I was always taught that a Masters wasn't intended to prepare one to be competent at novel research; that's what the PhD is for.

A Masters, based on this view, is designed to instill expertise in a field's main body of theory and techniques.

I think, like many things, this varies a lot by field. In computer science, which is still comparatively "new", it is not that uncommon (at top 5 schools, it may even be common) for grad students to publish real research in good venues in their first or second year of grad school (what is nominally coursework time). Even in areas like theory where you're not really hacking away at experiments or building stuff.

Contrast this with say pure math, where (I believe) most students even at the strongest schools do not publish or even produce publishable research in their first few years.

This could again be a consequence of the amount of knowledge out there in both fields. Math is a very mature field compared to CS, though of course CS inherits some of that maturity due to how closely related the two are. First year grad students are publishing CS research because there's still a lot of "low hanging fruit" out there. Lots of open problems have the right lack of popularity, ease of solving, and importance (coupled with the fact that IME many CS grad students actually aren't that good at higher proof-based math) for this to be occurring. Over time the portion of problems that are easy to solve and important will decrease, and I bet theoretical CS's research landscape will look more like CS.

I have also wondered about that. It seems to me that as we better understand an area we are able to develop more powerful abstractions that allow people to reach the limits of current knowledge faster.

> Isn't it inevitable that 4 years of undergrad + 1-2 yrs of grad school classes won't prepare you sufficiently to do novel research?

I think it depends what you call "novel research". Most of the time, novel research means making a small contribution to an existing paper (and for a graduate student, the paper to extend and the contribution are usually suggested by their supervisor).

One help: As in the OP, do field crossing.

Another help: Start with a problem significant in practice. If get a solution with new, correct results, then can say that the results are significant because the practical problem was.

Related Slate Star Codex story: http://slatestarcodex.com/2017/11/09/ars-longa-vita-brevis/

This is a real phenomenon by the name of credential creep:

https://en.wikipedia.org/wiki/Credentialism_and_educational_...

It has very little to do with credential creep. The problem is that, in order to do work at the frontier of a field, you first have to get to the frontier - you have to learn what's going on there, which means that you have to learn all the stuff you have to learn in order to actually understand what's going on there. The learning to actually understand enough to get to the frontier is taking longer and longer. That's not credential creep, that's creep in how far away the frontier is from the foundations.

That sounds like an abstraction problem. I’m totally willing to accept I’m crazy. But my gut feeling is there is a lot of technical debt in representation, so it’s hard to move to the frontier.

In this case, you're probably just wrong. Consider for half a second how many programmers fail Fizz Buzz, and then think about how much work it takes to get from that to the frontier of computational science (i.e. not just installing another javascript library, but actually computation).

There's a lot of low hanging fruit out there, but you still have to actually know how to find it.

I want to point out that Maxwell had like 40+ equations for electricity that wound up being 4. I suspect that many fields have rules, but those rules aren't the most concise representation.

it will always be hard to get to the frontier. but that path should be easy. we can get years and years of explorers by building good roads.

Regardless, will written undifferentiated paper is way better than a paper mill and badly written or worse, lying paper. These happen all the time in algorithms.

Cannot even rely on citation indeed as it is a common practice to cite papers from your surroundings without quality control. Or barely on topic.

This is the opposite of credential creep ... the mathematicians with the credentials were really open minded and excited that someone from the "outside" had a solution! They didn't give a rip about the math pedigree.

That's because CompSci and mathematics is definite. It doesn't matter if someone has PhD's up the wazoo, or didn't finish high school. If the theorem solves something, and it has a proof that shows it, it's accepted.

Other sciences are much more "soft" and have error bars and such. Someone with a "lesser pedigree" just couldn't cope with the special sciences. Nor could they afford the lab.