And how is the PhD student trained? By doing research and writing about it by him/herself.

> we don't claim that undergrad education has been blown to bits

But it has been. People pass CS classes without knowing how to write even a simple program. How do you think they'll fare?

As a meme said: you'd better start eating real healthy, because your future doctor will graduate using chatgpt.

▲

delis-thumbs-7e 5 hours ago | parent [-]

>> we don't claim that undergrad education has been blown to bits > But it has been.

Let me fix it: pen and paper. There you go, bada-bing bada-bum. In our uni the exam for Algorithms and Data Structures (one that most people struggle with) is done on paper in an exam room. You better know your binary search.

People, especially wealthy, have been coasting through education paying someone to write their papers since Great Pumpkin knows how long. Now you can do same thing cheaply with LLM’s. The solution is pretty simple, let them fail. Have them show what they can do live in front of examiners. Here’s a computer without network card, only Python and SQLite installed with the Python basic documentation, build me X and prove it works in Z, Y, and Q.

I am studying my second degree and although there’s million and one ways to cheat, I won’t, since A) I want to learn this shot B) I am not sure LLM’s will be available with these capacity and with theses prices in the future, at least I won’t count on it.

▲

tgv 3 hours ago | parent [-]

Live exam, perfectly fine. You can use an LLM as support while you study, no problem, as a glorified search engine. On steroids. From the future. Then you prove you understood it.

Writing an essay, or a thesis, with LLM assist, is difficult to avoid in this scenario, though. It would require your prof, TA or examination committee to read the thing fully and understand the topic deeply. For 101 courses, that's doable, but for a PhD, that's so much work, that it'll break the system immediately.

	▲	delis-thumbs-7e 3 hours ago \| parent [-]
		Montaigne originally used essay (from French essayer, to try, attempt) to test hos on knowledge and understanding, his own capability to create new knowledge. In plain English, you test yourself if you can use your brain. That’s why it is so powerful to write a blog post about a topic you have just learned. Obviously, if you just generate the thing with LLM, there’s no point. Essay is a fairly recent invention and perhaps it was not very good for testing students un the first place, since it kinda was not meant for that originally? So maybe we should test knowledge and understanding in different ways? In STEM-subjects this is not very hard, since basically you either can do the thing or not (like my example of coding above), but I have no idea how one would do this in history or philosophy. Should we go back to having long walks in the garden of the academy, where the professor guide’s you towards knowledge like some 21st Century Socrates? That doesn’t seem doable at this level of university students. And how would you test PhD -level knowledge? You have to build a nuclear reactor from scratch in front of live audience? Or perhaps the oral examination have to become a real exam, not just a pointless ritual? Anyway, I think my point on using essay as a testing structure still stands. Maybe the examination should be like BJJ belt exam, where you have to show what you actually can do in front of peers and your teachers. A lot more people are likely to fail, but so be it. Maybe having a PhD stop being a joke then.