Using ChatGPT is not bad for the environment | Hacker News Reader

Using ChatGPT is not bad for the environment

andymasley.substack.com

255 pointsreturningfory2320 comments

Yet we also see that hyperscale cloud emissions targets have been reversed due to AI investment, Datacenter growth is hitting grid capacity limits in many regions, and peaker plant and other non-renewable resources on the grid are being deployed more to handle this specific growth from AI. I think the author, by qualifying on "chatgpt" maybe can make the claims they are making but I don't believe the larger argument would hold for AI as a whole or when you convert the electricity use to emissions.

I'm personally on the side that the ROI will probably work out in the long run but not by minimizing the potential impact and keeping the focus on how we can make this technology (currently in its infancy) more efficient. [edit wording]

Is that true though? Data centers can be placed anywhere in the USA, they could be placed near a bunch of hydro or wind farm resources in the western grid which has little coal anyways outside of one line from Utah to socal. The AI doesn’t have to be located anywhere near to where it is used since fiber is probably easier to run than a high voltage power line.

That was already done years ago and people are predicting that the grid will be maxed out soon.

Build new data centers near sources of power, and grid capacity isn’t going to be a problem. Heck, American industry used to follow that (building garment factories on fast moving rivers before electricity was much of a thing, Boeing grew up in the northwest due to cheap aluminum helped out by hydro). Why is AI somehow different from an airplane?

They'll have to build new power generation and build the data centers next to it.

You are massively conflating what is possible with what is done.

I know data centers are built next to wind farms for these reasons already. We have an abundance of those projects out here in the PNW.

There are a large number of reasons the AI datacenters are geographically distributed--just to list a few off the top of my head which come up as top drivers: latency, data sovereignty, resilience, grid capacity, renewable energy availability.

Why does latency matter for a model that responds in 10s of seconds? Latency to a datacenter is measured in 10s or 100s of milliseconds, which is 3-4 orders of magnitude less.

Two reasons that I understand 1. not all these AIs are LLMs and many have much lower latency SLAs than chat and 2. These are just one part of a service architecture and when you have multiple latencies across the stack they tend to have multiplicative effects.

If you look at a model with a diverse competitive provider set like llama 3 the latency is 1/4 second, and it will definitely improve at a minimum incrementally if quality is held constant: https://artificialanalysis.ai/models/llama-3-3-instruct-70b/... Remember that as long as you experience the response linearly (very much the case for audio output for eg) then the first-chunk latency is your actual latency, not to stream the entire response.

The root problem there is that fossil energy is very cheap and the state sponsors production of fossil fuels. Consequently the energy incentives are weak and other concerns take priority.

This is coupled with low public awareness, many people don't understand the moral problem in using fossils so the PR penalty from a fossil powered data center is low.

Voluntary conservation was only working by accident and guilt tripping never works. The grid needs to become clean so that we can have new industries.

Yep, this is the real answer. It's also the only answer. The big fiction was everyone getting hopped on the idea that "karma" was going to be real, and people's virtue would be correctly identified by overt environmentalism rather then action.

Fossil fuel companies won, and they won in about 1980s when BP paid an advertising firm to come up with "personal carbon footprint" as a meaningful metric. Basically destroyed environmentalism since...well I'll let you know when it stops.

I made a point in the post to say that it's better to mostly ignore your personal carbon footprint and focus on systematic change, but that I was writing the post for people who still wanted to reduce their consumption anyway

It's a false dichotomy to say "either systemic change or individual change" - both have always and will always go hand in hand, influencing each other in the process. To say only systemic change is required leaves out the individual responsibility for those who have the means to choose. To say it's just individual change required leaves out the fact that people can only choose within the reality of their situation, which clearly is defined by the outcome of the system they are in.

> for those who have the means to choose

interesting qualification.

maybe conservation should start with the masters of the universe who fly private jets all over the world etc. and emit more than the rest of us do all year in a matter of hours.

The grid being clean means not having any fossil power. We can only get there by shutting down all fossil fuel power plants.

We can not get there by adding new power generation.

Well you need the latter to replace the former. So you need to add new power generation to allow you to shut down fossil fuel plants.

And to be honest what we need to do is replace them with nuclear power stations to manages the base load of nations power requirements. Either that or much better power storage is required

> We can only get there by shutting down all fossil fuel power plants. We can not get there by adding new power generation.

https://knowyourmeme.com/photos/1433498-no-take-only-throw

"No add new power plants, only transform our grid to greener".

Even if grid the was 100% renewable, this does not mean that there's no environmental cost to producing electricity. As a society, we need to decide what is important and try to minize energy consumption for things that are not important.

And shoving LLMs into every nook and cranny of every application, so just tech giants who run the data centers can make more money and some middle managers get automatic summaries of their unnecessary video calls and emails is, I would argue, not important.

But once again, the fundamental issue is late-stage capitalism.

What's the upside of moralizing energy consumption, especially once it's 100% renewable. Why not just let the market decide? If I'm paying for it, why does anyone else get a say in how I use it?

I kind of agree with this sentiment , though we haven't reached 100% renewable ...

Isn't that kind of a non-sequitur? The claim made was that renewable energy would still be a finite resource to some degree. It's possible that the available energy surplus will be too big for any decisions about usage to matter, but that's a strong claim and you're doing nothing to make it here.

A lot of people believe in a higher power. If trusting in this supposed "market" brings you comfort and clarity in a complicated world, I do not begrudge you it. But invoking it doesn't address the claim it's answering

It's also clear that "the market" does not care enough about environmental impact to even do stuff like remove the current significant fossil fuel subsidies present in most government budgets, nor stop individuals or organizations from consuming or selling said fuels, natural gas, or plastic products at massive scales, so it's unclear why it would allocate energy in a way that didn't deprive crucial priorities.

Like the theodicy on the invisible hand's problem of environmental collapse ain't lookin' good is all I'm saying

> What's the upside of moralizing energy consumption, especially once it's 100% renewable.

Because of the reason I just explained above. 100% renewable energy does not mean that producing that energy does not have an environmental cost.

> Why not just let the market decide? If I'm paying for it, why does anyone else get a say in how I use it?

Because the "free market" does not internalize externalities (costs to shared resources like the environment, air quality, public health, etc.).

Focussing on pricing those externalities (tiny as they'll be in a 100% renewable world), through laws + policy, is a better strategy than trying to convince people not to use their electricity for <thing I personally don't value>.

Having LLMs everywhere haven't helped me much, it just gets in the way.

Emissions are a collective action problem. Guilt tripping works poorly directly on behaviour but it works on awareness&public discourse -> voting -> policy.

Cf how we addressed the ozone hole, acid rain, slavery, etc.

Why do you belive this? Datacenter uses just a 1-1.3 percent of electricity from grid and even if you suppose AI increased the usage by 2x(which I really doubt), the number would still be tiny.

Also AI training is easiest workload to regulate, as you can only train when you have cheaper green energy.

I also had doubts, but asked chat and it confirms it’s an issue - including sources.

https://chatgpt.com/share/678b6b3e-9708-8009-bcad-8ba84a5145...

The issue is that they are often localised, so even if it’s just 1% of power, it can cause issues.

Still, by itself, grid issues don’t mean climate issues. And any argument complaining about a co2 cost should also consider alternative cost to be reliable. Even if ai was causing 1% or 2% or 10% of energy use, the real question is how much it saves by making society more efficient. And even if it wasn’t, it’s again more of a question about energy companies polluting with co2.

Microsoft, which hosts OpenAI, is famously amazing in terms of their co2 emissions - so far they were going way beyond what other companies were doing.

What do you mean by confirms the issue? What's the issue exactly?

The issue is that when you have a high local usage your grid loses the ability to respond to peaks since that capacity is now always in use. Essentially it raises the baseline use which means your elasticity is pretty much gone.

A grid isn't a magic battery that is always there, it is constantly fluctuating, regardless of the intent of producers and consumers. You need to be able to have enough elasticity to deal with that fact. Changing that is hard (and expensive), but it is the only way (such as the technical reality).

The solution is not to create say, 1000 extra coal-fired generating facilities since you can't really turn them on or off at will. Same goes for gas, nuclear etc. You'd need a few of them for your baseline load (combined with other sources like solar, wind, hydro, whatever) and then make sure you have your non-renewable sources have margin and redundancy and use storage for the rest. This was always the case, and it will always be the case.

But now with information technology, the degree to which you can permanently raise demand on the grid to an extreme degree is where the problem becomes much more apparent. And because it's not manufacturing (which is an extreme consumer of energy) you don't really get the "run on lower output" option. You can't have an LLM do "just a little bit of inferencing". Just like you can't have your Netflix send only half a movie to "save power".

In the past we had the luxury of nighttime lower demand which means industry could up their usage, but datacenters don't sleep at night. And they also can't wait for batch processing during the day.

Except neither ChatGPT and nor sources say this. First source says:

> Gas-fired generation could meet data centers’ immediate power needs and transition to backup generation over time, panelists told the Northwest Power and Conservation Council.

What you are saying has nothing to do with local, but has to do with large abrupt changes in electricity usage, and datacenter electricity usage is generally more predictible and smooth than most other industry.

I'm not talking about fluctuations (i.e. a datacenter with fluctuating usage). I'm taking about adding a datacenter to an existing grid. That significantly changes the baseline load on the grid, and that is a local problem because transmission is not universally even across an entire grid.

If your transmission line is saturated, it doesn't matter how much more generation you add on the source end, it's not gonna deliver 'more' over the transmission lines.

And that is just a simplistic local example, because it's not a single producer, single consumer, single transmission line scenario. ChatGPT and the article aren't diving in to that. The closest they might get is congestion but even then you already have to know the issue to be able to ask about it.

As far as the article itself is involved here, this tread mostly goes into the reason why global usage percentages doesn't mean there are no problems. It's like saying gerrymandering has no impact because of some percentages elsewhere.

ChatGPT didn't "confirm" anything there. It is not a meaninful reference.

This is a great article for discussion. However articles like this must link to references. It is one thing to assert, another to prove. I do agree that heating/cooling, car and transport use, and diet play massive roles in climate change that should not be subsumed by other debates.

The flip side to the authors argument is that LLMs are not only used by home users doing 20 searches a day. Governments and Mega-Corporations are chewing through GPU hours on god-knows-what. New nuclear and other power facilities are being proposed to power their use, this is not insignificant. Schneider Electric predicts 93 GW of energy spent on AI by 2028. https://www.powerelectronicsnews.com/schneider-electric-pred...

> Governments and Mega-Corporations are chewing through GPU hours on god-knows-what.

The "I don't know so it must be huge" argument?

Not knowing what it's being spent on is separate to knowing whether it's being spent

I’m not an expert, but I’ve seen multiple reports that predict very large increases in electricity demand.

https://www.goldmansachs.com/insights/articles/AI-poised-to-...

The question this is addressing concerns personal use. Is it ethical to use ChatGPT on a personal basis? A surprising number of people will say that it isn't because of the energy and water usage of those prompts.

I would be surprised if many people said it is unethical to use LLMs like ChatGPT for environmental reasons, as opposed to ethical principles such as encouraging unfair use of IP and copyright violation.

Still, LLM queries are not made equal. The environmental justification does not take into account for models querying other services, like the famous case where a single ChatGPT query resulted in thousands of HTTP requests.

I see people complaining that ChatGPT usage is unethical for environmental reasons all the time. Here's just the first example I found from a Bluesky search (this one focuses on water usage): https://bsky.app/profile/theferocity.bsky.social/post/3lfckq...

"the famous case where a single ChatGPT query resulted in thousands of HTTP requests"

Can you provide more information about that? I don't remember gearing about that one - was it a case of someone using ChatGPT to write code and not reviewing the result?

> Can you provide more information about that?

Sure, I looked it up and it was a security advisory: https://github.com/bf/security-advisories/blob/main/2025-01-...

I don’t have the time to re-read it now so if I am misremembering and it is a training-time effect rather than query-time effect then my bad, though the point largely stands.

Bluesky is a special place, I am not surprised you will bump into the sort of activists who will critique every bucket of water that go into making a hamburger there. For those of us who avoid contact with the axe grinding hoards of short form media, meeting somebody who spends any portion of their day fretting about the water usage of language models is indeed a rarity.

It’s not specific to Bluesky. I’ve seen it on Threads, X, Facebook, and Reddit. It’s a talking point for people who hate AI and they tell anybody who will listen.

> I would be surprised if many people said it is unethical to use LLMs like ChatGPT for environmental reasons, as opposed to ethical principles such as encouraging unfair use of IP and copyright violation.

Usually they complain about both.

I feel it's great that people have gotten invested in energy use this way, even if it's a bit lopsisded. We should use it in a positive way to get public opinion and political overton window behind rapid decarbonization and closure of oil fields.

> However articles like this must link to references.

There are links to sources for every piece of data in the article.

