Kulturkampf

[Besticus Maximus]

Though I also wonder if AI will make future plagiarism really difficult to discover. Here you can see him being pretty lazy and copying words verbatim but with AI it's very easy to ask it to rewrite the whole thing

It probably won't. Plagiarists are rarely smart people trying to pull a fast one. They almost always do clearly stupid things like copy things verbatim, entire paragraphs. With AI you run the risk of wrecking the citations, that only work because you copied verbatim. AI hallucinates citations because it doesn't know how they work, it understands them through reverse engineering when they're actually totally arbitrary and established by some academic organization by fiat. And their usage is controlled by convention in the field. AI will get better at this, but it will suffer the same issue as bad academics, not knowing what not to cite. Especially because its source of data is everything not just academic publications.

But the real reason is that AI will catch plagiarism faster than plagiarists will be able to use it to innovate. The existing programs that catch plagiarism are incredibly simple and they've still come a long way. It's been a decade since an undergrad would have been able to use anything verbatim. (Graduate work is rarely run through these, it's assumed you wouldn't do it if you've made it that far.) Attaching a modern LLM to them will all but eliminate it because the LLM is able to understand the sentences. Since it's all probabilistic, you just have to tell your LLM to rewrite it and check it against the version you have because there's good odds that the AI will rewrite it the same way unless given other instructions.

The reason so much gets through is because nobody actually cares it's too much work to check everything. The people who actually do hunt this stuff will absolutely check if an LLM says "this seems fishy" and gives them hints at what to look at.

Where AI won't help and the main issue in academia is the China/India paper mills, China especially has state support for these. Those are already gibberish. They get caught because they don't know how to do the actual methods and so their results tables look insane to the rest of the world. There's no actual data so the AI's couldn't check them.  

Too bad they've done away with the live streaming since then.

Yeah, because fascists like you take it out of context.

Hark at professor chuckles over here