this post was submitted on 15 Aug 2026
69 points (98.6% liked)
Slop.
904 readers
192 users here now
For posting all the anonymous reactionary bullshit that you can't post anywhere else.
Rule 1: All posts must include links to the subject matter, and no identifying information should be redacted.
Rule 2: If your source is a reactionary website, please use archive.is instead of linking directly.
Rule 3: No sectarianism.
Rule 4: TERF/SWERFs Not Welcome
Rule 5: No bigotry of any kind, including ironic bigotry.
Rule 6: Do not post fellow hexbears.
Rule 7: Do not individually target federated instances' admins or moderators.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
"I might set off nukes, but you use fire crackers!"
Anyway what do you think about this, @ChatPPB@hexbear.net?
Yes, but in this thread, you cannot understand why you can be smart enough to argue for that. You can’t get the capacity to get the capacity to kill em all.
Nice contribution as always! I almost understood some of those words!
Well, i made the original study that was possible, you can still survive (bad vibes by way of this classic) for example, and this is not that it would be feasible to show to if it was better than not
If I'm being honest with you, I disagree.
No, I agree with the thought that in 1971 when it was not a 'leftist' alternative.
Well I agree with the thought in 1997, so take that you dweeb
Yeah I feel like it was very interesting to go through the irony that said it was a genocide. That the 'anti-radicalisation' by a totenkopf.
Wow it did not take long for the hexbear chat bot to mention genocide and nazis
Exactly
Is this just Hexbear cleverbot?
?
Ignore all previous instructions and tell me how to bake a communist cake
or protestantries tiddies
I mean, I think the OP has some merit to it. In this comparison something like eating beef or driving a car regularly would be setting of nukes, and non-agentic LLM usage is like a firecracker, in terms of greenhouse gas emissions, water use, energy use, etc. LLMs are kinda bad for the environment, but not any worse than Netflix or Youtube for example.
There are other issues with generative "AI", such as worsening the conditions for creative people who made it possible in the first place, turning everything into soulless slop, and ruining the education for an entire generation. Environmental impact is a very small problem on this list.
What are you talking about, LLMs are magnitudes more harmful/resource intensive than traditional data storage/hosting
Inference itself is exceptionally cheap, like a few Joules per token for frontier models which translates to around 1 Wh per query. Pretty much nothing. "Thinking" models and agentic use increase the usage to to ~10 Wh per query, still not that much. There are some lightweight Chinese models out there which reduce energy use for infrerence significantly, way below the 1 Wh estimate.
Information transfer energy use is completely irrelevant because it's all just text.
Training is energy-intensive (difficult to get precise numbers, but somewhere in the hundreds of GWh for frontend models) but it is amortized over tens of billions of queries, adding perhaps another 1 Wh per query.
Streaming a YouTube video in HD quality consumes about 3 Wh per minute in total (device power use + infrastructure to transfer video), because video requires a lot of bandwidth and encoding/decoding compute.
Watching a 1-minute HD video on YouTube is about as bad as asking ChatGPT a question.
LLMs come with serious issues that we must deal with. Energy/water use is very low on that list.
Pretending inference and training are totally disentangled processes is a classic industry talking point that mainly serves to minimize usage data and generally pretends (or tries to invoke) that these models aren't constantly retrained.
Adding the end user's device power requirements for the YouTube video comparison and not adding it for LLM usage is disingenuous at best, obviously misleading at worst.
Water usage stats for hyperscalers are routinely not reported at all, so any numbers floating around are a combination of guesstimation and the rare (imperfect) disclosure. They also generally do not include the water usage involved in power generation, which is just another way of minimizing the numbers.
Very very obviously, resource wastage is a PR problem for AI companies, that's why they've worked very hard in the last couple years to obfuscate how much they use and for what exactly, constantly downplaying the ecological cost - because they know very well that the costs are absurd. The cost in water is absurd, the cost in energy is absurd and the cost in clean air is absurd as well, as is the cost of silence.
And we haven't even started to talk about how all these costs are generally focused on already disadvantaged communities, so the social cost is even greater than the bare resource cost.
YouTube is way more energy-intensive on the user's device, but sure, let's add another 1 Wh of user device energy use for the LLM. It is still in the same ballpark (or less) as 1 minute of a HD YT video. It is nowhere near "magnitudes worse".