Where?

One of the most crucial points "Training an AI model emits as much as 200 plane flights from New York to San Francisco"

This seems to come from this blog https://icecat.com/blog/is-ai-truly-a-sustainable-choice/#:~....

which refers to this article https://www.technologyreview.com/2019/06/06/239031/training-...

which is talking about models like *GPT-2, BERT, and ELMo* -- _5+ year old models_ at this point.

The keystone statement is incredibly vague, and likely misleading. What is "an AI model"? From what I found, this is referring to GPT-2,

If I understand TFA correctly that's a claim it's covering and arguing against, not arguing for.

By TFA do you mean the author of the article? It seems to be using an outdated [and incorrect] claim (as far as I know, GPT-4 has no note of taking 200 flights of energy to train), arguing against it saying that those numbers are especially small, when they are potentially significantly larger.

I assume this comes from the 60GWh figure, which does translate to about 200 flights (assuming energy density of gasoline; in actual CO2 emissions it was probably less since likely cleaner energy was used than for running planes).

The link the article uses to source the 60 GWh claim (1) appears to be broken, but all of the other sources I found give similar numbers, for example (2) which gives 50 GWh. This is specifically to train GPT-4, GPT-3 was estimated to have taken 1,287 MWh in (3), so the 50 GWh number seems reasonable.

I couldn't find any great sources for the 200 plane flights number (and as you point out the article doesn't source this either), but I asked o1 to crunch the numbers (4) and it came up with a similar figure (50-300 flights depending on the size of the plane). I was curious if the numbers would be different if you considered emissions instead of directly converting jet fuel energy to watt hours, but the end result was basically the same.

[1] https://www.numenta.com/blog/2023/08/10/ai-is-harming-our-pl...

[2] https://www.ri.se/en/news/blog/generative-ai-does-not-run-on...

[3] https://knowledge.wharton.upenn.edu/article/the-hidden-cost-...

[4] https://chatgpt.com/share/678b6178-d0e4-800d-a12b-c319e324d2...

Just 200 flights? I would expected a number at least 100 times that. 200 flights of that range are what, 0.1% of a single day of global air traffic?

All of that is crazy in terms of environmental destruction but this makes AI training seem nothing to focus on to me.

I'm the author, the 200 flights number is taken from posts made by environmentalists and seem to match public numbers given (about 50 GWh) https://www.forbes.com/sites/arielcohen/2024/05/23/ai-is-pus....

If you think the numbers I used are wildly off I'd really appreciate any source saying so and I'll update the post with the correct amounts.

One miss in this post is that the author tries to make their point by comparing energy consumption of LLMs to arbitrary points of reference. We should be comparing them to their relevant parallels.

Comparing a ChatGPT query to an hour long Zoom call isn't useful. The call might take up ~1700 mL of water, but that is still wildly more efficient than what we used to do prior - travel/commute to meet in person. The "10x a Google search" point is relevant because for many of the use cases mentioned in this post and others like it (e.g. "try asking factual questions!"), you could just as easily get that with 1 Google search and skimming the results.

I have found use for LLMs in software development, but I'd be lying if I said I couldn't live without it. Almost every use case of an LLM has a simple alternative - often just employing critical thinking or learning a new skill.

It feels like this post is a long way of saying "yes, there are negative impacts, but I value my time more".

I basically do think that at some threshold it's important to weigh your time against negative impacts. I personally avoid taking flights whenever I can because of the climate and think that's worth my time relative to the emissions saved, but I also never worry about optimizing the energy use of my digital clock because that would take too much time relative to the emissions I could save. ChatGPT exists somewhere between those two things, and my argument in the post is that it's much closer to the digital clock.

On a logarithmic scale, it’s closer to the flight.

Flying 1000 miles commercially only represents about 10 gallons of fuel.

10 gallons of fuel's worth of energy could be used to ask ChatGTP 100,000 questions (assuming 3 Wh per question) or power a digital clock (1-2 W) for 35 years. If you assume you ask ChatGPT 8 questions per day, it's using exactly as much energy as a digital clock. Personal use of ChatGPT is much closer to the clock.

The marginal difference of someone sitting or not sitting on a specific flight isn’t 10 gallons of fuel. It’s the capacity of the entire aircraft making that trip / number of passengers that gives 10 gallons on average because if more people make a trip eventually you need an extra aircraft.

So on an apples to apples comparison ChatGPT including training, hardware, etc uses a fuck of a lot more than just 3 Wh per question.

The training cost adds a minuscule amount to the marginal cost of each search. The best sources I can find on training GPT-4 say it used 50 GWh of energy to train. GPT-4 is being used for a lot more than ChatGPT, but even if we assume that ChatGPT is all it's used for if we divide the total cost of energy by the total number of ChatGPT searches it increases the cost of each search by 10% (I go into this in the post) raising the marginal cost including training from 3 Wh to 3.1 Wh.

Sorry but this doesn't add up and it's not fixable because the argument is broken and I doubt the numbers are even in the right order of magnitude when considered in context. (That's before considering that it also had the effect of you writing and sharing this post - what's the second-order-effecf of people seeing it increasing their use as result of that and resharing the same idea with their peers etc? What's the network effect of your digital clock?)

You are rationalizing your laziness in an attempt to cope with the negative effects that you truly deep down know are there.

If I'm going to be accused of rationalizing my laziness I need at least one source saying my numbers are off. I'm going off the numbers provided by environmentalist critics of ChatGPT. I hope that by sharing my post people feel exactly as guilty about using ChatGPT as they do about buying one additional digital clock.

Sounds to me like you're rationalizing your laziness in researching and sharing more accurate numbers because you want to justify "the negative effects that you truly deep down know are there".

You appear to be arguing based on vibes.

Not so fast, billions of investment wasn’t spent on just 50 GWh of electricity.

The final step of training one specific model might have used 50 GWh which only costs what 0.011 billion? What the fuck do you think they are spending the rest of that money on?

And what do you think environmental impact of that spending was?

Most of that money went into infrastructure + human talent, with energy costs at the tail end. I'd be happy to include the energy costs of producing the materials, feeding the humans and moving them around etc. but those aren't usually included in environmentalist critiques of the energy used in training. I couldn't use the original article to unpack every possible interpretation of how much energy was used without it turning into a book. If we did the same for plane rides we'd also need to include the energy cost of assembling the planes, training and feeding the human pilots etc.

> feeding the humans

You shouldn't include this, because they would exist and eat anyway, and probably the same thing (unless, maybe, their habits are changed).

Fair enough, I’m not saying you have to do the analysis.

I was working from a rough order of magnitude estimate from their spending and calling their numbers BS.

Airlines are a better studied industry so you can find academic literature or use direct fuel costs which is a large enough cost you can get fairly close just using it alone.

The section on training feels weak, and that's what the discussion is mainly about.

Many companies are now trying to train models as big as GPT-4. OpenAI is training models that may well be even much larger than GPT-4 (o1 and o3). Framing it as a one-time cost doesn't seem accurate - it doesn't look like the big companies will stop training new ones any time soon, they'll keep doing it. So one model might only be used half a year. And many models may not end up used at all. This might stop at some point, but that's hypothetical.

I am sure that’s intentional, because this article is the same thing we see from e/acc personalities any time the environmental impact is brought up.

Deflection away from what actually uses power and pretending the entire system is just an API like anything else.

I am to put it mildly not an e/acc and referenced being very worried about other risks from advanced AI in the article.

Then I would certainly be interested to know why you spent so much time making the same argument e/acc AI proponents make ad nauseam.

As it stands, the majority of your article reads like a debate against a strawman that is criticizing something they don't understand, rather than a refutation of any real criticism of environmental impact from the generative AI industry.

If your aim was to shut down bad faith criticism of AI from people who don't understand it, that's admirable and I'd understand the tone of the article, but certainly not the claim of the title.

The point of the article WAS to debate someone criticizing something they don't understand. AI as a whole is using a lot of energy and we should think about the environmental impacts, but I pretty regularly meet people who think that every individual ChatGPT search is uniquely bad for the environment. I tried to make it as clear as possible that that's the issue I'm responding to, not all AI energy use.

It briefly touches on training, but uses a seemingly misleading statistic that comes from (in reference to GPT-4) extremely smaller models.

This article [1] says that 300 [round-trip] flights are similar to training one AI model. Its reference of an AI model is a study done on 5-year-old models like BERT (110M parameters), Transformer (213M parameters), and GPT-2. Considering that models today may be more than a thousand times larger, this is an incredulous comparison.

Similar to the logic of "1 mile versus 60 miles in a massive cruise ship"... the article seems to be ironically making a very similar mistake.

[1] https://icecat.com/blog/is-ai-truly-a-sustainable-choice/#:~....

737-800 burns about 3t of fuel per hour. NYC-SFO is about 6h, so 18t of fuel. Jet fuel energy density is 43MJ/kg, so 774000 MJ per flight, which is 215 MWh. Assuming the 60 GWh figure is true (seems widely cited on the internets), it comes down to 279 one-way flights.

Thanks, I missed that 60 GWh figure. I got confused because the quotes around the statement, so I looked it up and couldn't find a quote. I realize now that he's quoting himself making that statement (and it's quite accurate)

I am surprised that, somehow, the statistic didn't change from GPT-2-era to GPT-4. Did GPUs really get that much more efficient? Or that study must have some problems

~90% of the plastic debris in the ocean comes from ten rivers [0]. eight are in china/SEA. millions and billions of single-use items are sitting in warehouses and on store shelves wrapped in plastic. even before the plastic is discarded, the factories these items are produced in dump metric tons of waste into the oceans/soil with little repercussion.

point is, none of our "personal lifestyle decisions" - not eating meat, not mining bitcoin, not using chatgpt, not driving cars - are a drop in the bucket compared to standard practice overseas manufacturing.

us privileged folks could "just boycott", "buy renewable", "vote with your wallet", etc, but sales will move to a less developed area and the pollution will continue. this is not to say that the environment isn't important - it's critically important. it's just to say that until corporations are forced to do things the right way, it's ludicrous to point fingers at each other and worry that what we do day-to-day is destroying the planet.

[0] https://pubs.acs.org/doi/10.1021/acs.est.7b02368

That's definitely not true. Let's take Americans, for example, driving their cars to work. Americans are about 15% of the world's emissions, of which 25% or so is transportation, of which well over half is cars. So you not driving to work is making direct impact on 2-3% of the world's overall emissions. Likewise, your decisions on all the other things, if taken in aggregate, will have a significant impact on overall emissions.

Meat and dairy specifically accounts for around 14.5% of global greenhouse gas emissions, according to the UN’s Food and Agricultural Organization (FAO).

If people collectively just ate a bit less meat and dairy, it would go a long way. Don't even have to be perfect. Just show a little bit of restraint.

Right just a little bit of restraint. On an unprecedented scale of coordination by hundreds of millions to billions of people - a scale of cooperation that has probably never occurred in human history (and there's no reason to believe it will any time soon).

But sure, if people "just" did a "little", it would go a long way. Just a _little_ restraint from the entire population all at once in perpetuity. No big deal.

You do this through the government. Implement meat rations. Everybody is only allowed so much meat per week. So you have to make it count.

I think this might be the first idea I've ever heard that I truly feel will backfire far worse than alcohol prohibition. I almost want to see it happen.

All families can't afford meat daily, let alone any type of meat they want. So, some families might favor chicken and certain processed meats here and there, but most cuts of beef and lamb for example might be out of the question, and so will daily meat consumption. The same applies for dairy, they might afford big blocks of tasteless cheese, but won't touch often cheese with taste, French or otherwise. There is no enforced rationing, but choice is reduced if not removed for a good fraction of the population, purely for economic reasons.

The USDA budget totals ~$200bn, of which 70% goes to SNAP (Supplemental Nutrition Assistance Program).

Table FNS-1, p58 https://www.usda.gov/sites/default/files/documents/2023-usda...

What's missing from this type of discussions is performance quantification and scientific attitude. It assumes solid food is indulgence and ultraprocessed vegan shakes paired with appropriate willpower is all it takes to do anything.

Things just don't work that way. Been there, done that. Willpower does nothing. Extra calories around your berry is extra calories around your berry. It's much more worthwhile to think about reducing GHG emission from farming than deceiving yourself into self-harming and projecting your own doing onto greater society.

> Implement meat rations.

I don't think we can ever be friends.

There's less extreme government action possible: tax meat & dairy higher.

There are already precedents for that, for example here in Germany cigarettes are taxed to about 70% and hard-liquor (altho it's more complicated to calculate as it depends on exact alcohol content) to about 40%.

Meat is only taxed at 7% via VAT (similar to sales tax in the US).

Vegetarian here. People who eat meat tend to reduce their consumption if you 1. charge them appropriately for it and 2. give them good alternatives. Telling them to not do something usually makes them angry unfortunately.

