The AI catastrophe

MIT made a big splash a few days ago when it issued a report claiming that the LLMs can now supposedly “do” most undergraduate level work, so that seems like a big problem.

I remain a deep skeptic about these sorts of claims, although I of course acknowledge that the various intellectual and pedagogical concerns driving this report about the threat AI poses to education and scholarship are very legitimate.

Somebody explain how LLMs are anything other than clever plagiarizing/copyright violating machines? I suppose they can “do” undergraduate work in the sense that plagiarism is an ancient and venerable student strategy for dealing with the apogee of a hangover meeting the perigee of an assignment deadline, but the argument that they represent anything more than a cheap party trick, as opposed to some sort of intellectual breakthrough, remains opaque to me.

I’m particularly suspicious there are literally trillions of dollars at stake, as Carmine Lupertazzi would say, in the selling of Silicon Valley gee-whiz bullshit as some sort of quantum leap in the cognitive evolution of the species.

I’ve seen a bunch of comments from legal academics about the purported fact — this isn’t meant to be dismissive; it may well be a fact, I just have no way of confirming that it is — that the legal academic publishing world is being overwhelmed by AI texts, or pseudo-texts, in the desperate race to publish rather than perish.

People are even going on social media to claim that this or that article was actually written at least in part by AI, and citing AI detection programs to “prove” this, although it turns out that because of the Magic of the Market ™ there are also know programs specifically designed to rephrase AI-generated texts to “humanize” them, that is, make them harder/impossible to detect. I’m aware that the detection programs already have very high error rates, even before getting into the technological arms race of masking programs. All this reminds me rather depressingly of the PED controversies in sports, although hopefully heavy use of Claude or ChatGPT won’t result in a gigantic misshappen head.