Vital insights by the engineers at the International Energy Agency about the energy footprint of AI. (Some of the consultants in this short docu are, admittedly, endearingly stuffy – something that I applaud. Not everyone can be a chirpy influencer.) https://www.youtube.com/watch?v=8M3MciEmC5c Some takeaways: * Singular AI search queries through f.ex. Google Search are not individually that energy intensive. One shouldn't scold people for occasionally using AI search. (All ye besserwissers of Mastodon, understand that it is bad optics to scold those that could become allies.) However, what they fail to mention in the video is the sheer volume and global aggregation of such queries, all of the time. What seems to be the case is that humanity once again keeps Jevons' paradoxing itself into oblivion, inventing ever-more byzantine concoctions of multi-query houses of cards ("agentic" AI; image and video generation; etc.) that eat up capacity. * The ONE BAD THING we ought to keep criticising are indeed the LLMs – not machine learning in general. It's the LLMs and the veritable clown circus around them that is causing the current bubble economy. * Something that might act to rein in some of this bubble energy is indeed the real-world inertia of things like planning permissions, limits to actual capacity, slowness of constructing machinery, and, perhaps, prohibitive energy costs, given the current state of world politics. Of the total data center build-out announced by the hyperscalers and AI corporations, only a small percentage has yet come online – building monstrosities like these takes years!