> Telling them to not do something usually makes them angry unfortunately

Then it's up to you to get even angrier.

Angry people are spiteful. I generally find it a lot more productive to get them to understand why their choices have consequences, how the can (progressively!) fix the problems, and watch things improve from there.

Environmentalists and those championing similar causes politically have taken this approach for decades. The result has been a perpetual shift to the right-wing who has no qualms about playing dirty, using fear, anger and other tribal emotions to gain support. Everywhere in the West, authoritarians are seizing power.

Good luck with continuing the virtuous approach while the world goes up in flames. It hasn't worked and has gotten us to where we are right now.

I think people should be angry about that, yes. I’m not particularly interested in tying eating meat to racism or whatever it takes to make people fearful enough to support you.

And this is why everything will keep going up in flames - you're not interested in doing what it has been shown to take to get people to support your cause.

Unfortunately it doesn't matter what we think people should be angry about. FWIW I obviously completely agree with you! People should be angry about those things! Turns out they aren't. What matters is the reality we live in.

I'd say accusing people of random stuff(racism,sexism and other isms) have been so overplayed in recent years, not only it doesn't work anymore, it works against you.

In the US, what Trump is doing is working. In the EU, what the authoritarian far-right parties are doing is working. Their tactics have been working incredibly well for them, use them, the inane pacifist take of "being above that" by environmentalists and progressives in general is what has caused them to be completely irrelevant and the rise of authoritarianism across the West.

I mean Hitler clearly did something right in getting elected. I'm not sure we want to emulate his policies, though.

How effective IYO was Morgan Spurlock (RIP) the activist filmmaker ("Supersize Me", [0]) at trying to educate the general (meat-eating) US public on US meat production, the restaurant industry, nutrition, marketing?

(PS:it's not a dichotomy between strict vegetarianism vs meat 4+ days/wk. Can encourage people to eat healthier and more sustainable.)

[0]: https://www.imdb.com/name/nm1041597/

I mean I’m not an expert but I think the general sentiment among Americans about meat is “yeah I hear meat production is gross/polluting/inhumane but I like meat and don’t want to think about that”. Charitably I think we have a lot of things we can care about and most people want to ignore this, plus there’s a lot of cultural baggage tied to diet people don’t want to leave. I think actual solutions will mostly involve the idea that reduction is good instead of throwing up your hands because you’re unable to commit to being vegan, and adding options to people’s diet that they actually enjoy. I think we have a very hard time solving “I can’t completely solve this and also it requires changes I don’t like so doing nothing at all is fine”.

He didn't try to educate. He sensationalized for profit. Supersize Me was a grift. He drank himself fat and fabricated his numbers.

Can't just dismiss his 17 films as a grift. Even accepting all the valid criticisms where he misled audiences, his shockumentaries did try to educate, somewhat, but also stretched facts and went for a personalized narrative, Michael-Moore-style. He did leave a legacy of nearly single-handedly reviving documentary as a film genre relevant to nutrition, one that mainstream audiences would watch.

His sequel "Super Size Me 2: Holy Chicken!" (2017) had fewer distortions and was pretty watchable and educational. Here's a review from https://www.agdaily.com/livestock/poultry/super-size-me-2-mo... which among other things covered in depth: "Many food labels, such as organic, natural, non-GMO, gluten-free, no added hormones, free range, green, artisan, antibiotic-free, are indeed quite misleading" and how those branding terms are misused in marketing to restaurant customers. I haven't seen any industry criticism of the segment with the two branding consultants.

As to the legacy of "Super Size Me", here's a review by an ag evangelist: https://www.eater.com/24173039/morgan-spurlock-legacy-super-...

Not by rationing, if you meant the US government: US per-capita consumption of meat is 328 lbs/year [UN FAO]. You'd try to reduce the subsidies and incentives for meat/ corn/ soybean production which are baked into the USDA budget ($200bn in 2024), since the 1930s (Great Depression), and more since the 1950s (and related stuff like the marketing tool of the "Food Pyramid"). These will be as politically hard to cancel as defense production or military bases or prisons.

Here's even a 2023 editorial from the Kansas City Star pointing the blame at Big Ag "Corn drives US food policy. But big business, not Midwestern farmers, reaps the reward" [0]:

> No, corn is not an evil crop, nor are farmers in the Corn Belt shady criminals. However, the devastating effect of corn owes to the industrialization of the plant by a small group of global agribusiness and food conglomerates, which acts as a kind of de facto corn cabal. These massive corporations — seed companies, crop and meat processors, commodity traders and household food and beverage brands — all survive on cheap commodity corn, which currently costs about 10 cents a pound. Corn’s versatility makes it the perfect crop to “scale” (commoditize, industrialize and financialize).

A libertarian take is the Cato Institute "Farm Bill Sows Dysfunction for American Agriculture" [1]:

> The Sprawling Farm Bill:... has its roots in the century-old New Deal and is revised by Congress every five years... "Supplemental Nutrition Assistance Program (SNAP), was added to the Farm Bill in 1973 to ensure support from rural and urban lawmakers, accounts for about three-quarters of the omnibus package. Some lawmakers and pundits have proposed splitting SNAP from the Farm Bill to stop the logrolling and facilitate a clearer debate on farm subsidy programs, which make up the rest of the bill." ...aslo criticizes crop insurance subsidies (which mainly go on the four big crops), Agriculture Risk Coverage (ARC) and Price Loss Coverage (PLC) ("Welfare for Wealthier Farmers"), Crop insurance subsidies (" originally envisioned as a more stable and cost-efficient alternative to ad hoc disaster payments, but they have acted more as a supplement than a replacement—and may have actually increased risks along the way.")

[0]: https://www.kansascity.com/opinion/readers-opinion/guest-com...

[1]: https://www.cato.org/policy-investigation/farm-bill-sows-dys...

That's an incredible amount of meat. That's 400g a day per person. I can't even eat that much meat in one go without feeling the meal is seriously unbalanced. Half that is a meat heavy meal for me. So that must mean a significant number people are eating huge amounts of meat for every meal. That or the waste is huge.

Yes that stat does make me wonder. Even if 21% of meat is wasted [0]. (Possibly they count "reaches a supermarket shelf" as = "not wasted", even if it isn't ultimately cooked or eaten?). But even then.

([1], the 2023 study by Tulane University supposedly finding "12% Of Americans 1/2 Of The Nation's Beef: How a mere 12% of Americans eat half the nation's beef, creating significant health and environmental impacts" is a red herring, all it says is that on any given day, some fraction of meat-eaters exceed portion sizes ("Those 12%—most likely to be men or people between the ages of 50 and 65—eat what researchers called a disproportionate amount of beef on a given day").

[0]: https://ballardbrief.byu.edu/issue-briefs/food-waste-in-the-...

> "Wasted food ranks as the number one material in US landfills, accounting for 24.1% of all municipal solid waste. Americans waste 21% of meat, 46% of fruit and veg, 35% of seafood, and 17% of dairy products. Altogether, Americans waste between 30% and 40% of the total US food supply."... "Poor packaging techniques causes 10% of grain products, 5% of seafood, and 4% of meat to be lost."

[1]: https://www.reddit.com/r/RedditForGrownups/comments/16cpmni/...

People, especially men, get irrationally angry by even the slightest thought of their precious meat being touched (including here on HN; there are more than a few crackpots over here who promote a meat only or even red meat only diet). If there would be a reason to violently overthrow the gov, this would probably be it; not for the things that actually matter to better or save humanity, but to save their daily meat platter.

There are mechanisms for such coordination. Diets have shifted around the world to more McD's, more western cuisine. Wild catch changes all the time as we overfish stocks to endangerment.

Always driven by money. McD's is cheaper and easier to mass produce. Wild catch changes when it is no longer economical to fish one kind of fish. There's never an "enlightenment" of people.

Internationally in markets undistorted by meat subsidies, plant-based food is cheaper and easier to mass produce...

- Ozone

- Efficient Lighting (LEDs now, but others before)

- Santa Claus

There are certainly examples of a large portion of the population of the planet working together towards common goals. A lot of people putting in a little bit of effort _does_ happen, and it _does_ produce results.

Smoking, drinking - both on the decline for the last decades.

There was a profit motive in all of those except Santa. For instance until the main players in CFCs (which were the main cause of ozone depletion) were able to invent, patent, and market an alternative to CFCs they successfully lobbied against any change. Then they got a new patent, turned 180 degrees, claiming CFCs were the worst thing to ever exist - and made a ton of money after they were banned. Same thing with stuff like asthma inhalers.

When people want to do something, we're unstoppable. But unfortunately that means the good and the bad, and right now polluting actions are incredibly beneficial, and the alternatives are mediocre and not only unprofitable, but generally incur a substantial cost. There's a reason places like the US/Canada/Australia talk a mean game about climate change, yet remain some of the biggest polluters per capita, by far.

Uh, Santa is the biggest advertising campaign for consumerism the world has ever known

I agree with you and that’s why I’m believe soy will conquer the humans nutrition sooner or later. It already is the undisputed king of world-scale animal nutrition because it’s cheap and super nutritious. The drawbacks in Brasil happens because of its success, replace by another crop and the problems will be even worse.

Replace animals proteins for humans by soy for human and you can divide world soy production by ~5.

Improvement of the ozone hole wasn't from millions of consumers making different choices, it was from government regulation and choices of the few leading corporations in particular industries.

Efficient lighting is a mix of regulation against worse lighting and individual consumer economic self-interest, lowering their electrical bill (and sometimes longer-lasting bulbs).

Neither of these are examples of large numbers of people choosing to sacrifice something for a common goal.

Yes, a big multi government marketing campaign "Just don't eat meat 1 or 2 days a week" can definitely work. E.g. friday is "Fish day" in a lot of places (education/public) due to Catholic church influences.

Just need a catchy word and campaign for "Veggie Wednesday".

Meatless Mondays are a thing.

Technically most hindus don't eat meat / maybe eats eggs rarely , So its definitely possible but it would probably have to be ingrained in the religion itself.

Another reason is that we had plenty of land and rice and wheat etc. just made more sense than eating meat I suppose.

I have never tasted meat , eggs sometimes but I am stopping that as well , not for hinduism but because I am so close to this ideal , might as well do it 100% lol

I am majorly convinced that there is no god but if I wish there was a god , I wish for Hindu deities (though I am obviously biased and this explains why people have such biases even being half atheists)

I was able to cut out 95% of meat without it being much trouble.

Greenhouse gas emissions are only a fraction of terrible things that humans are inflicting on the environment, and meat/dairy are both nutritious food that provides requirements for sustenance, and if not eaten need to be replaced by something else that will also cause greenhouse gas emissions (aka, a 10% reduction in meat consumption does not equal to a 1.45% reduction in greenhouse gas emissions)

I think it's kind of crazy to place the burden of environmental destruction on individual buying habits, rather than the people in power who actually have the ability to make sweeping changes that might actually move the needle.

Let's start with not incentivizing, then disincentivizing the mass production and importation of plastic garbage waste and e-waste that not only create greenhouse gas emissions but pollute the environment in other, irreversible ways.

And if your government and leaders don't make this a priority, and regardless of who you vote in, big-name corpo donors get their way instead, then maybe it's time for a new government.

It won’t be 1.45% but it will be 1.3%. Meat is exceptionally inefficient compared to non-meat. It sounds false but it really is like that :(

Meat and dairy are bad for GHG, but they are also terrible for most of the other bad things we do to the planet. Eutrophication, species extinction, habitat loss, water use.

And the people in power do what the people want, that's why they are in power. Imagine telling people they couldn't eat as much meat. Would be political suicide.

That's why individual buying habits are important. If many individuals change, we might change as a society. And at some point there might be a tipping point where the people in charge can make a change.

Also, blaming other people for all the problems is not a great way to solve a problem. Take some responsibility.

Not encouraging population growth everywhere but particularly in the highest per-capita consuming and polluting countries, but rather allow them to naturally level off and even gradually decline would go a much longer way. It would enable significant emissions reductions and reduction in all other environmental impacts of consumption without impacting quality of life.

Eating bugs and living in pods sounds great and all, but if the end result is just allowing the ruling class to pack more drones and consumers in like sardines then it's not really solving anything.

The US government could help by ending subsidies towards meat and dairy production, which will prevent those products from being artificially underpriced.

https://www.ewg.org/news-insights/news/2024/10/usda-livestoc...

Since my birth the population of earth has almost doubled.

This assumes all emissions / externalities are created equal, which they are not.

You are right. Though for CO2 that simplification comes pretty close to true.

Could you say more?

Are you talking about comparing CO2 to N2O to CH4 to fluorocarbons, for example?

Yes, that. But also that these different externalities will have different responses to reduction at scale, which also impacts how effective action on a more "personal" level really is.

