Is anyone surprised? "researchers estimates that the training of GPT-3, which ChatGPT is partly based on, consumed 1,287 MWh, and led to emissions of more than 550 tons of carbon dioxide equivalent" https://www.wired.com/story/the-generative-ai-search-race-has-a-dirty-secret/ #climatechange #AI
The Generative AI Race Has a Dirty Secret

Integrating large language models into search engines could mean a fivefold increase in computing power and huge carbon emissions.

WIRED

@peterroopnarine

So as much as 39 Americans emit per year.

@buermann @peterroopnarine so we could say it's a 39yo American (when looking at how much energy it consumed)
Issue with that kinda is it consumed it over a very short space of time instead of 39 years

@peterroopnarine

#AI may be alluring or frightening, depending on your viewpoint.

The implications for #ClimateChange will be the last thing on the developer's or the users' minds.

Hasn't this been the case with new technology since the #IndustrialRevolution.

#LinearEconomy #CircularEconomy

@peterroopnarine the cherry on top would be an AI trained spending the profits from Bitcoin in order to mint new NFTs... Wait they've already did that, did they?
@peterroopnarine don't tell me from now we are actually going to ashame every fucking innovation for their emission
@peterroopnarine so, that's a lot?
It is yearly consumption of approx. 80 people ( EU citizen yearly footprint is approx. 7t CO2e).
@peterroopnarine I find these analysis interesting but also often misses the idle tax the machines would have had in some other role - its not as if the compute would be unused - a bigger threat I see often is equipment powered on but not yet in service or worse, decommissioned due to age (and efficiency) and still powered on as nobody wants to break anything
@jaredmauch agreed that those are all additional problems. But none of these caused the problem of global warming, they are simply magnifying and accelerating it. Why do we insist on piling on the factors, often excusing it by saying, well, someone else is doing, or would do it anyway?
@peterroopnarine One return flight of a Boeing 767-300 ER from New York to London generates 585 tonnes of CO2-equivalent, using DEFRA's new recommended conversion factor for radiative forcing on long-haul air travel of 1.9.
https://travelnav.com/emissions-from-london-united-kingdom-to-new-york-ny
Carbon emissions from London to New York

Calculate the flight emissions from New York, New York to London, United Kingdom to offset your CO2 footprint.

@Wikisteff yes
Just another part of the problem.
@peterroopnarine nobody should be surprised, really. The “scale up!” paradigm that has prevailed in the last few decades has lead to a dominant strain in engineering, in which brute force approaches are not only the first, but even the only used to tackle (real or perceived) problems. This form of “AI” (yes, these are “George Santos” quotes) are merely the next burp in that gluttenous scheme towards annihilation.
@peterroopnarine Sounds a lot, but the carbon footprint of 35 programmers is about the same.
@peterroopnarine I reckon this might fall back on power companies and supply of materials needed to build renewable infrastructure. There is no end to how the anti AI crowd will distill events to meet their narrative.
@peterroopnarine That is a lot, but US plane flights produce 200 million tons of CO2 emissions per year, and only a few companies have the resources to produce a large language model like GPT-3.
@peterroopnarine frankly, that seems miniscule. I like to compare with leading-nowhere spent resources and energy like fireworks, gambling, and war. Look at how much impact and potential the language model has. Everything has a cost and for scientific progress I am willing to pay a reasonable price.

@peterroopnarine

i guess this does not include the workers... A US person produces 19,9 tonnes of CO2eq per year. So, just 28 workers of OpenAI would emit more that in a year.

@viktorbir no it does not include the workers, but if you read carefully it does include the eventual scaling per user, which will be far greater. And 2 points. 1) I have seen too many arguments that excuse this type of additional emissions by comparison to what individuals already emit. Individual emissions are the sum of usage of services, eventually to include those one. 2) AI emissions will magnify tremendously as number of users do.

@peterroopnarine

If you reread the article you'll see it gives no data about what CO2 equivalent emissions are per search. So, we cannot talk about this.

And, if bing is like google, google's carbon footprint is zero. And it is thanks to use AI in the optimization of the energy use, in fact.

https://www.bbc.com/news/technology-54141899

Google says its carbon footprint is now zero

The technology giant has also pledged to be using only carbon-free energy by 2030.

BBC News
@viktorbir sorry, but Google's announcement deserves more scrutiny than it has been given. While laudable, the footprint is supposedly zero only because of purchased offsets. And the extent to which purchased offsets decrease actual emissions is a difficult calculation. And, that is a footprint calculation that does not account for increases that will result from an expanded use of AI in searches, which is currently unknown, but will be large, and is a focal point of the article that I posted.

@peterroopnarine

The focal point... of which the give no data.

@viktorbir Touche, as does Google. It is a news article that raises a very serious question in need of investigation and cannot simply be ignored because there are insufficient data. Those questions are settled with the gathering of data. Unfortunately, as with many issues of climate change, the impacts accompany the data.
@peterroopnarine and it's not even being used for anything good or useful