On Mon, 17 Aug 2026, 21:09 Jonas Smedegaard, <[email protected]> wrote: > Quoting [email protected] (2026-08-17 20:13:16) > > > > > > > > On Aug 17, 2026 17:20, Joerg Jaspert <[email protected]> wrote: > > > > > > On 18054 March 1977, Aigars Mahinovs wrote: > > > > > > > Really? Is it? I do *not* see that at all. > > > > > > You must be blind. > > > > > > > There is nothing inherently dirty in the AI technolgies or > datacenters > > > > in general. > > > > > > "There is nothing inherently evil in knifes, guns or weapons at all". > > > Etc. pp. > > > > > > > Just because weak oversight in USA allows *a few* bad apples cut all > > > > the > > > > corners and power their AI datacenters with gas turbines or old coal > > > > power > > > > plants does not mean that *all* AI is dirty and damaging to the > > > > environment. > > > > > > You must have your head very deep down in the sand to not see all the > > > news about the various datacenter / ai companies planning ever larger > > > "factories". With extra coal/gas power plants just for them. Larger > > > values of those for single data centers than needed to run whole > > > countries on this world (and notan under-developed country even). Lots > > > of them companies entirely ditching their "green it" plans (however > they > > > named their plans to go co2-neutral, whatever one may think about those > > > plans in general). > > > > > > That is, for sure, entirely imagined by everyone, right? > > > > > > I'm sure there are ways how the whole AI thing could be run in > something > > > that more people would label as "useful, acceptable" and not "world > > > burning sh*t". > > > > > > You mean like this ? > > > > > > https://d4project.org > > > > > > Disclaimer: this is my employer. Never the less: 100% green energy, 100% > of heat reused in homes. > > > > > > Truth, we don't do training, only inferencing. But that doesn't change > the fact that IT IS possible to run AI in a responsible way. > > Yes, it is possible to solve a part of the equation. > > Similar to nuclear "green" while skipping the question of nuclear waste, > it is possible to "run" AI in a responsible way while skipping the > question of creating it - and recreating it, in a competitive mad race. >
These is nothing magical or inherently destructive about training an AI. It is just running a data center, just like inference, just for a longer stint. But that also is a big benefit for sustainability - you can actually do the training in batches, focusing most load on the times when most green and cheap power is available. The race becomes irrelevant when a model is already there that is good enough for the task at hand. It is possible to solve all parts of this equation as nothing about it is actually all that hard. Technically speaking. And many do. Again, the perfect is the enemy of the good. Just like in the case of nuclear "waste". There are solutions for it. From neutron flux accelerated decay to different reactor designs and reprocessing technology. But people want the impossible - super cheap and reliable technology with zero risk that has been proven over the whole multi-decade lifecycle .... but without ever actually doing it .... because it's too risky *until* it is proven, but can not be proven because it is too risky to prove it. It seems that China might be cracking that nut, but we will see. -- Best regards, Aigarius >
