https://eraldkolasi.substack.com/p/the-global-energy-and-emissions-footprint

Trying to find realistic estimation of energy and water usage for AI. I’ve read water estimates from “less than most agriculture” to “more than a city”, and similarly wide estimates of electricity consumption and waste heat/emissions.  This paper is one data point that purports to be more accurate, although as with most others the text is cloaked in hyperbole.

Not sure what’s real here, but want to set a marker on the more extreme position.

Note that some analysts also posit that, since the chips involved have a useful life of months before being hopelessly obsolete and requiring replacement, it’s categorically impossible for the industry to make money on scaling up (https://share.google/TCs1bLRPT2T6G6rPw)

https://www.lawfaremedia.org/article/fighting-while-talking–the-iran-war-enters-its-bargaining-phase

https://laurenleek.substack.com/p/temperature-zero-for-culture-why

Why everything is becoming bland.  I get this when playing music on Tidal; each new selection it makes eventually ends up in Pachelbels Canon or some other old chestnut, regardless of where I start: Brahms string quartets, 20th century experimental vocal music, …. She looks at three different systems: shopping streets in London, lyrics in music, movies people watch, and draws the threads together as a failure of exploration of a space vs exploitation of the space, resulting in convergence of local minima rather than an explosion of personalization options.  The solution, she posits, is finding a threshold of variance that allows such systems to climb away from local minima, a value she finds is around 5%.  As she says, this won’t work in a market, because it effectively requires a business to propose one twentieth of its offerings not knowing whether or not they’ll be leaving money on the table, a hard sell for a successful business.  So she returns to regulatory intervention in the markets.