" ... they are asking "is it more energy efficient for an AI to do a painting, or for a human to exist for several hours?".
That is, it doesn't take into account that the human would exist and have a carbon footprint anyway, even if not employed as an artist. It's devised as if employers created their employees, and seems to speculate about the potential energy saving in exterminating them.
> emissions of the computing devices used during the writing process are also considered
And their graphs show this as exceeding AI emission by a lot (due to AI being quick). This is a fair point, but they only needed this part. They shouldn't have thrown in the person's annual carbon footprint on top of this. Also, traditional art is presumably much less expensive in carbon.
And as somebody else pointed out, the use of devices will be part of the carbon footprint data in the first place, so it gets (partly) double-counted.
This is like measuring your active calories vs your total calories in a workout. Turns out that a meaningful percentage of calories are burned just by being alive, not from the work that was done.
I will only mention that I don't see LLM output as overly satisfactory.
Now. I would compare this to how humans use more carbon to move than a car for example. This could lead to a conclusion that we should stop moving and use more efficient propulsion to move. In car centered societies people move less and then need to exercise after hours more to have any sort of fitness at all.
Of course there is a grunt body (and soul) crushing physical movement as well as there is a grunt soul (and body) crushing writing work. But there is also a middle ground that would be something we aim for.
> I will only mention that I don't see LLM output as overly satisfactory.
This is changing very quickly, but even today you many not be aware of how well LLMs can create plausible drafts if you haven't seen what Claude 3 Opus can do with good prompting.
It's better than me at soft skills and communication drafting, editing and critique. It's a profound help for task planning, execution, and management.
And it's a profound help for requirements generation, architecture/design explorations, legacy code annotation, test planning, and other technical things that I struggle to do on my own
So honestly it's not just replacing writers. It's eliminating my need for executive officers, technical officers, and PMs
I'm spending 20 month on GPT4 and I'm 2-3x more productive. Good stuff!
Nevermind the total absence of an attempt to control for quality, or, y'know, that the human goes on breathing after she finishes drawing, there are some real problems here
1. the human CO2e takes national, annual averages of CO2e and attributes the fraction of a year it takes to make a drawing to that drawing. they then _add_ to that the operating cost of whatever machine the person is using, which presumably is already a component of the national individual average?
2. The 'ChatGPT' baseline is GPT-3, not -3.5 or -4.
3. Training CO2e is from a cite I can't access, but is based on GPT-3 only.
4. They cite to some guy's post on Medium to estimate ChatGPT inference CO2e!? That admits it doesn't know the number of queries per day, or the compute time used, or the energy per compute (besides the TDP of A100s), or the carbon intensity of that energy???
This is a mess, and doesn't deserve a Nature publication.
Depends on whether the generators are running, or if the power company it's hooked to is using coal, oil or gas... or when the standby generators spin up, if they're one of these.
Is the goal to eliminate most humans? Think of how much less meat and gas we’d use if we just killed everybody and replaced them with efficient computers.
It means that the current low fertility rate associated with economic progress and education might not lead to the collapse of technological civilisation even if the population drops to a more sustainable level.
Nah we'll just have to move to better communication systems - something like web of trust. We tap our phones, and from now on I know that anything that claims to be from you is from you. My friends trust that what you claim to have written came from you because they trust me. If you start spouting nonsense, I downgrade your contact information in my list and my friends get the same hint.
The open internet itself, yes, it won't survive the next few years - but then there's no stopping this. We could ban AIs in the west, which is the only place that sort of law might work, but Russia, North Korea, Iran, China - they'll still use it to mess with us.
This article suggests, in the year 2000, that replacement migration might be a way to combat low fertility rate in the west.
I don't think you can encourage low fertility rate. I don't think we did in the west.
I also know there's nowhere to get our replacement from once the whole world is educated. Technological accelerationism might save us, or it might doom us.
> I don't think you can encourage low fertility rate. I don't think we did in the west.
Pushing young women into the workplace so they can waste their lives grinding away at meaningless soul crushing fake corporate jobs instead of raising families. Degenerate Tinder/hookup culture. The "west" certainly encouraged it.
I'm not moving anywhere. I'm just an observer and that is simply what I have observed and experienced in my lifetime. Some women have told me they prefer to be a housewife, but modern society looks down on that. Nothing that I can or will do about that.
Natural selection will sort it out. No reproduction leads to extinction.
