(C) Daily Kos
This story was originally published by Daily Kos and is unaltered.
. . . . . . . . . .



Imitative AI Doesn't Need to Be an Environmental Disaster. Maybe. Or Regulation and Incentives Work [1]

['This Content Is Not Subject To Review Daily Kos Staff Prior To Publication.']

Date: 2025-01-28

I am not impressed with power.

I had an engineering professor who hated F1 racing, but not because the racing was largely decided in the garage — he was, after all, an engineering professor. No, what offended him was that the cars did not need to refuel. The sport was all about power. Any idiot, he said, could handle power. Efficiency is where real engineers made their mark.

Of course, that is an exaggeration in many ways. Racing is not all about power, there are other considerations like tire management, aerodynamics, etc. And of course, at the edge of capabilities, constraining and channeling power has its own difficult problems. but overall, efficiency is harder to achieve than power in most contexts. There is even a saying in IT when you run into an issue of constraints on servers — throw more iron at the problem. Throwing more iron — adding more hardware — is seen as cheaper and easier than rewriting the programs to be more efficient. And it almost always is. That is why imitative AI is such an environment catastrophe — it was much easier to throw iron at the problem than come up with more efficient solutions. Unless of course, the environment that you worked in required you to come up with more efficient solutions. And that brings us to Deepseek, the Chinese AI model that is threatening all the Western imitative AI companies.

The Biden administration placed export controls on chips and other components that are used in imitative AI processes. Nvidia’s imitative AI specific chips, for instance, were banned from being shipped to China. So, a group of Chinese scientists did what smart people in their circumstance usually do — they worked within the bounds of their constraints and came up with a model that is up to fifty times less expensive to run than competing models from Western companies. Oh, and it is at least as good as those models. The implication, of course, is that it requires a corresponding lower amount of energy and cooling.

There are caveats. They aren’t disclosing how much it cost to train the model, an important part of the economic picture for the company. Western companies tend to have the same hole in transparency, but it does prevent people from gaining the whole picture. And these are still imitative AI models. They are still hallucination machines trained on material that was taken from artists, writers, and coders without their permission with the intention of using that work to replace them. It is still meant to outsource thinking with likely terrible consequences. And even if Deepseek is 50 times cheaper to run than other models, that does not necessarily mean that it is fifty times less environmental invasive. And even if the relationship between cost and environmental impact is linear, does it matter that much that we can use AI to generate fifty emails as opposed to one before using up a bottle of water?

But Deepseek does show, I think, that constraints lead to technological creativity. There is an argument that we cannot do anything to slow down AI, like make them responsible for the environmental impacts of their models and products, since AI is so important to the future of the US. Put aside whether or not it actually is that important (actually, don’t put that aside. It’s not. Fancy Clippy is not so important that we should ignore its impacts on the rest of society), Deeseek’s ability to work withing constraints pretty clearly demonstrates that imitative AI can be developed without giving tech companies carte blanche. If we made them pay for their training data, if we made them pay for their environmental damage, they would not be dead in the water like they claim. If it was important, if it had actual value, they would do what Deepseek did and find a way.

These people like to portray themselves to the rest of the world as geniuses, people working on the cutting edge of the most important technology in human history. Well, if they really are clever buggers and if they really are working on the most important technology in human history, then they will find a way. If they are just grifters and con-men trying to ride the latest hype wave, then, well, they’ll just whine about how unfair it is that they treat society fairly and work within the bonds of constraints the recognize the externalities they impose on the rest of us.

That their reaction to Deepseek has been panic and claims that Deepseek is just an attack on Western AI that no red-blooded American should use tells me pretty much where their leadership lies on the clever bugger to complete grifter spectrum.

[END]
---
[1] Url: https://www.dailykos.com/stories/2025/1/28/2299542/-Imitative-AI-Doesn-t-Need-to-Be-an-Environmental-Disaster-Maybe-Or-Regulation-and-Incentives-Work?pm_campaign=front_page&pm_source=more_community&pm_medium=web

Published and (C) by Daily Kos
Content appears here under this condition or license: Site content may be used for any purpose without permission unless otherwise specified.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/dailykos/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.