I would argue it's translation with summarization as a close second, but not for people who spend much time on HN since their command of English is almost alway

I would argue it's translation with summarization as a close second, but not for people who spend much time on HN since their command of English is almost always good enough. Tools like DeepL already were cutting into the field of translation/localization, but now we can have something which can basically translate ANY language to ANY language (even things like fictional languages from e.g. LoTR, although that may be very hallucinogenic due to the paltry amount of training data) rather rapidly (gpt3) or rather slowly but more accurately (gpt4) including understanding of things like metaphors and idioms. Not to mention you can probably fix some of the choice of vocab and style just by specifying (once converted to target language) "reword in the stylings of X" or countless other multi-invocation approaches to the previously very expensive and artistic task of converting human thought from one variation to another. The biggest problem is probably over-training on english (and being dependent on openai), but even Microsoft made a LLaMA[!] finetune just for translation which iirc was significantly better than the non-translation versions from everyone else