For example the OP was talking about plastic. A 2% reduction in plastic waste has a clear benefit, because any amount of plastic reduction is a bonus. However it is not clear that a 5% reduction in CO2 emissions due to Americans driving their cars less will have any meaningful difference when it comes to climate change.

Still not understanding you, if you meant "a 2% reduction in plastic waste in country A" is of no global benefit if there's a comparable increase in country B (assuming their disposal is roughly equally bad for the environment globally, since only 9% of plastic globally is recycled). So I assume you meant "a 2% reduction in plastic waste, globally". But then the same criticism applies to country-level reduction in CO2 emissions.

Anyway, maybe the second Trump admin will exit the Paris Accord again, putting strain on China and India participating, so climate summits will be weaker till 2029 (earliest).

Likely the domestic economic impact (on the US) of major climate-related events will have more influence on US policy than international. e.g. whether one of the many causal factors to the huge 2025 LA fires was climate change; or California's water setup where Senior Drawing Rights legally predating the existence of the state of CA allow agribusiness essential unlimited water use to grow pistachios in a semidesert during a drought, or bottle and export artesian water. As well as all the other more proximate causes. Anyway, expect a long and politicized fight between state, federal, city and insurers about the LA fire root-causes, liability and federal compensation. Especially given there's a plan to weaken or abolish the US EPA at the same time.

Ditto, expect nation-level debates (in OECD countries) reappraising whether nuclear (alongside fossil-fuel and renewable) is considered safe, desirable and necessary and how that affects/limits GDP growth, and whether datacenters and AI continue to be seen as a proxy for GDP growth, or whether that assumption breaks down some year soon.

"Driving to work" is hardly a "vote with your wallet" style consumer choice. Our housing, building, and transportation policies have been geared towards encouraging car-dependence for nearly a century. In places with better public transit and bike lanes, people spontaneously choose to use those modes of transport. Just like with companies dumping as much plastic waste/CO2 as they can get away with, this is a policy problem, plain and simple. No amount of pro-environment metal straw campaigns will solve it. At best environmentally-conscious messaging could encourage changes in voting behavior which influence policy. At worst people could be convinced that they're "doing their part" and fail to consider systemic changes.

Regular voting is usually what affects things such as the transportation infrastructure in your country or city. It’s a slow proceed though.

Here in Oslo there has been a lot of investment in bike lanes, but just because one part of the local government builds more bike lanes doesn’t mean that other parts of the government will follow suit. Police still doesn’t care about cars illegally blocking the bike lanes. The people ploughing snow see bike lanes as the last thing that should need ploughing, preferably no earlier than 2 weeks after it snowed. A dedicated bike path I use to work is supposed to be ploughed within 2 hours of snow, but it took a week before anything was done and now three weeks later it’s still not to the standard that the government has set.

Speaking of Oslo and bicycles, I just want to add an amazing statistic that surfaced a couple of years ago here on Hacker News:

Oslo has a zero pedestrian and bicycle mortality rate!

https://thecityfix.com/blog/how-oslo-achieved-zero-pedestria...

> In 2015, the political climate and public will in the City of Oslo changed the tone on accepting continued surface transportation fatalities. The mayor, city council and transport division staff all supported a shift in roadway decision-making from car-centric to people-centric. [...]

Neighboring capitals with similar progressive bicycle cultures (Denmark, Sweden) have somewhat low but non-zero mortality rate as Oslo had 6 years ago. So the policies definitely make a change, but it's the consequence of a culture. You won't see American politicians suggesting a ban on cars in big cities.

We’ve had at least one cyclist killed since that article came out.

I would agree with you, but Americans intentionally reinforce car dependence whenever it's discussed.

It's bad enough that even non-US people regurgitate those talking points despite them being significantly less true for them; because they see it so much online.

> I would agree with you, but Americans intentionally reinforce car dependence whenever it's discussed.

They do, because their experience is that transit and biking really do suck and are useless. Which is accurate, for where they've lived.

The problem is that you have to convince people that things could be better, when their lived experience is that it's always terrible.

I live in place known for its rainy weather, 15 km from work (because of the housing crisis). Being overweight, biking to work never crossed my mind for two years. Once I tried to commute during weekend, as a challenge. I realized a few things: - same duration as the train - it give me the exercise I needed - it relaxes me - it is free since I already have a bike

Yes, it still take me 50 min to commute, but now I enjoy it and even volunteerly go to the office more often. It have been 6 months.

My point is: those who complain about biking being terrible or impractical should give it a real try. It may fit you.

And if you need help and can afford it, ebikes exist. Some of them really cheap. Hopefully not skimping on safeties though.

Ebikes are really great and I often ride one to work. They do help with speed, range, motivation, and of course, hills.

They don't do anything about unsafe infrastructure though, which is by far the US' biggest problem for biking. To say that American bike infrastructure is garbage tier is an insult to garbage, that's how bad it typically is. It's extremely unsafe, and people can feel it, they can tell, hence why hardly anyone actually bikes to work.

I've biked around in a bunch places in the US, and the reality is that it really is terrible. Bike infrastructure is F-tier almost everywhere, rising to D-Tier in supposedly bike friendly cities like Portland.

Bike infrastructure generally

* Is designed to be unsafe. Door zones are common, actual physical protection or segregation is rare, ESPECIALLY for intersections.

* Stops and starts randomly. Just look at Google maps for a city, you'll be able to see that the bike network is completely fragmented, with many bike lane suddenly disappearing on a road for no apparent reason.

* Randomly changes style/design principles even within the same city, so both you and drivers are constantly confused unless you're already used to a route.

* Is poorly enforced, with drivers routinely driving or parking in bike lanes with punishment being a rarity.

Now, some spots have good bike trails that work for their commute, and that's a great option when it's available. But I'm a bit tired of the "biking is actually okay in the US!" gaslighting.

Some people still manage, I do sometimes, but after getting hit by cars a couple times I tired of making excuses for it. There's a reason hardly anyone bikes for their commute in the states: overall, biking in the US is simply awful, and that's the truth.

I ride a bike, and doing so has saved me about $450k in transportation costs over 3 decades. The effort it takes to earn $450k is something to include amongst the unpleasantries and pathologies associated with driving.

Now, of course, I've had my whole life to set up my whole life the way I want it, and with a little foresight it really wasn't that difficult to set it up in a way that facilitates getting around on a bicycle. It involved making choices. Choices about where to work and live. If more people made such choices, there would be more options available to facilitate them.

450k tax-free, too, because it's a cost saving and those aren't taxable*.

Myself I've not driven at least since late 2017, thanks to excellent and cheap public transport; even before the Germany-wide €58/month ticket, the more expensive Berlin-AB ticket that I used to get was much much less than your $15k/year.

Do most people plan like that?

* usually, though IIRC buying a house in Switzerland gets you a tax equal to the money you saved from not renting it?

We also have a carbon tax where I’m from. Since I don’t own a car I don’t pay much into it. However I get a nice big rebate on my tax return. I basically get paid a small amount to not own a car.

I've ridden bikes for transportation in a handful of regions in the US. It is essentially always bad, it's just a matter of some people being able to look past or tolerate the awfulness.

Yes, some people are okay sharing a lane with cars, or using a strip of paint for protection, but accepting a poor status quo doesn't stop the status quo from being poor.

Occasionally there's a route that's legitimately good, almost always due to having an off street trail you can use for the bulk of the commute. Those can be great.

But once you hit bike lanes in an urban area, it's virtually always terrible. A lot of Americans are just used to the terribleness and don't notice it anymore, the same way they don't notice how almost every neighborhood is designed to be unwalkable (and how most are designed to be economically segregated besides).

> The problem is that you have to convince people that things could be better, when their lived experience is that it's always terrible.

Ironically, international air travel to places where it works great may help with this.

For some people yes.

But a lot of people are mentally stubborn, and seem to have inbuilt excuses of, "it wouldn't work here!" despite not spending even ten seconds thinking about it.

See, my point is that everyone first goes “it’s not me”, then they understand it is them and go “but it’s not my policies” and then they vote in the policies which are the problem. It’s totally fine to go “we need collective action to fix this”. But you have to actually join the collective action. You think billionaires are getting rich by committing environmental arbitrage? Then don’t oppose attempts to make the costs appropriate, even if you must now pay your fair share too.

Sure, recognizing that the problem is political is step one. Step two is... political activism, I guess? Lack of local political engagement and organizing is part of what allows problems like these to form.

Not just lack but outright apathy and villainizing of these attempts. If you try to tax gas some oil CEO will run a campaign explaining to the working class that their commute will now cost 10% more, which obviously makes people upset. Part of it is that we, unfortunately, can’t actually subsidize car commuters anymore. But part of it is that CEO is going to incur a cost of 50%, as he should, which is why he’s bothering to spend money riling people up.

It is not feasible to take a bike or public transportation to work in many parts of the world so a policy change like that would just incur an additional cost to those who have no alternative.

A better idea is to encourage other types of transportation in private companies rather than penalize existing companies with taxes. If taxing the companies will raise prices on consumers, you might as well consider it an additional tax on the people.

Redistribute the tax back to the people. That way you're eliminating the price externality but offsetting the cost to individuals. The fossil fuel industry can't continue to plug their ears and ignore the costs associated with putting more and more carbon in the atmosphere. We need to hold them to account.

The point is that you make the other alternatives affordable with such a fee. You can take the tax and put it directly into improving that infrastructure.

The emissions from vehicles are different from plastics produced by factories.

Also, while important, 2-3% of world emissions is a drop in the bucket compared to the other 97%. Let's consider the other causes and how we can fix them.

Think about this: for many people, not driving to work is a big deal. If people collectively decide to do that, that's a lot of effort and inconvenience just for 2-3%.

while 3% might sound like a drop in the bucket, there isn't any single specific chunk of the rest of the 97% that will immediately cut, say, 30-40% of emissions (also remember that 2-3% is the super specific "Americans not driving cars", not "everyone in the world not driving cars").

There isn’t really a magic wand we can wave and get 50% back for free and without inconvenience. The other 97% involves things like individually figuring out where our electricity generation goes. Or figuring out which farms to shut down, or what manufacturing we don’t like anymore. All of this must happen. It will be inconvenient. I picked a slice that is immediately relevant to a lot of people here. But there are a lot of axes to look at this.

> Let's take Americans, for example, driving their cars to work.

This is easily solved by switching to EVs. A small-size EV (perfect for personal transportation) is only slightly less CO2-efficient than rail ( https://ourworldindata.org/travel-carbon-footprint ).

I wish the world would ditch public transit entirely. It's nothing but a misery generator. It's far better to switch to remote work and distributed cities.

I unironically like public transport because someone else does the “driving” for me. I’m sure someone is going to explain how Waymo solves this but sitting on a train with my laptop and breakfast while I magically get teleported to my office is much nicer than even being in the back seat of a car.

> I magically get teleported to my office

You work on a train station platform? My condolences.

And indeed, self-driving will solve this. You'll be able to just step into a car, and get teleported to the door of your office. After stopping at your favorite coffee shop midway.

It doesn't, actually: there is no good place to work in a car. They don't have tables in them, and they accelerate both laterally and in the forward and back directions which makes most people carsick.

Public transit is not miserable everywhere. In Central European countries it can be quite enjoyable.

How much of Americans driving to work is because they choose too though? Amazon's 5 day RTO policy is a good example. How many of the people now going to an office 5 days a week would've done so without the mandate? I see the traffic every day, and saw the same area before the mandate, so I can tell you with confidence that there's many more cars on the road as a result of this commute. this all funnels back to the corporate decision to mandate 5 days in office.

Exactly. IMO, any politician who's serious about saving the environment or reducing the number of cars should be proposing bills to heavily tax employers for every unnecessary commute they require of their employees (maybe $100-$500 per employee per unnecessary day required in the office).

Employers are unfortunately very bad at deciding what is "necessary" work.

if taken in aggregate, will have a significant impact

This is a good sentiment. But, in context, it is a fallacy. A harmful one.

Consumer action on transport and whatnot, assuming a massive and persistent global awareness effort... has the potential of adding up to a rounding error.

Housing policy, transport policy, urban planning... these are what affects transport emissions. Not individual choices.

Look at our environmental history. Consumer choice has no wins.

It's propaganda. Role reversal. Something for certain organizations to do. It is not an actual effort to achieve environmental benefit.

We should be demanding governments clean up. Governments and NGOs should not be demanding that we clean up.

I think many Americans driving to work would be happy to work from home if not RTO mandates (encouraged by the government at least on a local level).

> That's definitely not true. Let's take Americans, for example, driving their cars to work.

Even an example like this that is carefully chosen to make consumers feel/act more responsible falls short. You want people to change their lives/careers to not drive? Ok, but most people already want to work from home, so even the personal “choice” about whether to drive a car is basically stuck like other issues pending government / corporate action, in this case to either improve transit or to divest from expensive commercial real estate. This is really obvious isn’t it?

