@Gargron @liaizon That might be the case for European languages. But for distant language pairs like English → Korean, LLM-based translation is necessary. This is especially true for deep discussions like in @tante's writing.

i think it's really useful that you're trying to separate the technology from how it's situated socially @hongminhee.

some of the critique in this thread, for example how llm's need constant training to remain useful, actually don't hold for your translation example. language changes, but it doesn't change at the speed of hammering the web constantly with ruthless scrapers. firefox' built-in translation would be useful if it were to ship an llm-based model with the browser (they might actually already do that).

there was some initiative to make llms less energy intensive, but so far that's basically zero because economically dirty energy is just cheap enough, and the current u.s. administration is very intent on keeping it that way.

why is the web hammered constantly with obnoxious scrapers? why are gas turbines used to power grok, and why are its data centers built in poor disprivileged neighborhoods? that's the logic of capital, and even more so of specific historic capital relations.

given all that, llms as a technology still have attributes that i find undesirable from a leftist perspective: the transformation of input to output is so complex that it becomes opaque, and training is so resource-intensive that it becomes heavily centralized, and that should be taken seriously.

@liaizon @tante @Gargron