> Research suggests, for instance, that about 700,000 litres of water could have been used to cool the machines that trained ChatGPT-3 at Microsoft’s data facilities.
> Suggests
So it’s a complete guess, cool.
> Trained
I… don’t care? That’s a one-time cost, talk about how much inference costs and also make me understand why that matters. The water wasn’t destroyed in the process.
I’m having a hard time taking this article seriously, so data centers had more emissions than commercial flights (I’d like to see the numbers instead of this person’s word given other questionable things they say, like above), data centers provide a lot more value IMHO so that’s not surprising to me.
The answer to high emissions from energy production used to run something is not to shut down the thing using power (if it’s useful, crypto is a different beast), it’s to change the source of the energy. What would this author have us do? Shut down data centers than run the internet because they use power that comes from less desirable sources? Absurd.
We're shaming most of the population for way less. Somehow when it comes to corporations, it's all good when they maximise shareholder pro... wait, when they advance technology for all mankind.
For its energy usage claim, this article links to another article[0] which gives "about 500 metric tonnes of CO2" as the figure for the training of "ChatGPT-3". Unless I'm making a mistake, that's about 6% of Taylor Swift's annual private jet usage[1].
ChatGPT has 180 million users, and the underlying model is used through API access for a huge range of genuinely useful tasks. I think we absolutely should give more leeway to energy uses that advance technology and benefit many (ideally by being open-source) than to those that just serve the exorbitance of a few. Obviously we should still aim to reduce both.
Yeah, the article is just seemingly throwing out some random numbers without context or what the exact implication of those numbers would mean. But it's labelled as "ugly".
I have never been able to understand the argument about the supposed high water use - the water doesn't magically cease to exist after it's been used to cool a datacenter. You put freshwater in and get the same, but warmer, freshwater out. Probably doesn't require much (if any) in the way of treatment to become potable again.
Am I missing something or is it a bit of a disingenuous argument?
I'm not actually familiar with current DC cooling equipment...but I suspect they use a lot forced-draft, evaporative cooling towers ( https://en.wikipedia.org/wiki/Cooling_tower ). That is far more efficient than "dry" cooling (so long as the outdoor humidity isn't too high). But the waste water from it (what hasn't evaporated) has too high a mineral content to be re-used. Unless you ran it through a desalination plant, which would get very expensive.
To my understanding it's common to use evaporative cooling or discharge into a natural body of water, which would require full treatment again.
Though ultimately if a data center is carbon negative and water positive (which to be clear is not generally the case yet, but there is progress) I think raw energy/water usage numbers are less relevant.
I think in a lot of cases, there are already solutions for recovering at least the heat from the water as energy. I'm not sure about the water itself, but I think OP's point is that there's not really such a thing as 'waste water'—it'll just go back into the normal water cycle, either being pumped out into a nearby river, or evaporated up into the air.
I think the key concern is the wasted energy, as a lot of energy is used to clean the water prior to it being used in the data centre.
There is another problem with evaporative cooling in regards to the effect on climate: water vapor has an incredibly strong greenhouse effect. It's complicated though, because white clouds have the opposite effect: they reflect heat back into space. I have no idea how you could begin to calculate the net effect from additional humidity introduced into the environment, you would have to somehow figure out how much becomes just humidity and how much contributes to more cloud formation.
There might also an effect on fish spawning if the temperature is higher, some fish may not return to spawn until the water is cool enough or they have lower chances of survival if they hatch early due to increased temperatures. That is, assuming the remaining water flow allows for passage of fish runs.
> The ugly truth behind ChatGPT: AI is guzzling resources at planet-eating rates
This is nothing special.
Some months ago was considered normal that Bitcoin farms eat the energy of one town.
And it is still normal for Electron programms to eat a lot of resources. Because they can.
There is a lot of people that claim to care about climate, gender equality, human rights, working conditions etc... and then they keep using this shiny useless thing that goes in the opposite direction...
Claiming to care about X, when that is a currently-trending performative virtue, is very similar to claiming that you love dressing in a certain sort of clothing, which just happens to be the latest fashion.
Except clothes cost money, and claiming to care does not.
hbomberguy’s “Woke Brands” video is pretty good on how corporations use these things that people care about and disagree about to sell more product and obscure their true behaviour.
No one individual can bear the burden of trying to address all social issues. I think it's good to live according to one's political and moral beliefs, but ultimately there is a point to where this is reasonable to expect of someone.
That is to say that I think one can criticize the tech industry and its e-waste production, while also using a modern smartphone, for example.
> Suggests
So it’s a complete guess, cool.
> Trained
I… don’t care? That’s a one-time cost, talk about how much inference costs and also make me understand why that matters. The water wasn’t destroyed in the process.
I’m having a hard time taking this article seriously, so data centers had more emissions than commercial flights (I’d like to see the numbers instead of this person’s word given other questionable things they say, like above), data centers provide a lot more value IMHO so that’s not surprising to me.
The answer to high emissions from energy production used to run something is not to shut down the thing using power (if it’s useful, crypto is a different beast), it’s to change the source of the energy. What would this author have us do? Shut down data centers than run the internet because they use power that comes from less desirable sources? Absurd.