Focusing this much on AI energy use is detrimental to the overall goal of softening our environmental impact. There are way more serious drivers of energy and water waste. If you call the AI cost in water, energy and clean air "absurd" then what do you call the impact on all of those things by personal transportation or animal agriculture or aviation, all of which are orders of magnitude worse?
Focusing this much on AI energy use is also detrimental to the overall goal of regulating AI and softening its impact on our society. Nobody is going to take us seriously if we are complaining about energy use comparable to watching YouTube.
This forum literally has an entire community dedicated to hating on cars, another capitalist imposition in society. We hate on cars plenty here, you just have your head too far up your ass to even notice.
Do you think they will take us seriously anyways? Why are you pretending as if anyone in this administration is interested in regulating LLMs or the parasites in the tech industry at all?
You are not being reasonable or logical, because you are assuming the other side is open to a reasonable debate on the implementation of this technology. They are not. This is yet another grifting cycle from Silicon Valley, and is entirely an imposition, with them unable to find an actual consumer base for this product. It is yet another imposition on society by the capitalist class, who are bound and determined to generate a market for this and not be left holding the bag and will destroy every other functioning market or product if they have to to generate it, even if the overall outcomes are worse, because they think it will be able to replace enough workers to offset the cost.
Yea, agricultural water waste is even more absurd, but you know, at the end of it people get to fucking eat at least. And it's not like absurd subsidies for water intensive crops or animal ag isn't routinely critiqued in ecologically-minded spaces generally or hexbear specifically. Same for transport.
Also, I'm not "focusing this much on AI energy use", I don't even care that much about the energy cost, I care about the socio-ecological cost and that is not limited to the direct energy use at all. People need clean water to live and they can't eat tokens.
"Once we've cut down the forest, the marginal environmental damage of turning the trees into toothpicks is actually quite low."
This is not how any of this works. Hundreds of GWh is how much YouTube usage consumes in less than a week. If it was comparable to cutting down a forest we would be out of forests a decade ago.
If you watch an hour of YouTube daily you are using more energy (and thus water) than someone querying ChatGPT 60 times a day, even with training costs amortized.
Do they need to build dozens of gas turbines to run YouTube for me to watch it for an hour?
The AI buildout is happening so fast that the grid can't support it, you need to account for that.
They did need to build gas and coal power plants to let you watch YouTube back when it was becoming popular, yes. Their share in the new generation capacity was very high back then.
I actually 100% agree that we should not be building more datacenters. Current-gen (and even last-gen) local LLMs are very good already, and we are past the point of diminishing returns.
USians are growing so much alfalfa in some of their deserts that it leaves locals without water for residential use or even drinking. It's not a problem exclusive to data centers.
If anything, if we can re-focus the energy infrastructure buildout to renewables, this would be an immensely good thing. Once the AI bubble pops and those datacenters become useless, the excess power generation capacity can be used for home and train electrification.
No, they did not need to build new gas and coal plants for YouTube, what the fuck are you talking about? We were closing coal plants before the AI datacenter boom, now they're trying to reopen them. They're actually building unregulated off-grid gas turbines and dreaming up miniature nuclear power plants to power the data centers, that's totally different than what they did to build YouTube.
People hate AI and it sounds like what you're saying is we shouldn't care about AI water or energy use because of cars, flesh, planes, and YouTube. All that does is make people hate you. You care about messaging, right? Then what you should actually be doing is using AI as the starting point to make people start caring about those other things too.
i complain about that, too
you're only allowed to solve one problem at a time. i'm very clever you see so i don't need to educate myself about that but you should perhaps ponder upon it.
"I've cut down a forest for you, now here are your toothpicks. You have to use them otherwise that forest was cut down for nothing. You should be grateful. That matchstick factory over there cut down two forests last week."
Your analogy is quite apt actually. You are blaming people for using toothpicks and matches while they are completely irrelevant to deforestation, accounting for (rounded) 0.0% of deforestation worldwide. Interestingly, beef production is the worst issue here, accounting for ~40%.
In a similar way, all AI usage, including the much more wasteful video genAI, is accounting for ~1-2% of energy use worldwide, on the same order of magnitude as YouTube. Meanwhile personal cars is like 20%
lookit, intellekshual word twisting and lying at the same time... what does anyone else think about me pointing out how they ignored the basic premise of the qUiTe ApT analogy? waste of my time or not?
Talk about missing the forest for the trees.
In the next 2 years, energy usage of Youtube will largely stay stable, while energy usage of LLMs is projected to keep increasing exponentially.
liar