day ruined

okay, so almost 6 months ago, in march 2023, someone named Steven Piantadosi published what might be called a massive linguistic shitpost, a paper named "Modern language models refute Chomsky’s approach to language". it brought so much controversy that it's been the most popular article on lingbuzz since publication to this day.

Piantadosi is a psychologist with a degree in cognitive science who occasionally touches linguistics (which is normal for cognitive science). what he does in the paper is basically soyboy'ing over AI/ML and LLMs in particular, talking with chatgpt and claiming that language models change everything we knew about language learning. he specifically targets Noam Chomsky (a.k.a. "the father of modern linguistics [and cognitive science]") and his theories. a portion of his critique actually makes sense and is worth discussing, but 70% of the paper, as i see it, is essentially a delusional [schizo]theory apparently caused by poor understanding of LLMs, despite some theoretical understanding of machine learning, on par with that google engineer who believed that LaMDA chatbot is sentinent and yudkowsky whose dumb and groundless takes on AI make the machine learning community laugh at him and hate him every time he once again says something about it.

so it's no wonder that since march there's been at least 3-4 big articles containing "a Reply to Piantadosi" in the title, and i've stumbled upon one of them once again recently, each of them heavily criticizing the original work. but should they? what if it was indeed just a shitpost, trolling, irony? after all, despite virtually all of his preceding and subsequent papers being published in rather good journals, this one wasn't accepted anywhere, not even arxiv, and was published only on lingbuzz, which has no reviewing or other quality control, only relevance control (and apparently has a history of slighlty trollish works by certain authors — wasn't diving deep into that)

anyway, i invite my subs with basic ML understanding to read Piantosi's paper and at least the most recent "reply" article to get an idea of how the chatgpt overhype affected linguistics (as expected). no special linguistic background required, it's kinda more about ML and "science as a science" than the actual linguistics.

(for those 0 people interested in more criticism — [1], [2])

also, folks, especially arch/artix users, if you still have the xf86-video-intel DDX driver installed, consider removing it
i installed it for some reason over 3 years ago, probably because i was under the impression that it's essential or maybe "just in case", but a recent update of it fucked my graphics really really badly and i've spent an hour trying to figure out what exactly caused that
turns out it doesn't work any worse without it in my case, the generic driver provides acceleration for my card just as well
god bless archwiki!

mfw

and some shitty original content like in the old good times

SunVox is a modular synthesizer + tracker, and each module by default has its own little oscilloscope, which can be configured to represent sound signal in many ways: as a normal music player (dots, lines, painted areas etc) or as a "real" oscilloscope, like the stereo phase scope mode (beginning is sunvox software OC, then a real hardware OC for comparsion, very beautiful) or XY mode (similar).

you know what it means. yes. again. 🍎