The strongest societies and cultures will survive. We'll see how it plays out over the next few generations.
I'm not moving to India, China, or Arabia, but they are certainly moving here. My city of ~2M people is now 50% foreign born and mass immigration is still at an all time high. The native populations here are reproducing at below replacement level. I have observed extreme demographic changes happening in real-time within a generation.
It is incredible to experience and none of us voted for it. There was no referendum. Nothing. Apparently this was UN policy and we had no choice.
Was it UN policy? Was it adopted in a resolution? Or maybe was it that somebody suggested that this might happen in response to that and the other problem might not happen?
It was globally implemented in developed countries exactly as the UN recommended, without any local populations being informed or having any say on the matter. It was dictated by some globalist elites. We have an illusion of a democracy.
No, you don't get to say that and not be laughed at. You can believe it, but you can't claim it. Democracy isn't just voting for what you want or complaining you didn't get a vote. It's also running for office on a platform that you believe is important. Unless you got elected and tried to implement your ideas and men in black came to stop you, you don't get to complain.
You have completely missed the point. Whether I ran for office or not is completely irrelevant.
There are powerful influential people and entities in this world that make far reaching and wide impacting decisions without consulting or informing the public.
You have to be incredibly naive to believe all government officials, corporations, and financial elites are entirely transparent and honest
Perhaps you missed my question the first time. :))))
Are you as upset that young men are "wasting their lives grinding at meaningless soul crushing fake corporate jobs instead of raising families" as you are about women? Or are you just upset that the wimmenz escaped the kitchens they "belong" in?
If there were fewer people forced into work, maybe that would cut down the number of pointless/meaningless jobs.
Maybe it is better that only half of a couple are forced into corporate soul crushing jobs. It is subjective. Some women look at child raising / housewife duties as bondage and slavery. Natural selection rewards those who prioritize family. In any case we need generous maternity leave and options for part time employment.
Yeah so I don't think women "belong in kitchens shooting out babies". That is a total misunderstanding of the point.
EDIT: Personally I don't give a shit. If a woman wants to be an astronaut or a mathematician or a ruthless psychopath CEO then go for it. I would encourage it. However if all women dedicated all of their energy to a career and have no babies then we become extinct. This is simple mathematics. Just because I point out a simple obvious mathematical fact doesn't mean that I am God and I created the universe this way. It was already this way long before I showed up and will continue to be this way long after I am gone from this world. Don't shoot the messenger.
This outcome of course needs the volume of writing / illustration to stay constant.
Instead AI has just flooded the market with more content in a way that’s net additive. PLUS we’ll need to consider the energy impact of the AI we’ll all need to filter all this incoming AI spam.
Does society have an obligation to maintain a plethora of 'Poor Creatives'? What if a portion of society just decided that they must absolutely be paid for their hobby, or rich people are evil.
Capitalism has really done a number on the human pshcye.
Luckily, my occupation of collecting gains, dividends, and interest is on the order of playing Snakes & Ladders, so it'll (at least according to Munroe) be difficult for an AI to outdo me: https://xkcd.com/1002/
What else was the human doing besides generating text and images while that CO2 was being produced? Is the human supposed to cease producing CO2 when they stop generating text and images?
This is some galaxy-brained shit. All this study really says is you should be using pen and paper.
Well, of course it is. The information density is less in AI writing, and as more AI writers are trained on more AI writing, will approach random entropy. This is merely thermodynamics at work. /s
> For the human writing process, we looked at humans’ total annual carbon footprints, and then took a subset of that annual footprint based on how much time they spent writing.
So, they aren't asking "does a human painter use more electricity to do a painting than an AI?" ... they are asking "is it more energy efficient for an AI to do a painting, or for a human to exist for several hours?".
That is, it doesn't take into account that the human would exist and have a carbon footprint anyway, even if not employed as an artist. It's devised as if employers created their employees, and seems to speculate about the potential energy saving in exterminating them.
> emissions of the computing devices used during the writing process are also considered
And their graphs show this as exceeding AI emission by a lot (due to AI being quick). This is a fair point, but they only needed this part. They shouldn't have thrown in the person's annual carbon footprint on top of this. Also, traditional art is presumably much less expensive in carbon.
And as somebody else pointed out, the use of devices will be part of the carbon footprint data in the first place, so it gets (partly) double-counted.