Part II
"Quals will still require reading deeply from the literature. You will be expected to know all the fundamentals in your computer science area to pass."
First, I saw very little that was "deep" anywhere in computer science, yes, from a career in computing and for some years as a researcher at Yorktown Heights. Yes, a proof of P versus NP would likely be be deep, but that's not in the literature.
Second, I've seen a lot in graduate academics and/or research in math, physics, engineering, and computer science, and I've never seen a graduate program where preparing for the qualifying exams really required reading the "literature", i.e., the journals, deeply or not. Instead there's plenty in the better texts.
For reading the "literature" in a field of specialization: Did I mention that the goal was research and, there, publication? The journal flatly doesn't ask that you have read all the literature in the field beyond what is crucial for your paper.
Indeed, in math, physical science, and more mature parts of engineering, what's in the texts is plenty deep. I can give you a list of texts, say, heavily from Springer, in stochastic processes and stochastic optimal control so that you could count without taking your shoes off everyone in the US who could pass a test on that content. E.g., can filter just a huge fraction of math profs and/or math grad students, likely over 99%, with just the statement, not even the proof, of the Lindeberg-Feller central limit theorem. Can filter a huge fraction of statistics profs with just the Radon-Nikodym approach to sufficient statistics as in the Halmos-Savage paper in the late 1940s. Could throw out nearly all the rest with just the Hahn decomposition approach to the Neyman-Pearson lemma. Just showing that every arrival process with stationary and independent increments is a Poisson process would stop nearly everyone. I know only one proof in print; I improved on that but couldn't pass a test on it; heck, I never committed the whole proof to memory, even when I improved it. Lesson: No one can carry the QA section of the library, even just the texts, between their ears. Can't be done.
I wrote:
"Maybe the US DoD VA would give you a grant."
and you responded:
"Often only professors can apply for the big grants, and writing a grant is actually non-trivial. They need to see a fantastic track record, a solid proven team, and often nods to diversity and educating the public. Some students will help their advisors write the grant, but in the end it's the advisor who doles out the money and is the PI."
We are both correct: Still, for some nice R&D work in human-computer interface for the handicapped, there could be a grant. Many grant sources can just give the grant and not follow the paradigm you outlined. E.g., the US DoD VA has a big hospital across the street from the big NIH campus in Bethesda, and both sides of the street would like to be seen as helping the handicapped, especially US soldiers wounded in battle. There's nothing to keep them from giving a grant.
You should understand the 'hidden agenda' behind the grant situation you described: At the beginning WWII, the US DoD ('War Department') laughed at science. By 1945 the laughing was over, and D. Eisenhower said, "Never again will US science be permitted to operate independently of the US military." Then several faucets for funding were set up: ONR, elsewhere in DoD, AEC (DoE), etc. J. Conant's intention was to have so many faucets that there would be no one place to cut them all off. Then the top US universities got "an offer they couldn't refuse": Take the grant money or cease to be a top US university.
So to the present about 60% of the budgets at the usual suspects are from grants such as you outlined. The overhead per grant is also about 60% and supports the English, history, and art history departments, the Lacrosse team, the art museum, the weekly string quartet concerts, etc.
The hidden agenda is just Eisenhower's, and in particular to have US Federal funding of the top US research universities.
The actual PIs are heavily pawns in this game: As you outlined, the research is very competitive. But as is too easy to see, the research is commonly a bit far from anyone's concerns, of Eisenhower, the DoD, the US economy, etc. So, the NSF keeps trying to make the research more 'relevant', e.g., with 'cross-cutting' programs, etc. Still, one way or another, about 60% of the budgets of the top few dozen US research universities are funded by the US Federal Government, and Congress is not about to change this.
That big old system aside, again, there is nothing to keep someone doing good work in human-computer interface for the handicapped from getting a good grant; all that's needed is just to please some one grant administrator. Just one.
"I feel the attitude of professors/academia being so easy to fool to be somewhat overstated here."
No one's being "fooled" in what I wrote. Again, in really simple terms, to define 'research', look at what's in the journals. That's nearly all the definition. For criteria for a Ph.D. dissertation, don't look for more.
"I won't go into the 'ease of publishing' comments, but my first top-tier publication took 3 years with many rejections, when I was an a non-student doing research. I also did a lot of research that was rejected and went into the 'paper graveyard'."
I have no academic aspirations, but for various strange reasons I've published a stack of papers in applied math, mathematical statistics, and computer science. Statistics and computer science? I almost took one elementary course in each, but not really! In addition my dissertation was clearly publishable. All my papers have been accepted with little or no revisions to the first journal where a submission was made except one case: As a compliment, they wanted me to rename one of my new results from a lemma to a theorem. I waited some years; the journal got a new editor; and he said that the paper was beyond what he could review. So I submitted to another journal. I stated an old theorem with a not very well known stronger version, and the reviewers wanted the original, weaker statement. Okay. Done.
How'd I do that? There is a theme that is implicit already in this thread: Independence.
Or, here's a not so good way: Diligently follow some profs around for a few years and then write something that will impress them. Generally here are close to asking for the impossible.
If crank down the 'diligence' level, this can be made to work, and here's one way: Maybe a prof, say, at Stanford or Berkeley, gets some cash from, say, Microsoft and Google, to do some things in computer and network system management. So, the prof has a dozen or so grad students and lets them go for it. The prof doesn't actually do much or any of the technical work, and the grad students are free to spread out like a dozen scared rabbits.
We should insert: In nearly every field, especially in science and engineering, what is considered the best work 'mathematizes' the field. If you are not so mathematizing, then you are at a disadvantage. But to mathematize, really need an ugrad major in pure math plus some additional topics. Since so few people in engineering and computer science have these prerequisites, for someone with the prerequisites mathematizing the field is relatively easy.
For what I did, it was heavy on independence. So, I started in pure math, and at the ugrad level that's a good thing to do. The stuff in grad pure math looked to me as hopeless; I was wrong; it's only 99 44/100% hopeless! Actually, there is some hope in there, but nearly no one can see it!
So, with independence, I followed various other directions on applied math, while getting paid for it. Did a lot with independence.
Back in grad school, I emphasized selected, advanced, high quality topics in stochastic processes and optimization. The statistics, computing, etc. around was low quality stuff I avoided.
For specific research topics, the high quality background, the selected topics, the independent approach, and usually starting with an applied problem were all crucial.
In particular, in computer science and statistics, I did 'field crossing', that is, brought some of what I knew from outside, built on it, and got results.
Flatly on the research, I never got the problem or any significant guidance from anyone else.
So, when my stuff was reviewed, it was "new, correct (usually theorems and proofs), and significant (solved a practical problem)".
To computer science students I'd say: F'get about computer science. Study math, ugrad pure math, optimization, stochastic processes, mathematical statistics, abstract algebra grad math, get some practical problems roughly in 'computing', do research, publish papers, and pick up a Ph.D. in computer science.
If all the grad students in computer science have "top-tier publications" and still no Ph.D. degrees, then change fields to something in applied math in an interdisciplinary applied math program or something in engineering.
For more, if you really want an academic career, then think carefully a little on where you can make a 'big splash' and how you can do that. But, again, I'd recommend that your work be an example of the 'mathematization' of the field.