Hacker News Reader

I think an important thing to take into account is that a very important part of what makes a good scientist is not just how good they are at "doing science", but how good they are at communicating it. A good scientist is a good communicator. Even if that means finding people who you can work with who are good communicators, it's part of the job. Coding up a really clever solution and just kind of hoping people will come across it and appreciate is not really doing science. (Not saying that's what you are doing, I'm just going on a tangent here.) The point is, ever since the enlightenment, a strictly critical part of the job of being a scientist has been the ability to elaborate your ideas to others, and to argue for its merit.

So yeah, doing the work is the fun and motivating part, but a big part of the reason that the academic community values well-written scientific papers is because all that work can go to waste if it is not described (and defended!) in a way that is clear and convincing (to the relevant audience of course).

Grad students may complain about having to learn to write effective papers, but it's literally just as important as the technical work they are doing --- maybe moreso, since in the future they may very likely change topics, and the writing and communication skills is the part that will carry forward throughout their career.

I'm reminded of the saying that "Publishing research without data is simply advertising, not science". Not even publishing the research means it doesn't have a chance at advancing the state of the art. I don't understand how such research can be successful at anything other than selling a few extra copies of a software product likely to disappear within a handful of years. Unless it's implemented as the absolute killer feature of a very successful product it's very unlikely anyone will ever know there's anything novel in there.

The research does get disseminated, just slowly. The transmission vector is contact with other people with firsthand knowledge of the research. It becomes an oral tradition, which makes it difficult to break into and is inefficient. I've come across many non-publishing research communities in various parts of computer science that continue to develop and share the state-of-the-art amongst themselves. The research isn't being lost, it is just compartmentalized.

For example, there is an elegant cache replacement algorithm I learned from a database kernel engineer a decade ago, one of those algorithms that is so obvious in hindsight that you can't believe it never occurred to you. It solves a key problem with the algorithms everyone knows in literature. I've used it ever since and taught a bunch of other database kernel engineers the same algorithm. I have no idea where the person I learned it from got it, but the literature still treats it like an unsolved problem. Many engineers are using it in more and more systems, and there are a few different variants in the wild, but no one has bothered to write it down AFAICT.

The salient point is that academia is not the only community that develops and defines the state-of-the-art in computer science, but at least they publish. Without an incentive for non-publishing research communities to publish, it becomes difficult for an outside academic to break into some fields because there isn't an obvious entry point.

Is there a chance you can point us to a code snippet of said cache algorithm, or elaborate a bit?

If it's simple and elegant in hindsight this insight might be interesting in other CS applications. Not cross sharing and reinventing wheels is a somewhat sad implication from the compartmentalization you mention.

Is the issue with publishing your findings the grind of submitting to a journal and going through the standard review process? Because in that case, would just throwing up a high level overview on arxiv work for you? It sounds like you aren't adverse to sharing the results of your research if the activation energy was eliminated (or substantially reduced).

It is possible to recognize how much the publishing process is useful and still feel it is perfectible. It made me think a bit about the state of research when Paul Krugman was explaining that at his point, communication between researchers basically happens between arxiv and blogs. Official publications come at a later time as part of a validation of the whole process, but its cycle is too slow to be useful anymore.

In CS and especially in things related to deep learning, I can understand that pausing to publish can be felt like a delay.

I feel like the typical publication format is not well adapted to CS. For instance, a big point of formal publications is to give other researchers the tools to reproduce your experiments. Why is it not the norm to offer source code or even docker containers then?

Grad students complain about a lot of things, but I’ve literally never heard one complain about having to write good papers.

They could complain, however, that PhD training contains shockingly little on how to write effectively, given the skill’s importance to most post-PhD jobs. To make matters worse, any training is often front-loaded at the beginning of a program so those skills are atrophied by the time PhD candidates have their own data to write up.

If I am ever in charge of a graduate program, I’m putting everyone through a 10 week expository writing class around year 3.5-4. I took one in grade 12 and that public high school class literally beat most Ivy League grad seminars for (academic) career prep.

Literally downthread: https://news.ycombinator.com/item?id=17432281

I think the dissemination part is a collective action problem. Simply put, society has not set up opportunities to induce/encourage researchers outside the academic scaffolding to share their work. Given that, it comes down to a question of who cares most about the dissemination. If neither the person who would be writing, nor the putative audience cares about it enough, who would want to put effort into the dissemination process?

Creating such a two-sided market for research output is a hard problem, roughly analogous to what Uber has done for ride-sharing. The current academic/think-tank setup is an intricate triangle consisting of the grant process to match donors with researchers seeking funds, the publication process to match research output to downstream demands, and publication metrics (citation count, reputation, impact factor, etc.) to match downstream demands with funding agencies.

The academic publication process is a mess, but you don’t have to go down that route: write documentation and blog about it. The scholarship is in how you compare and contrast with other similar solutions, and how you describe where you got ideas from.

I had this issue during grad school. The research was the fun part. Writing my results up into a paper and formatting it for submission to journals... not as much fun.