Grabbing back our feeling of agency should not come at the expense of blaming the public under the ridiculous pretense of “educating” them, because after years of that it just obscures the issues and amounts to misinformation. Fwiw I’m more inclined to agree with admonishing consumers to “use gasoline responsibly!” than say, water usage arguments where cutting my shower in half is supposed to somehow fix decades of irresponsible farming, etc. But after a while, people mistrust the frame itself where consumers are blamed, and so we also need to think carefully about the way we conduct these arguments.

I didn't really carefully choose this, it was just what I came up. As others have mentioned, meat is another big one. FWIW I have no disagreement with letting people work from home, or pushing for other changes to make them less car-dependent.

The amount of people who choose to not drive to work is significantly impacted by policy.

This massively lets the Philippines off the hook. China has a gazillion people, and so does India, and the rest of SE Asia is bad for pollution, but the Philippines — with 1.5% of the world’s population — is an incredible 36% of ocean plastic pollution.

Also a call-out to Malaysia who are an upper-middle income country and contribute far too much per capita given their income situation, but again, they are a drop in the ocean compared to the (much, much poorer) Philippines.

Having spent half my life in South-East Asia, there’s a cultural problem that needs fixing there.

A pretty graph that make it clear just how bad the most egregious polluters are comparatively: https://ourworldindata.org/grapher/ocean-plastic-waste-per-c...

What are they doing in the Philippines? Dumping household waste straight into the rivers?

I think they are fishing commerially a lot. Most of the ocean trash comes from the fishing industry (e.g. abandoned nets).

Pretty much. Buying everything in sachets and then not having any real waste disposal; https://earth.org/philippines-plastic/

> it's just to say that until corporations are forced to do things the right way

But this isn't going to happen by itself. We need to vote for people who believes in regulating these corporations (rather than deregulating them).

It’s helpful to put this issue into perspective. But dismissing issues as not worth caring about on the grounds that there exist larger problems is fallacious and, to me, quite a dangerous way to live life.

“Why worry about your town’s water quality when some countries don’t have access to clean water?”

“Why go to the dentist for a cavity when some people have no teeth?“

“Why campaign for animal rights when there are some many human rights abuses going on?”

> "vote with your wallet", etc, but sales will move to a less developed area and the pollution will continue.

But voting with your wallet is literally moving sales to a more developed area with less pollution?

Per capita beef consumption is down by 35% in the US since the 70s. From 62kg/person/year to 35.

Beef produces ~100kg CO2 per kg of meat. That's a reduction of 27,000kgs of CO2 reduction, per capita.

That's not nothing. By simply reducing beef consumption by 1 kilogram a month, you can prevent more than a metric ton of CO2. If 5% of Americans cut 1 kilo of beef a month, that'd knock out 15 million tons of CO2.

Small changes can have an impact on aggregate, and just because someone else is not making those changes doesn't excuse us looking at ourselves and saying, "I could choose something else this time".

And 80% of all the trash in the oceans come from fishing industry (e.g. abandoned nets).

This is nothing but head-in-the-sand, arms-in-the-air, feel-good baloney to convince oneself to sleep well at night.

Guess what happens when you buy a used laptop instead of a new one?

That's right: less "standard practice overseas manufacturing".

Lifestyle change right there.

Buying less, using the same for longer, buying used goods instead of new are lifestyle changes that anyone can make and have an undeniable very clear impact by reducing the amount of stuff that needs to get made. Using my smartphone for 6 years instead of changing every 3 years doesn't mean the one I didn't buy gets sold elsewhere. It means one less sale.

I have always felt this way too. Our personal choices do not move the needle on fossil fuel and plastics. One could embrace aversion to these out of a sense of sustainability to signal virtue, but lets not pretend it will save the planet. It won't. Restricting aviation flights, stopping wars and minimizing the dirty fuel used in maritime freight does much more. But the world will not do it.

While I agree in general, my opinion is that customer choices do also matter and can move the needle, slowly, with larger cultural change.

Personally, trying to make better choices, big or small, isn't about "virtue signalling". It's about acknowledging the issues and living according to ones values.

Do we talk about pollution or emmissions?

This line of thinking is what undermines democracies and ruins the environment. Your choice might just be a drop in the ocean, but guess what the ocean is made out of.

Finally someone who speaks this out. What we do is more or less fly poop ... good four own well being but with almost zero impact. I'll go on doing some things because I think that some of that are the better ways to handle this or that or it's better for my health but with no expectation that I'll change anything.

This reads like an attempt to pass the blame to others. Per capita CO₂ emissions in the US are one of the highest in the world, and significantly higher than those in China or SEA. This is despite the US/Europe moving some of our dirtiest/cheapest manufacturing to that region.

Personal choices matter. See the amount of energy used on air conditioning in the US compared to areas of Europe with comparable weather for a banal example. If we want to significantly reduce emissions it will happen through a combination of personal choices, corporate action and government policy.

One of the most imminent problems with the environment isn't due to plastic pollution (which is of course terrible, might well have unforseen ramifications via micro plastics, and is impacting negatively biodiversity), but CO2 and other gases impacting climate.

While we should strive to fix both, it's more important in the short term to limit the amount of CO2 pollution before it's too late.

well, it's really not about the destruction of the planet but making our habitat more hostile and humans more sick.

sure, STEM will continue to find remedies and cures but at some point we're fucked just because the gene pool was reduced to an unnaturally selected bunch that survived & thrived completely alienated from the actual world.

sure, no biggie, wahaha, that's the name of the game, the old will die, the young repeat the same nonsense and that microbiome and all that other stuff we carry with us as hosts, potentially most likely in a beneficial symbiotic relationship, have no implicit mechanisms to cancel the contract and pivot towards some species or other that won't be d u m b enough to shit all over it's own home & garden, consequently ruining the bio-chemistry with the smell, taste and look of feces everywhere - in the body as well as outside - and all that while it's getting a bit hot in here.

and I doubt that the consequences of controlled demise in a deteriorating environment all while the meds and drugs of leadership and the people fade out quite a few of the brains and the bodies implicit reactions to a lot of sensory perceptions to everything that was vital, crucial to notice for a 'million' years can't be projected to at least some degree. I mean "blindspots" are a thinking tool, after all, but those thinking brains and minds believe in black swans and the better angels of our nature so that doesn't really mean a thing.

the population itself is fine, a habit of psycho-social education and all consecutive upper levels being insanely afraid of competition and insights from below. thing is, whatever financial survival schemes people are running, they all have death cult written all over their faces.

btw, most of this was for fun, I'm really not worried at all. climate change is more a cycle than man-made acceleration. my only point of interest is the deterioration of the species due to all the things that we do and then worry more about the habitat than our and all kinds.

we absolutely can turn the planet into a conservatory. through any climate.

Regarding the immediate effect I am sure your point is valid. But it’s also a bit of a cynical point of view, wouldn’t you say? People make these statements and pursue these personal lifestyle decisions because of their dreams for a better future - not its immediate effect. Just as companies need a vision to succeed, societies need vision as well. If a lot of people are vocal about something and live it, it has a chance of becoming anchored in laws and so force companies to do the “right thing”. Regulation follows collective values.

I think this is wrong.

Descriptively / "objectively" if you make your demand cleaner, you decrease demand for dirty consumption. You can't say individuals don't matter by comparing them to the world, that's invalid.

Normatively, is it a useful lie? Maybe, to some extent. People are lazy, selfish, and stupid. Peter Singer points out that we might be nice to people nearby, but we don't give money to people starving in other countries even if we think it will make a real difference. And no human can really know how even a pencil is made, so we make poor decisions. A carbon tax would unleash the free market on the problem. But saying individuals can't act is not good leadership, if even the people who say they want to fix the issue won't make personal sacrifices, why should the average voter?

"Personal impact" is just laundering the responsibility of government and corporations so it looks like it's our fault.

It is true that everyone everywhere all at once could suddenly make the right decision forever and save the planet. But is a statistical anomaly so extreme it's not worth pursuing as a policy. No policy maker worth their salt would look at that and consider it valid long term.

We have a playbook. We refuse to use it. We ban products, and then the companies that refuse to change or cheat get shuttered, and we move on.

Isn’t it also the case that a lot of those plastics come from those rivers because the first world ships is plastic water to China to be processed (“recycled”) and that this is what skews the numbers like this?

Even assuming everything you state to be true, moving sales to a less developed area also means selling for lower price in most cases. That is an incentive at least to try and comply with regulation. Only that the US is notorious underregulated.

The absolute best thing I've read on this subject is this article here: https://about.bnef.com/blog/liebreich-generative-ai-the-powe...

It talks at great length about data center trends relating to generative AI, from the perspective of someone who has been deeply involved in researching power usage and sustainability for two decades.

I made my own notes on that piece here (for if you don't have a half hour to spend reading the original): https://simonwillison.net/2025/Jan/12/generative-ai-the-powe...

I find the following to be a great point regarding what we ought to consider when adapting our lifestyle to reduce negative environmental impact:

> In deciding what to cut, we need to factor in both how much an activity is emitting and how useful and beneficial the activity is to our lives.

The further example with a hospital emitting more than a cruise ship is a good illustration of the issue.

Continuing this line of thought, when thinking about your use of an LLM like ChatGPT, you ought to weigh not merely its emissions and water usage, but also the larger picture as to how it benefits the human society.

For example: Was this tech built with ethically sound methods[0]? What are its the foreseeable long-term effects on human flourishing? Does it cause a detriment to livelihoods of the many people while increasing the wealth gap with the tech elites? Does it negatively impact open information sharing (willingness to run self-hosted original content websites or communities open to public, or even the feasibility of doing so[1][2]), motivation and capability to learn, creativity? And so forth.

[0] I’m not going to debate utilitarianism vs. deontology here, will just say that “the ends justify the means” does not strike me as a great principle to live by.

[1] https://news.ycombinator.com/item?id=42486481

[2] https://news.ycombinator.com/item?id=42549624

Hello Simon,

You mention that

> Google, Microsoft, Meta and Amazon all have net-zero emission targets which they take very seriously, making them "some of the most significant corporate purchasers of renewable energy in the world". This helps explain why they're taking very real interest in nuclear power.

Nuclear is indeed (more or less) zero-emission, but it's not renewable.

Thank you for the synthesis and link to the original article, it's a good read!

Such a stupid post, I know people on HN don’t like absolute descriptors like that and sorry for that.

Obviously the LLMs and ChatGPT don’t use the most energy when answering your question, they churn through insane amounts of water and energy when training them, so much so that big tech companies do not disclose and try to obscure those amounts as much as possible.

You aren’t destroying the environment by using it RIGHT NOW, but you are telling the corresponding company that owns the LLM you use “there is interest in this product”, en masse. With these interest indicators they will plan for the future and plan for even more environmental destruction.

Similar feelings about the repeated references to the apparently agreed consensus that individual action is pointless vs systematic change like switching to a renewable energy system. Jevons Paradox would like a word.

Plenty of companies have revealed exactly how much energy and CO2 they have used training a model. Just off the top of my head, I've seen those numbers are available for Meta's Llama models, Microsoft's Phi series and DeepSeek's models - including their impressive DeepSeek v3 which trained for less than $6m in cost - a huge reduction compared to other similar models, and a useful illustration of how much more effect this stuff can get on the training side of things.

It's not like they are mixing that water with oil and pumping into the aquifer. Water evaporate, turn into clouds, that precipitate into rain that fall on the ground and water bodies, where it can be used again. So what's the problem, with datacenter water usage? Has the water cycle has stopped and I was not informed?

Fresh water is finite. Infinite in reuse, but we can only take so much from a river before that river ceases to be. If you have a megabit connection, it doesn't matter that your cloud backups have infinite storage, you are limited by bandwidth.

Water vapor stays aloft for wild, so there's no guarantee it enters the same watershed it was drawn from.

It's also a powerful greenhouse gas, so even though it's removed quickly, raising the rate we produce it results in more insulation.

It's not a finite resource, we need to be judicious and wise in how we allocate it.

Anyone care to have a go at back of the envelope number for training energy use amortized per query for ChatGPT's models? Is the training or the inference going to dominate?

Hey all I wrote this post. To clear up a few points:

I meant this post to tell individuals that worrying about the emissions they personally cause using ChatGPT is silly, not that AI more broadly isn't using a lot of energy.

I can't really factor in how demand for ChatGPT is affecting the future of AI. If you don't want to use ChatGPT because you're worried about creating more demand, that's more legit, but worry about the emissions associated with individual searches right now on their own is a silly distraction.

One criticism is that I didn't talk about training enough. I included a section on training in the emissions and water sections, but if there's more you think I should address or change I'm all ears. Please either share them in the comments on the post or here.

I saw someone assumed I'm an e/acc. I'm very much not and am pretty worried about risks from advanced AI. Had hoped the link to an 80,000 Hours article might've been a clue there.

