I'm not convinced. This article focuses on individual use and how inconsequential it is, but it seems like to me it dismisses the training part that it does mention a bit too fast to my taste.
> it’s a one-time cost
No, it's not. AI company constantly train new models and that's where the billions of dollars they get go into. It's only logical: they try to keep improving. What's more, the day you stop training new models, the existing models will "rot": they will keep working, but on old data, they won't be fresh anymore. the training will continue, constantly.
An awful quantity of hardware and resources are being monopolized where they could be allocated to something worthier, or just not allocated at all.
> Individuals using LLMs like ChatGPT, Claude, and Gemini collectively only account for about 3% of AI’s total energy use after amortizing the cost of training.
Yeah, we agree, running queries is comparatively cheap (still 10 times more than a regular search query though, if I'm to believe this article (and I have no reason not to)) after amortizing the cost of training. But there's no after, as we've seen.
As long as these companies are burning billions of dollars, they are burning some correlated amount of CO2.
As an individual, I don't want to signal to these companies, through my use of their LLMs, that they should keep going like this.
And as AI is more and more pervasive, we are going to start relying on it very hard, and we are also going to train models on everything, everywhere (chat messages, (video) calls, etc). The training is far from being a one shot activity and it's only going to keep increasing as long as there are rich believers willing to throw shit-tons of money into this.
Now, assuming these AIs do a good job of providing accurate answers that you don't have to spend more time on proofreading / double checking (which I'm not sure they always do), we are unfortunately not replacing the time we won by nothing. We are still in a growth economy, the time that is freed will be used to produce even more garbage, at an even faster rate.
(I don't like that last argument very much though, I'm not for keeping people busy at inefficient tasks just because, but this unfortunately needs to be taken in account - and that's, as a software developer, a harsh reality that also applies to my day to day job. As a software developer, my job is to essentially automatize tasks for people so they can have more free time because now the computers can do their work a bit more. But as a species, we've not increased our free time. We've just made it more fast-paced and stressful)
The article also mentions that there are other things to look into to improve things related to climate change, but the argument goes both ways: fighting against power hungry LLMs don't prevent you from addressing other causes.
I'm not convinced. This article focuses on individual use and how inconsequential it is, but it seems like to me it dismisses the training part that it does mention a bit too fast to my taste.
> it’s a one-time cost
No, it's not. AI company constantly train new models and that's where the billions of dollars they get go into. It's only logical: they try to keep improving. What's more, the day you stop training new models, the existing models will "rot": they will keep working, but on old data, they won't be fresh anymore. the training will continue, constantly.
An awful quantity of hardware and resources are being monopolized where they could be allocated to something worthier, or just not allocated at all.
> Individuals using LLMs like ChatGPT, Claude, and Gemini collectively only account for about 3% of AI’s total energy use after amortizing the cost of training.
Yeah, we agree, running queries is comparatively cheap (still 10 times more than a regular search query though, if I'm to believe this article (and I have no reason not to)) after amortizing the cost of training. But there's no after, as we've seen.
As long as these companies are burning billions of dollars, they are burning some correlated amount of CO2.
As an individual, I don't want to signal to these companies, through my use of their LLMs, that they should keep going like this.
And as AI is more and more pervasive, we are going to start relying on it very hard, and we are also going to train models on everything, everywhere (chat messages, (video) calls, etc). The training is far from being a one shot activity and it's only going to keep increasing as long as there are rich believers willing to throw shit-tons of money into this.
Now, assuming these AIs do a good job of providing accurate answers that you don't have to spend more time on proofreading / double checking (which I'm not sure they always do), we are unfortunately not replacing the time we won by nothing. We are still in a growth economy, the time that is freed will be used to produce even more garbage, at an even faster rate.
(I don't like that last argument very much though, I'm not for keeping people busy at inefficient tasks just because, but this unfortunately needs to be taken in account - and that's, as a software developer, a harsh reality that also applies to my day to day job. As a software developer, my job is to essentially automatize tasks for people so they can have more free time because now the computers can do their work a bit more. But as a species, we've not increased our free time. We've just made it more fast-paced and stressful)
The article also mentions that there are other things to look into to improve things related to climate change, but the argument goes both ways: fighting against power hungry LLMs don't prevent you from addressing other causes.