Hacker News Reader

Comment by this_user | Hacker News Reader

I am not sure what point the author is even trying to make here. On the one hand, he seems to complain about how AI has virtually made the traditionally PhD thesis obsolete, but on the other hand, he also states:

> Consider that most PhD theses were never good. How often do you rush to read a PhD thesis? The vast majority of them are painful to read. You learn little if anything.

So, it sounds like nothing of much value has been lost.

I think what he really complains about is that AI is starting to show that the emperor called academia has no clothes. So much of working your way through that system has always been about being able to master largely pointless rituals.

Yet, the people on the inside have no interest in making any improvements, because academia has always been institutionally conservative. But now AI is starting to put pressure on them to rethink their way of doing things, and they really don't like it.

View on HN

Most thesis are never read because anything worth sharing with the wider world ( and some that isn't ) is highly likely to have been published as a paper - not because the work in thesis has no value.

The whole point of a PhD is not to create a thesis - that's just a mechanism to measure - it's to be trained as a scientist or researcher.

Doing a degree in chemistry for example, is largely a knowledge building phase - and in my view it doesn't make you a scientist - being a scientist is about discovering new things about the world that nobody else has - ever - that's what you are learning how to do when doing a PhD.

> Most thesis are never read because anything worth sharing with the wider world ( and some that isn't ) is highly likely to have been published as a paper - not because the work in thesis has no value.

Yes - my wife’s advisor literally told her that the goal should be getting as close as you could to stapling your published papers together with an introduction. They saw this as recognizing that you had learned how to ask good questions, run experiments, and iterate on the results — and since their field had fairly expensive overhead requirements everyone was mindful of the need to repeatedly get grants to keep the experiments running.

> Yes - my wife’s advisor literally told her that the goal should be getting as close as you could to stapling your published papers together with an introduction.

Indeed - in fact some countries ( I think Scandinavian ) do exactly this - the equivalent to a thesis is simply your publications.

Has it's Pro's and cons - you can learn a lot as a scientist but be unlucky in not having an publishable results ( as the answers to your scientific questions were either 'no' [1] or somebody scooped you ).

[1] https://en.wikipedia.org/wiki/Betteridge%27s_law_of_headline...

I think the premise was, before you had to actually do some work to create the thesis. And there was always the concern that yours could be one which was read deeply, so that thesis work had to at least show that you did some work. That there was meat behind the paper.

But now, it could simply be all a couple of prompts to an LLM.

The bar is just lower for not doing the work, now.

But really, that's the fact everywhere.

But it was always about the questions. Physics is difficult because you must formulate the right question to ask, and often once you do, the answer is revealed. Science, I think, it about creating many questions, then many hypotheses and discarding and weeding out the ones that don't work. This is the hard part: The creation of the report SHOULD be the easy part, and why do we care if it becomes even easier? That way we can develop more and more questions and hypotheses. AI cannot really help with the hard parts, it is not creative enough and lacks understanding of the real world.

None of that is relevant without you, the person architecting the questions or the idea, being able to cogently convey this to others. If you cannot understand it enough to transfer the knowledge, or if you are incapable of expressing your inner voice enough to transfer the knowledge, it's all for not.

That's one of the concepts this process is supposed to handle. Validation that you can transfer knowledge. Broken or not, that's the point of it, and what you're replying to indicates that at least in the past, you had to "do the work" to express knowledge, and also demonstrate that you could transfer that knowledge.

Fair point when it comes to training a PhD, but the creation of the actual report is no measure of that. The oral presentation is, along with speaking opportunities and conferences perhaps. Perhaps there are other, more contemporary approaches that could be encouraged in the PhD process?

But the report is! Simply because some people don't value them, doesn't mean all institutions don't. And being able to convey information by text, is as important as the verbal part too. That may not be fashionable in some places now, but I recall being forced to take several english classes, as an engineer, on the premise that "being able to communicate is important".

I agree though, we'll have to look forward, not backwards to solve this.

A good place to plug my favorite thesis to read: Okasaki’s “purely functional data structures” https://www.cs.cmu.edu/~rwh/students/okasaki.pdf .

Very soothing and moves fast.

Man, I had forgotten about this one.

Thanks for the highlight, I read Okazaki's the first time about 2014/15 IIRC.

Another great read is "A Theory of Regularity Structures" - https://arxiv.org/pdf/1303.5113. by Martin Hairer.

It builds on his thesis, Comportement Asymptotique d’ ´Equations `a D´eriv´ees Partielles Stochastiques. - https://www.hairer.org/papers/these.pdf

The English language preprint from just prior to the French is at https://arxiv.org/pdf/math/0109115 "Exponential Mixing Properties of Stochastic PDEs Through Asymptotic Coupling" great, but I read them in revers order from publication and kind of prefer them in that order.

I cant disagree with the premise of the OC, however, there's a feel to Hairer's and Okasaki's insights and presentation that simply is not going to be reached via the lossy compression that is at the heart of LLM.

If that sets a high bar for who gets a PhD going forward, this would be a good thing.

> So, it sounds like nothing of much value has been lost.

Do you actually mean this? To me this sounds like someone said "you never learn anything by reading a high school essay", and you reply "so stop writing them". The point is not the product, you obviously train people by making the product.

> AI is starting to show that the emperor called academia has no clothes.

Seriously, what are you talking about. Academia in the last century has been the most successful engine of knowledge and technology in human history. Lots of papers are junk, like lots of businesses are junk, lots of books are junk. But I don't know how any serious person can say academia has no clothes.

> AI is starting to show that the emperor called academia has no clothes

Why single out academia? We're seeing vast majority of knowledge work had no clothes.

>working your way through that system >pointless rituals >rethink their way of doing things, and they really don't like it

Reply on news.ycombinator.com