Academic here. I agree with you, and I think the majority of academics would. After all, most of us get into the field because we like research itself, not because we like writing. But the thing is that research is useless if you don't communicate it. A researcher that doesn't write is like a cook that throws their food directly to the bin, or a singer that only sings in the shower. It's legitimate that they do it for fun, but no one would consider them to be a cook and a singer, or take their work seriously.

Anyway, if you had kept doing research after grad school and climbing the ranks of academia (I assume from the wording that you didn't) you would see that writing papers is far from the worst part... in grad school I often did research to procrastinate writing a paper, now I often write a paper to procrastinate writing a grant request or doing paperwork... :)

Have you considered finding a partner who has time and is willing to write things up for you?

You need grad students to write papers for you...

Teaching them how to write good papers takes way more time than writing papers yourself.

I think I'm a good bit earlier in my career, but I've been in a similar situation for a number of years now. It was extremely discouraging to me when I discovered that the typical result of my spending years (in some cases) researching and developing some software/ideas[0] would be far from gratitude and offers of financial assistance or employment to pursue these obviously important subjects ;)

Instead the breakdown of responses has been something like 20% dismissals following a cursory reading, and suggestions that what I was doing was some other extremely commonplace thing I was already well aware of and could precisely relate to what I was actually talking about[1]; 30% people saying essentially, "ooh, neat!" and then moving on; 20% inquiries from genuinely interested folks, often times professional researchers (often possible to get the main idea across with them, but still: unless what you're doing relates in a very direct and positive way to their work, that's not going to go anywhere); 30% various, with a surprising amount of hostility mixed into all the categories (except the "ooh, neat!" people).

It has certainly helped in the professional realm (for software engineering, not research)—though not in the way I would've hoped. By far the most typical thing is some potential employer sees your work and is like, "oh wow, he must be good to have made this—I can get him to write LOTS of CODE for me!!" but they'll use their knowledge that you value developing original ideas to lure you in, pretending they also care about your ideas, and then allow the memory to slowly fade as your new role doing totally vanilla software engineering solidifies.

I actually had someone offer to give me a place to live and pay monthly expenses and tried to arrange for me to move to another state (all the time claiming the world needed the ideas I was working on). I told them about my experience with the above and said that I was maybe paranoid but wasn't getting into something with potential for that again; he replied that it wasn't paranoia but wisdom and we ended our discussion.

So now... I realize if you work hard and develop new ideas that may (or may not) end up being important, no one's going to help you unless you turn it into a product or do it in the context of Academia. It's hard work to communicate new ideas, and all too easy to just let them sit rather than doing that hard and often unforgiving work. But I'm still excited about working on the ideas so whatever, for now.

----

[0] To give some context for the kind of ideas/software I'm talking about: I (re-)discovered the idea of a structure editor/projectional editor, not knowing what a grammar or AST even was at the time (I was going to develop a 'linguistic constraint description language' before discovering formal grammars), and built a novel incarnation that may still have lots of potential (almost forgot: it also included a modular system that did something like language-agnostic full-language auto-complete; essentially the inverse of a 'compiler compiler' so I call it a 'generator generator'). That project was largely re-discovery, but led to a formulation for a new general architecture for programming tools built around very simple 'program models' and 'language models' (this is what I'm most interested in pursuing now). I also developed a new kind of general purpose interactive data structure visualization system, which I see as an incarnation of a more general notion of an 'abstractoscope' (something to allow us to usefully/systematically remove detail, rather than adding it like e.g. a microscope).

[1] I read recently that Douglas Engelbart had so much trouble with this precise phenomenon (people assuming whatever new idea you're talking about is some common thing from the framework they're used to thinking in), that it inspired him to give a more explicit philosophical grounding/context/framework for his idea, which resulted in a famous piece of writing of his (can't remember the name atm).

> By far the most typical thing is some potential employer sees your work and is like, "oh wow, he must be good to have made this—I can get him to write LOTS of CODE for me!!" but they'll use their knowledge that you value developing original ideas to lure you in, pretending they also care about your ideas, and then allow the memory to slowly fade as your new role doing totally vanilla software engineering solidifies.

I laughed. While I've never heard employer say that explicitly, it fits with my experience. Many employers seem to think solely about what a candidate can do for them and not in terms of what they can do for the candidate.

When phrased like that, the employer might say something like "But we have generous pay/benefits/whatever!", which just shows that they don't understand their problem. I'd gladly take a reduction in pay and benefits if it gave me more control over my work.

It might be worth asking a potential employer who claims to value your work to state in the employment contract that you can choose X hours per week for your own work. This would be a more formal way to test how serious a potential employer is about your work. Unfortunately few employers seem to be willing to make that deal, but after I finish my PhD I'll be asking.

Hehe, yeah, it's been surprising how common that pattern is. I fell for it once when a prospective employer found a demo video of mine on the front page here and sent me an email saying, "I can offer you the opportunity to continue and extend your work here at <company>"—but since then I can smell the trap coming and have been able to avoid it.