by JohnFen on 6/9/2025, 2:19:56 PM
> Am I missing something here?
Perhaps that the base environmental cost of a living human being is incurred no matter what they're doing. Replacing them with LLMs is just adding a new cost on top of the existing one. It is not removing the cost of the human.
by incomingpain on 6/9/2025, 2:15:01 PM
>could LLMs actually be a net positive for the environment?
They produce a tremendous amount of heat in usage and at power generation. In most circumstances are powered by fossil fuels. I cant see how this would ever be justified.
>Should it earn all the benefits that come with ESG status?
Not a chance. AI is literally an antithesis to climate folks and ESG.
The ESG/climate folks should be opposing AI as much as possible.
by Jackpillar on 6/9/2025, 2:24:48 PM
What? AI is horrific for the environment from the embedded energy from the production necessary for the hardware and the energy required to train & run the LLMs. They use an insane amount of water and the resulting emissions are that of small countries.
Not only that, your argument depends on the saved emissions/energy consumption from laying off someone. Which is first and foremost just dark - but also how do you know that the laid off SWE isn't 10x more energy intensive when they're at home? Driving a lot, flying a lot, has a Ford F350, 5 kids and now runs AC all day while they're home?
I've been wrestling with a notion lately from a recent HN discussion, and debating with a colleague: could LLMs actually be a net positive for the environment? Should it earn all the benefits that come with ESG status? It's a bit wild, but we haven't found a solid, economically defendable argument against it. it assumes a claude (200$/month) subscription can in many cases replace a 10K/m SWE.
The best we've managed to push back with is that the environmental cost of a human worker, like their commute or office space, would be there anyway. But that feels like a cop-out; it basically means no technology can ever really improve productivity in an environmentally beneficial way, and we don't apply that logic to anything else.
So on one side: the environmental cost of the LLMs themselves (as indicated in the subscription price). On the significantly higher cost of a human engineer. Am I missing something here?