> If Proposal H has a weakness, it's that it does not distinguish between local and cloud-based LLMs

Is the energy usage so different between local and cloud inference? Both require electricity, the local option even more than the better optimized cloud variant even perhaps. How either is powered makes the crucial difference I suppose. Both can potentially run on solar as well as gas or nuclear.

It's the training that takes the most energy, and that needs to happen for locally running or cloud models regardless.

What am I missing here? EDIT:

Proposal H mentions "LLM usage accelerates the destruction of our ecosystem"

I was thinking solely about energy usage, but there is of course also water usage. A local setup is not water-evaporator cooled most likely.

> Is the energy usage so different between local and cloud inference?

In throughput mode for agentic loads, the energy usage (tok/s/MW) of the new NVidia Vera Rubin hardware is 30x lower than that of the B300 and perhaps 450x lower then the H200 was, which in turn is hundreds of times lower than the inference for single users at home in any non-data-center hardware. It feels like comparing the momentum of an ant to the momentum of an elephant.

This doesn't have anything to do with water usage? The rationale of banning it is that it's copyright status is ambiguous and at the least unethical, it produces lower quality code, and it stifles new developers from getting involved.

People want to ban LLM contributors not because of energy usage, but because AI generated code is unmaintainable and the people who generate it tend to engage with others in domineering and brazenly manipulative ways.