Someone else assumed I work for Microsoft. I actually exclusively use Claude but wanted to write this for a general audience and way fewer people know about Claude. I used ChatGPT for some research here that I could link people to just to show what it can do.

The title does not match the content.

A more appropriate title is "Emissions caused by chatgpt use are not significant in comparison to everything else."

But, given that title, it becomes somewhat obvious that the article itself doesn't need to exist.

> given that title, it becomes somewhat obvious that the article itself doesn't need to exist.

Why? I regularly hear people trying to argue that LLMs are an environmental distaster.

Because LLMs are an environmental disaster.

It's not about any individual usage. It's the global technology that is yet to prove to be useful and that already have bad for the environment.

Any new usage should be free of impact on the environment.

(Note: The technology of LLM itself is not an environmental disaster, but how it is put in use currently isn't the way).

> It's the global technology that is yet to prove to be useful

Useful for whom, by what definition? I personally find it very useful for my day to day work, whether it be helping me write code, think through ideas, or otherwise.

I don't want to be rude but I don't think that your personal experience is of any interest. The only question is "are LLM relevant ?"

And the only way to assert that they are is to get numbers (big or small) and to compare them to alternatives.

> I don't think that your personal experience is of any interest

But you didn't specify what you meant by "useful," hence why I asked the question I did. So under such ambiguity, my assertion is absolutely a counterpoint to what you just said. I will ask again, useful for whom, by what definition?

Oh! Good point, we may not have the same definition of usefulness.

By useful I mean, to the world. That it affect the world in a good way. Maybe it's not the best technology to do something but replace the best way in an CO2 effective way. Maybe it is not a clean technology but increase overall fairness.

I don't know, but it had to have a good income to the world in a way.

> yet to prove to be useful

I don’t understand this perspective. It should be abundantly clear at this point that these systems are quite useful for a variety of applications.

Do they have problems? Sure. Do the AI boosters who breathlessly claim that the models are super intelligent make me cringe? Sure.

But saying that they’re not useful is just downright crazy.

> I don’t understand this perspective. It should be abundantly clear at this point that these systems are quite useful for a variety of applications.

LLM are polyvalent. But in most of the tasks they are not the most efficient way to do the task.

Want to play chess ? Use Stockfish or Leela. Want to do image recognition ? SAM or TinyViT like models. Want to know if your are sick ? Go to the doctor or at least do a search on the web.

Yes, there is tasks where LLM are perfect for (speech analysis/classification for example). But omnipotent chatbot isn't one for example.

If there were a revolutionary use, we would have a productivity boom. We don't. This article is from 2021: https://www.technologyreview.com/2021/06/10/1026008/the-comi...

> If there were a revolutionary use, we would have a productivity boom. We don't.

What evidence do you have for this assertion? It seems like you are asserting something as fact when in reality it's your own personal opinion, yet ironically you are dismissing everyone else's personal experiences as mere opinion too.

> What evidence do you have for this assertion?

No predicted productivity boom (check last US data), no GDP boost yet (again last data). Even LLM enthusiast like McKinsey or Goldmansachs expect nothing before 2027.

And it's not about LLM, it's about the whole AI progress. That is, obviously, a revolution.

https://www.goldmansachs.com/insights/articles/ai-may-start-...

But just to be clear. I'm denying something said to be obvious. I should not be the one who give sources about something that doesn't exist. If there is a productivity boom, I may not have seen it. Show it to me.

The article needs to exist because the idea that ChatGPT usage is environmentally disastrous really has started to make its way into the human hive mind.

I'm glad someone is trying to push back against that - I see it every day.

> "Emissions caused by chatgpt use are not significant in comparison to everything else."

Emissions directly caused by Average Joe using ChatGPT is not significant compared to everything else. 50,000 questions is a lot for an individual using ChatGPT casually, but nothing for the businesses using ChatGPT to crunch data. 50,000 "questions" will be lucky to get you through the hour.

Those businesses aren't crunching data just for the sake of it. They are doing so ultimately because that very same aforementioned Average Joe is going to buy something that was produced out of that data crunching. It is the indirect use that raises the "ChatGPT is bad for the environment" alarm. At very least, we at least don't have a good handle on what the actual scale is. How many indirect "questions" am I asking ChatGPT daily?

Learning a new model (like GPT-4) is way more costly than running it.

Where in the world are you getting the numbers for how much video streaming uses energy? I am quite sure that just as with LLMs, most of the energy goes into the initial encoding of the video, and nowadays any rational service encodes videos to several bitrates to avoid JIT transcoding.

Networking can’t take that much energy, unless perhaps we are talking about purely wireless networking with cell towers?

LLM Inference is still quite power-hungry, Video decoding with hardware acceleration is much more efficient.

But we can do some estimates, heck, we can even ask GPT for some numbers.

Say you want to do 30 minutes of video (h265) or 30 minutes of LLM inferencing on a generic consumer device, ignoring the source of the model or source of encoded video, you get about 4x difference:

  Energy usage for 30 minutes of H.265 decoding: ~15–20 Wh.
  Energy usage for 30 minutes of Llama3 inference: ~40–60 Wh.

This is optimised already, so a working hardware H.265 decoder is assumed, and for inferencing, something on the level of an RTX 3050, but can also be a TPU or NE.

While not the most scientific comparison, it's perhaps good to know that video decoding is practically always local, and for streaming services it will use whatever is available and might even switch codecs (i.e. AV1, H.265, H.264 depending on what is available, and what licenses are used). And if you have older hardware, some codecs won't even exist in hardware, to the point where you start doing software decoding (very inefficient).

AI inferencing is mostly remote (at least the heavy loads) in a datacenter because local availability of hardware is pretty hit and miss, models are pretty big and spinning one up every time you just wanted to ask something is not very user friendly. Because in a datacenter you tend to pay for amperage per rack, you spec your AI inferencing hardware to eat that power since you're not saving any money or hardware life when you don't use it. That means that efficiency is important (more use out of a rack) but scaling/idling isn't really that big of a deal (but it has slowly dawned on people that burning power 'because you can' is not really a great model). That AI inferencing in a datacenter is more power-hungry as a result, because they can, because it is faster, and that's what attracts users.

I would estimate that the local llama3 inferencing uses less power than when done in a datacenter, because there simply is less power available locally (try finding an end-user device that is used mass-market with enough power available, you won't; only small markets like gaming PCs and workstations will do).

20 Wh for 30 minutes of hardware accelerated h265 decoding is an order of magnitude too high at any bitrate. Please cite your sources.

As I wrote in my reply, I don't have "sources".

Pure decode excluding any other requirements is probably pretty low, but running a decoder isn't all you need. There's network, display, storage and RAM so your OS can run etc. There will probably be plenty of variation (brightness, environment, how you get your stream in since a 5G modem is probably going to be different energy-wise compared to WiFi or Ethernet), and if you have something like a decoder in the CPU or in the GPU and if that GPU is separate, more PCIe involvement etc. But we can still estimate:

Hardware decoding (1080p video): ~5–15 W for the CPU/GPU

Overall system power usage (screen, memory, etc.): ~25–45 W for a typical laptop.

Duration (30 minutes): If we assume an average of 35 W total system power, the energy consumption is:

Energy = 35W × 0.5h ours = 17.5 Wh

We can do a similar one for inference, also recognising you'll have variations either way:

CPU inference: ~50 W. GPU inference: ~80 W. Overall system power usage: ~70–120 W for a typical laptop during LLM inference.

Duration (30 minutes): Assuming an average of 100 W total system power:

Energy = 100W × 0.5 hours = 50Wh

We could pretend that our own laptop is very good at some of these tasks, but we're not taking about the best possible outcome, we're talking about the fact that there is a difference between decoding a video stream and doing LLM inference, and the fact that that difference is big enough to make someone's point that video streaming is somehow 'worse' or 'as bad as' LLM usage moot. Because it's not. LLM training and LLM inference eats way more energy.

Edit: looking at some random search engine results, you get a bunch of reddit posts with screenshots from people asking where the power consumption goes on their locally running LLM inferencing: https://www.reddit.com/r/LocalLLaMA/comments/17vr3uu/what_ex...

It seems their local usage hovers around 100W. Other similar posts hover around the same, but it seems to be throttle based as other machines with faster chips also throttle around the same power target while delivering better performance. Most local models use a quantised model which is less resource-hungry, the cloud-hosted models tend to use much larger (and thus more hungry models).

Edit2: looking at some real-world optimised decoding measurements, it appears you can decode VP9 and H.265 on 1 year old hardware below 200mW. So not even 1W. That would mean LLM inferencing is orders of magnitude more power hungry than video decoding. Either way: LLM power usage > Video Decode power usage, so the article trying to put them in the same boat is nonsense.

"I would estimate that the local llama3 inferencing uses less power than when done in a datacenter, because there simply is less power available locally"

Is this taking into account the fact that datacenter resources are shared?

Llama 3 on my laptop may use less power, but it's serving just me.

Llama 3 in a datacenter or more expensive, more power-hungry hardware is potentially serving hundreds or thousands of users.

Luckily we don’t have to do such a calculation. All this energy use will be factored into cost which tells us which is using more resources.

Ah yes high tech, an industry where there's famously no weird distorting influence from VCs subsidizing unprofitable business models to grab market share.

It doesn’t matter that you are paying for, someone’s paying for it, and that economizing force always is putting pressure on.

If you use chatgpt somehow saves you from making one trip to the doctor in your car it can offset the entire year worth of chatgpt usage in terms of co2 impact.

if your use of chatgpt saves you from a trip to the doctor I would be very concerned

Most doctor visits are for benign matters...

The point of the doctorate is for them to make that determination.

More constructively, the middle ground of asking a doctor about some minor symptoms over async messaging or tele-appointments (whatever the term for those is), is probably always something to consider for the trivial stuff.

My partner works in emergency. Someone came in with a blister on their finger yesterday. They had to see the patient and send them home with a bandaid. Not a good use of resources.

There is no need to nitpick obvious comments. If you have a running nose and fever in winter, it's 99.99% a cold and a doctor visit will do nothing to help you, yet millions of people waste doctors time for that

Yes and no, you really don't need a deep expertise to diagnose the very common illness.

And that's where chatgpt is doing great.

Early days, but not as crazy as it sounds: https://jamanetwork.com/journals/jamanetworkopen/fullarticle...

"The LLM alone scored 16 percentage points (95% CI, 2-30 percentage points; P = .03) higher than the conventional resources group."

Would be much more interesting if this ranked based on severity of misdiagnosis. An LLM that is 50% better at diagnosing a common cold but missed sepsis 10% more often would not be an overall improvement.

ChatGPT is probably adequate to provide a slightly more user-friendly but also slight-less-reliable replacement for a reliable consumer-oriented medical reference book or website for the task of determining whether self-care without seeing a doctor or seeing a doctor is appropriate for symptoms not obviously posing an immediate emergency.

If ChatGPT somehow makes you eat more burgers it could make it makes water consumption worse.

To me the "ChatGPT is destroying the environment "card always felt somewhat like bad faith arguing from the anti-AI crowd trying to find any excuse for being against AI. Like, the same people who complained about "using AI is destroying environment" seemed to have no issue with boarding a plane which would emit a bunch of CO2 so that they can have a vacation in Europe or the like. Selective environmentalism.

It's less bad faith, more a meme that has become so prevalent that it's impossible to dispell as it's something too nuanced for social media. I've seen more than a few social media posts asking "do they really cut down a rainforest every time someone generates an AI image?"

I mean we can go back and forth all day with credulous clowns on both sides. There are certainly plenty of misguided AI advocates who think what they're using is magic.

I mean, it's a literal net new expenditure of power and water. I also deeply doubt they have "no issue" with plane travel. You're just assuming the worst and most hypocritical position to someone, which seems deeply bad faith as well.

It's literally true that most of the AI vendors and their data center partners are writing off energy and water conservation targets they'd had for the near future because of LLM money. That is actually bad in the short and likely long term, especially as people use LLMs for increasingly frivolous things. Is it really necessary to have an LLM essentially do a "I'm Feeling Lucky" Google Search for you at some multiple of that environmental cost? Because that's what most of my friends and coworkers use ChatGPT for. Very rarely are they using it to get anything more complex than just searching wikipedia or documentation you could have had bookmarked.

A person has a choice of if they take a flight and if it's worth it for them. They have no power except for raising a complaint in public on if OpenAI or Google or whoever spends vast amounts of money and power to train some new model. If your bar is that no one is allowed to complain about a company burning energy unless they live a totally blameless life farming their own food without electricity then random companies will get to do any destructive act they want.

Who is this person you’re constructing? Being concerned about plane emissions and travel is an incredibly common thing and people are adjusting their lifestyles accordingly - lots of overnight sleeper train lines are reopening due to demand.

You are talking about hypocrites.

What about the people who take the protection of the environment seriously?

They got now a setback because not only didn’t we reach our previous goals on lowering energy consumption but know we put new consumption on top of that. Just because the existing one ate worse doesn’t make it good.

There is a reason why MS missed its CO2 targets and why everyone is kn search for more energy sources.

They all create more CO2.

Sort of off-topic, but it does make one think about usage of compute (and the backing energy / resources required for that)...

i.e. it doesn't seem too much of an exaggeration to say that we might be getting closer and closer to a situation where LLMs (or any other ML inference) is being run so much for so many different reasons / requests, that the usage does become significant in the future.

Similarly, going into detail on what the compute is being used for: i.e. no doubt there are situations currently going on where Person A uses a LLM to expand something like "make a long detailed report about our sales figures", which produces a 20 page report and delivers it to Person B. Person B then says "I haven't time to read all this, LLM please summarise it for me".

So you'd basically have LLM inference compute being used as a very inefficient method of data/request transfer, with the sender expanding a short amount of information to deliver to the recipient, and then the said recipient using an LLM on the other side to reduce it back again to something more manage-able.

That sounds like the opposite of data compression (inflation?) where the data size is increased before sending, then on receiving it is compressed back to a smaller form.

Lossy data inflation.

Data center emissions probably 662% higher than big tech claims. Can it keep up the ruse?: https://www.theguardian.com/technology/2024/sep/15/data-cent...

what does "water used by data center" even mean? Does it consume the water somehow? What does it turn into? Steam? So uploading a 1GB file boils away nearly 1 liter of water? Or is it turned into bits somehow in some kind of mass to energy conversion? I sorta doubt that. Also this means data centers would have cooling towers like some power stations. Are we talking about the cooling towers of power stations?

I think at least that graph is complete non-sense. I will try and have chatGPT explain it to me.

Yes, datacenters have cooling towers. There are lots of good articles about this topic. A good starting point is "water usage effectiveness" (WUE) which is one way this is tracked.

The one near here just has heat exchanges. But even if all the others use evaporators then potential water usage is extremely misleading because its not like the water is consumed - its just temporarily unavailable.

Also why doesn't uploading a 1GB file to my NAS boil a liter of water? are maybe all the switches and routers used between me and the datacenter water-cooled? I mean I can see such switches existing but I don't see them be the norm. Why doesn't the DSLAM on the Street outside emit steam. Is there maybe one bad switch somewhere that just spews steam?

What I am saying is that at least that graph is without further explanation... bad.

> The one near here just has heat exchanges. But even if all the others use evaporators then potential water usage is extremely misleading because its not like the water is consumed

Water consumption in all contexts is mostly fresh water returned from immediately usable form to either evaporation or the ocean. It is not "extremely misleading", because when it returns to immediately usable form by, e.g., precipitation, that's when new water is considered to be made available. The normal definitions are internally consistent and useful.

I did a bit of research and even water just "involved" in the process is counted as used in this context. For instance river water that is used for cooling and returned is counted.

I think these sort of graphs are simply misleading and should not be used.

Is that actually a common configuration? I could find very little on data centers being cooler by river water, and the few I found sounded like they were doing something novel.

Using natural water is more common in power plants, where the volume of water is sufficiently large to use concrete aquaducts and the precision of the temperature isn't as important. Most datacenter liquid cooling is treated municipal water with glycol additives to prevent corrosion and pump jamming.

> what does "water used by data center" even mean?

This doesn't clarify what exactly it includes, but there are two main things that generally are included:

(1) Direct water use for cooling, (which, yes, ends up as steam rom cooling towers), and

(2) Water used in generating electricity consumed by data centers, which, yeah, is again evaporated in cooling towers.

> what does "water used by data center" even mean?

It’s referring to water lost to evaporation in evaporative cooling towers, both at the data center and at the power generating plant.

I wrote the article. Water use is used really ambiguously by people talking about data centers and LLMs in general. To get the numbers I just took the estimated water used per kWh (https://www.sunbirddcim.com/glossary/water-usage-effectivene....) and multiplied it by the number of kWh used in data centers for different tasks

"personal carbon footprint" is a term invented by BP and is the single hack that derailed the environment discussion by making people personally responsible and removing the actual polluters from the discussion.

The article could indeed be written by the same kind of people, given it glances over training cost as if AI companies aren't pushing datacenter power/gpu capacities to the limit to produce incrementally better models by brute force. It all falls apart as soon as you stop individualizing the numbers and add back in all of the supposedly non-recurring (but actually recurring because we can't stop ourselves from redoing it) cost.

I specifically mentioned in the article that I think it's mostly a waste of time to worry about your personal carbon footprint and you should aim for big systematic change instead, but because so many people seem to be worried about their personal chatgpt use I wanted to clarify that it doesn't use much energy anyway

You are kidding right?

OpenAI -- for now -- is planning to build 5 gigawatt data centers (yes, plural) to continue its quest towards AGI. -- see HN archives. Meanwhile they are also looking into private nuclear power for the same purpose.

Any serious competitor will likely need to do the same.

So there's a rational choice and tradeoff to make:

Net zero or AGI.

We can't have both.

> and it’s completely clear to me that one side is getting it entirely wrong and spreading misleading ideas

What a great way to start an article. I get it as: "I am not open to listening to your arguments, and in fact if you disagree with me, I will assume that you are a moron".

It reminds me of people saying "planes are not the problem: actually if you compare it to driving a car, it uses less energy per person and per km". Except that as soon as you take a passenger in your car, the car is better (why did you assume that the plane was full and the car almost empty?). And that you don't remotely drive as far with your car as you fly with a plane. Obviously planes are worse than cars. If you need to imagine people commuting by car to the other side of the continent to prove your point, maybe it's not valid?

The fact is that the footprint of IT is increasing every year. And quite obviously, LLMs use more energy than "traditional" searches. Any new technology that makes us use more energy is bad for environment.

Unless you don't understand how bad the situation is: we have largely missed the goal of keeping global warming to 1.5C (thinking that we could reach it is absurd at this point). To keep 2C, we need to reduce global emissions by 5% every year. That's a Covid crisis every year. Let's be honest, it probably won't happen. So we'll go higher than 2C, fine. At the other end of the spectrum, 4C means that a big stripe (where billions of people live) around the equator will become unlivable for human beings (similar to being on Mars: you need equipment just to survive outside). I guess I don't need to argue how bad that would be, and we are currently going there. ChatGPT is part of that effort, as a new technology that makes us increase our emissions instead of doing the opposite.

I take your general point, but:

> Except that it doesn't work if you don't drive your car alone (if you assume the plane is full of passengers, why not assuming that the car is, as well?)

These can be measured for averages. Lots of cars with one person in them, seldom cars fully packed; lots of planes fully packed, seldom (but it does happen) that the plane is almost empty.

> we have largely missed the goal of keeping global warming to 1.5C (thinking that we could reach it is absurd at this point).

Probably, yes; last year passed the threshold — it would be a pleasant *surprise* if that turned out to have been a fluke 14* years early.

* 14 because it would take 14 years for the exponential — seen for the last 30 years — for PV to replace all forms of power consumption; not just electricity, everything. But even then we'd also need to make rapid simultaneous progress with non-energy CO2 sources like cattle and concrete.

> around the equator will become unlivable for human beings (similar to being on Mars: you need equipment just to survive outside)

In so far as your bracket, sure; but there's a huge gap in what equipment you would need.

The comparison I often make is that Mars combines the moisture of the Sahara, the warmth of the Antarctic, the air pressure of the peak of Mount Everest, and the soil quality of a superfund cleanup site, before then revealing that it's actually worse on all counts.

> These can be measured for averages. Lots of cars with one person in them

Sure, but the point should be that we should strive to share cars, not that it's okay to take the plane! Especially given the second argument which is that you don't drive 1000km every time you take your car. The footprint per km is not enough: when you take the plane you typically go much further!

> Probably, yes; last year passed the threshold

That, plus the IPCC scenario that keeps us under 1.5C says that in a few decades, not only we won't be extracting any carbon anymore, but we will be pumping carbon underground faster than we are extracting it now! And that's with the IPCC models which tend to be optimistic (we measure that every year)!

> 14 because it would take 14 years for the exponential — seen for the last 30 years — for PV to replace all forms of power consumption

And you would have to take into account that PV today entirely relies on oil. We are going towards a world with less and less oil, and we don't know how it will impact our capacity of production for PVs. But probably it won't help.

> In so far as your bracket, sure; but there's a huge gap in what equipment you would need.

Sure. It was a quick way to say that the combination of humidity and temperature will be such that sweating won't help humans regulate their temperature. And when we can't regulate our temperature, we die. By any account, this means that billions of people will have to relocate, which means global wars (with entire countries moving with their entire armies).

Now of course that would be infinitely better than trying to live on Mars, which is why it is preposterous to even consider Mars.

> but we will be pumping carbon underground faster than we are extracting it now!

While I know about "we need to sequester carbon", I thought the assumption was more for the last 10% (which makes sense, last 10% of anything is often expensive), not >100% of current?

> And that's with the IPCC models which tend to be optimistic (we measure that every year)!

Indeed, unfortunately.

> entirely relies on oil

I don't believe "relies on" is correct: while I would agree that e.g. plastics are made from oil, that oil currently powers some of the energy generation capacity used for the manufacturing plants that make the panels, that shipping and air transport are at present almost entirely oil-based, these are not "entirely relies on oil", they are "the economy in which they emerged happens to have been built on oil".

This is importantly different, because as renewable energy ramps up, the CO2 emissions resulting from each of these steps also goes down — even for the plastic, as the carbon in the oil itself is much more valuable as plastic than as a fuel waste product.

> By any account, this means that billions of people will have to relocate, which means global wars (with entire countries moving with their entire armies).

Aye.

Lots of room for massive disasters there, even if it were not for the fact that at least one affected area already has nukes.

> not >100% of current?

I am not completely sure about the exact numbers, but my understanding is something like this: currently we extract 7 billion tons of oil (or is it fossil fuels in general?) per year, and the IPCC scenario for 1.5C says that in a few decades we will have to not only be zero emissions, but also sequester 10 billion tons per year. So yeah, that's another way to say "impossible".

> "the economy in which they emerged happens to have been built on oil"

Sure, but... it's not clear at all if globalization the way it is now is even possible without oil. We currently use oil for transports because it is much denser. We can't move a supertanker with PV, for instance. Extrapolating the evolution of renewables from the last decade is definitely optimistic because we will have (that's just natural limits) and must use (if we don't want to reach 4C) less fossil fuel, so it will most definitely become harder and harder.

I believe that we need as much renewables and nuclear as we can, because even that will not compensate for oil. So we will live in a world with less energy and a harder climate, that's a fact. The challenge now is to deal with it, and do as much as possible to keep as much energy as we can while preserving the climate as much as we can. This is the biggest challenge in the Human history, by far. And instead of focusing on that, we try to send a few people to Mars for no good reason...

> It is extremely bad to distract the climate movement with debates about inconsequential levels of emission

This. So we should focus on optimizing transport, heating, energy and food.

A very nice article. But the google search and LLM energy estimates are outdated. More recent work put both at 10x less.

https://engineeringprompts.substack.com/p/does-chatgpt-use-1...

Yup I decided to go with the worst estimates given by environmentalist critics of ChatGPT to try to show that even there it doesn't seem like a problem.

It's discomforting to me when people compare resource usage of ChatGPT, a computer, to the resource usage of a human being.

I've seen charts like this before that compare resource usage of people to corporations, implying corporations are the bigger problem. The implication here seems to be the opposite, and that tone feels just a little eugenicist.

Thinks like heating and car heavily depends on the usage, which I guess is based on the USA average.

US houses are HUGE and even here in Europe square m2 / person double in the last decades.

- we don’t have a housing problem, we have a surface inflation problem.

- heating is directly correlated to the volume to heat. Heating 100m2/person with (coal Chinese steel, resource extracted, logistics…) solar and batteries or heat pump isn’t necessarily more carbon or water efficient that 20m2/person with gas.

Bonus point: the resident will have to think twice before filling his property with garbage consumerism.

Ps: my GF and I live in 80m2 house, the precedent family where… 2 adults and 3 children! I thing the space is wayyy enough for us but people visiting regularly remark "it’s so tiny/small! "

Enforce a global carbon tax, price it in, and tax the land

International tax agreements are crucial if any progress is to be made on preventing climate disaster.

Which major parties support it? Who is even talking about it?

It’s such an obviously needed mechanism, but hard to get anyone enthused about it.

This group have some proposals on the topic —

https://en.m.wikipedia.org/wiki/Association_for_the_Taxation...

Oof. This article misses some important details.

Training is not a "one time cost". Training gets you a model that will likely need to be updated (or at least fine-tuned) on newer data. And before GPT4, there was a series of prior, less effective models (likely swept under the rug in press releases) made by the same folks that helped them step forward, but didn't achieve their end goals. And all of this to say nothing of the arms race by the major players all scrambling to outdo each other.

It also needs to compare this to the efficiency modern search engines run at. A single traditional query is far less expensive than a single LLM query.

The use of ChatGPT doesn’t replace the others it comes on top of that.

MS is missing its CO2 targets because of AI not because of burgers.

The whole argument is, it’s not bad because other things are worse.

We are racing towards the abyss but don’t worry AI only accelerates a little more.

This used to be the stick they used to beat bitcoin with. I guess it's a good stick because you can hit any technology with it and you can conveniently forget all the terrible uses to which electricity is put.

The part on training is misleading and full of shit.

Training is not a "one-time cost". There is an implied never-ending need for training. LLMs are useless (for one of their main purposes) if the models get stale.

I can use Musk's own argument on this one. Each model is a plane, fully built, that LLM researchers made into a disposable asset destined to be replaced by a newly built plane on the next training. Just incredibly stupid and inneficient.

I know what you're thinking right now: fine-tuning, etc. That is the "reusable" analogy to that, is it not? But fine-tuning is far, far from reusability (the major players don't even care about it that much). It's not even on the "hopper" stage.

_Stop training new shit, and the argument becomes valid. How about that?_

---

I am sure the more radical environmentalists know that LLMs can be eco-friendly. The point is: they don't believe it will go that way, so they fight it. I can't blame them, this has happened before.

_This monster was made by environment promises that were not met_. If they're not met again, the monster will grow and there's nothing anyone can do about it. I've been more moderate than this article in several occations and still got attacked by it. If not LLMs, it will target something else. Again, can't blame them.

> LLMs are useless (for one of their main purposes) if the models get stale.

I don't think that's entirely accurate. A lot of people deliberately continue to chose to use the older GPT-4 despite it not being updated since June 2023.

GPT-4o has had releases in May, August and November of 2024 - so about one every 3-4 months.

Anthropic's Claude 3.5 Sonnet was released in June and had a single update in October.

Personally I'd rather have a model with excellent summarization / tool using abilities that can look up recent facts about the world.

LLMs that can replace search (their primary and self-declared goal) cannot survive without repeated training. As I said, this is one of its main purposes.

The other main purpose (military application, surveillance, autonomous psyops) is also highly dependent on continous training. Without it, properly educated healthy humans can overcome its reasoning power very quickly.

All other user profiles are just cannon fodder. Companies don't give a fuck about people running older models. They'll do whatever they can to make you use a more recent one.

That's why I'm being provocative with the "let's stop training new shit" argument. I'm aiming for the heel.

If somebody told you that LLMs are a good replacement for search that person was misleading you.

LLM companies are doing that themselves with their actions. There is a clear connection between search and assistants, historically way back to "Ask Jeeves".

People literate in IT know the implementation difference. For those not literate, the difference is way less proeminent. To most people, it's the same thing and companies know it and abuse this.

They are obviously competing for the same market. That market being "the stuff you go to when you need knowledge".

Anyway, you are deviating from the point. Even if that's not the case, my argument still holds: the article is full of shit regarding the environmental sustainability of the lifecycle of model training.

If we're trying to figure out a reasonable number for how much energy a single ChatGPT search uses, it seems weird to factor in all future training of all future models. It would be like trying to figure out the carbon cost of a single plane ride and trying to figure out what fraction of additional plane flights that one ticket incentivizes. It's too murky to put a clear number on and probably doesn't add much to the cost on its own. I tried to make it clear that the post is about your own personal use of ChatGPT, not the entire AI industry as a whole.

I agree, it's an impossible task to measure it. Almost like a labyrinth you can get lost if you worry too much about it.

The environment is not a personal issue. You can't solve it just for you. The whole idea of making it personal is so that a collective aspect of it would flourish. What are you trying to flourish in people's minds?

I added in the article that I think this is all the wrong way to think about environmentalism anyway and that people should spend most of their time working on systematically changing the energy grid to renewables. But I keep meeting people who are worried about their personal emissions from ChatGPT, so I wanted to make it clear that if you are worried about that ChatGPT isn't really a drop in the bucket compared to other things you could change.

I am also wondering, why they don't do online learning with their LLMs. Seems silly not to. Maybe they are performing some irreversible post-training steps, that do not allow for picking up training again afterwards. In any case, being able to feed an in production model new training data would be one of the simplest ideas to come up with.

In my country there's a lot of institutional hype about green algorithms. I find the whole idea quite irrelevant (for the reasons explained in this post) but of course, it's a way to get funding for those of us who work in AI/NLP (we don't have much money for GPUs and can't do things like training big LLMs, so it's easy to pitch everything we do as "green", and then get funding because that's considered strategic and yadda yadda).

It's funny, but sad, how no one calls the billshit because we would be sabotaging ourselves.

A lot of conversations regarding the environment feel so frustrating because they are either qualitative or use aggregate high level data or are like we'll be dead in 50 years (lol my personal favorite)

Why not start capturing waste/energy data for all human made items like nutritional data on food? It won't add much overhead or stifle economies as people fear

That way when I log in to use any online service or when I buy/drive a car or when I buy an item I can see how much energy was consumed and how much waste I produced exactly

I just created an account and submitted a post but it does not appear on new section just to me, anyone knows why?

can anyone submit this:

TikTok Ban in USA and the Hypocrisy of the USA Regime https://justpaste.it/tiktok_ban

I wish charts like this would score the carbon impact of having children. It would be off the scale.

yea lets not do that. if we all decided not to have kids we would not have to worry at all about the fucking environment eh? :)

I’m not saying people shouldn’t have children. Just that if you do, you have to stop flying, go vegan, and sort your recycling properly for ever after!

thats funny :)

I check every point of your list however I have an European-size house so my biggest consumption is probably heating, after schooling (done) and medical (most still to come).

Also beware telling people "not to have children". I know population is the biggest treat because it's a multiplicative factor on hour lifestyle, that we don't like to downgrade. However saying "not having children" is easily arguable as bdangubic did and he's still a bit right: we don't want everybody stop having children altogether. However talking about population have the power to seed ideas in others head, and let themselves make the relation with the number of children they'll have.

You convince people by making them convince themselves :)

The major players in AI are collectively burning 1-2 gigawatts, day and night, on research and development of the next generation of LLMs. This is as much as my city of a million people. The impact is real, and focusing on inference cost per query kind of misses the point. Every person who uses these tools contributes to the demand and bears some of the responsibility. Similar to how I have responsibility for the carbon emissions of a flight, even if the plane would have flown without me.

I'm saying this as someone who finds LLMs helpful, and uses them without feeling particularly guilty about it. But we should be honest about the costs.

Agreed, I feel like the main response seems to be "Does Not!", but it's reasonable to accept that a thing you like has a cost. We all emit carbon every day to do things we don't 100% need, and we should just be willing to admit there's a cost and try and move towards paying them.

Personally, I'm not tripping too hard about datacenter energy long term because it's very easy to make carbon free (unlike say ICE cars or aircraft). But it would be nice to see some efforts to incentivize green energy for those datacenters instead of just saying "whatever" and powering them with coal.

"But it would be nice to see some efforts to incentivize green energy for those datacenters instead of just saying "whatever" and powering them with coal."

The major datacenter operators all have very aggressive targets for clean energy which they have been mostly sticking to.

Google: https://sustainability.google/progress/energy/

AWS: https://sustainability.aboutamazon.com/products-services/aws...

Azure: https://azure.microsoft.com/en-us/explore/global-infrastruct...

Anyone who says that LLMs are terrible for the environment will never be swayed from that belief.

It is like a new shibboleth for idiocy.

When someone says it just reply with, “I see” and move on with your life.

"Other bad things exist" does not mean this thing isn't bad. Or absolutely could be that all these other things are huge energy wasters AND chat gpt is an energy waster.

We have to stop thinking about problems so linearly -- it's not "solve only the worst one first", because we'll forever find reasons to not try and solve that one, and we'll throw up our hands.

Like, we're well aware animal agriculture is a huge environmental impact. But getting everyone to go vegetarian before we start thinking about any other emissions source is a recipe for inaction. We're going to have to make progress, little by little, on all of these things.

Except that's not the point of this: the point is that humans have absolutely finite time to think about problems, and so the way you distract from a problem is by inventing a new more exciting one.

LLMs are in the news cycle, so sending all the activists after LLMs sure does a good job ensuring they're not going after anything which would be more effective doesn't it? (setting aside my thoughts for the moment of the utility of the 'direct action' type activists who I think have been useless for a good long while now - there could not possibly be more 'awareness' of climate change).

Then again, if you can stall a nascent polluter before it becomes entrenched, maybe that's the right time to intervene. Getting people to not eat meat is hard, we've been eating meat forever. Getting people to not use LLMs? That's where most of us were up until very recently.

"Don't use LLMs" is just another variant of "don't use electricity".

Reframe the problem like that and then realize that no one's going to do it: global electricity use is constantly increasing. Fortunately, global renewable energy use is also growing incredibly rapidly.

Which problem seems more tractable? Because reality has already proven it: people will happily switch to clean electricity and keep using electricity. They won't voluntarily use less electricity unless they get some benefit from that - i.e. reduced expenditure, or just plain more stuff (i.e. my LED lights consume a fraction of the power of my previous halogens, but are brighter and I have more of them and also can change light color on a schedule).

Maybe because we didn’t make real progress in the existing pollution and have a greater chance stopping a new polluter.

How long are climate change and its reasons known?

In the end people vote climate change deniers because they don’t like the inconvenient truth

Comparing magnitudes does not prove the headline, poor analysis.

Let's compare everything presented there with just one day of Russian war in Ukraine and we can forget about all CO hesitations in our normal lives.

Perhaps off topic, what exactly does the “one way european flight” mean in the context of avoiding co2 emissions? I.e. what is the choice or scenario here?

I am pretty sure that everybody else has pointed out but the training costs should also be accounted I suppose?

Why did this get flagged?

I emailed hn@ to try to find out, I’ve no idea why.

Using it, sure, now do operating it and building it.

Just build nuclear.

That doesn't solve the problem

Indeed, it doesn't solve the problem that people will misinterpret data and spread misinformation to justify their bad feeling about AI with invalid arguments.

I mean.... doing most things are bad for the environment. Do less of everything that you don't need to.

Number are wrong. Just trying on topic. Data centers and AI realty consumes a lot more resources that you article poses.

I published this as a comment as well, but it's probably worth nothing that the ChatGPT water/power numbers cited (the one that is most widely cited in these discussions) comes from an April 2023 paper (Li et al, arXiv:2304.03271) that estimates water/power usage based off of GPT-3 (175B dense model) numbers published from OpenAI's original GPT-3 2021 paper. From Section 3.3.2 Inference:

> As a representative usage scenario for an LLM, we consider a conversation task, which typically includes a CPU-intensive prompt phase that processes the user’s input (a.k.a., prompt) and a memory-intensive token phase that produces outputs [37]. More specifically, we consider a medium-sized request, each with approximately ≤800 words of input and 150 – 300 words of output [37]. The official estimate shows that GPT-3 consumes an order of 0.4 kWh electricity to generate 100 pages of content (e.g., roughly 0.004 kWh per page) [18]. Thus, we consider 0.004 kWh as the per-request server energy consumption for our conversation task. The PUE, WUE, and EWIF are the same as those used for estimating the training water consumption.

There is a slightly newer paper (Oct 2023) that directly measured power usage on a Llama 65B (on V100/A100 hardware) that showed a 14X better efficiency. [2] Ethan Mollick linked to it recently and got me curious since I've recently been running my own inference (performance) testing and it'd be easy enough to just calculate power usage. My results [3] on the latest stable vLLM from last week on a standard H100 node w/ Llama 3.3 70B FP8 was almost a 10X better token/joule than the 2023 V100/A100 testing, which seems about right to me. This is without fancy look-ahead, speculative decode, prefix caching taken into account, just raw token generation. This is 120X more efficient than the commonly cited "ChatGPT" numbers and 250X more efficient than the Llama-3-70B numbers cited in the latest version (v4, 2025-01-15) of that same paper.

For those interested in a full analysis/table with all the citations (including my full testing results) see this o1 chat that calculated the relative efficiency differences and made a nice results table for me: https://chatgpt.com/share/678b55bb-336c-8012-97cc-b94f70919d...

(It's worth point out that that used 45s of TTC, which is a point that is not lost on me!)

[1] https://arxiv.org/abs/2304.03271

[2] https://arxiv.org/abs/2310.03003

[3] https://gist.github.com/lhl/bf81a9c7dfc4244c974335e1605dcf22

In the same way that leaving food out doesn’t create a cockroach infestation.

What about training it?

Yeah, there are lots of things worse than LLMs. But, at least, they are useful and profitable.

This guy must work at Microsoft.

Reply on news.ycombinator.com