Want to take part in these discussions? Sign in if you have an account, or apply for one below
Vanilla 1.1.10 is a product of Lussumo. More Information: Documentation, Community Support.
@52: MO moderation policy works a bit like the law in the US: We lay out general principles in the FAQ. Whatever users with votes to close happen to look at page then make a decision based on their understanding of the principles involved. If things get too contentious, it will go to a discussion on meta. In very rare cases, generally involving things that require technically moderator intervention (banning users, for instance), the moderators will step in and do something.
The key point is one Scott made earlier: MO requires a lot of moderating, it doesn’t require a lot of moderator attention. This is a very important distinction.
@John Baez: On the stackexchange network, moderation rules and user powers are organized to evolve with the sites’ maturation and growth (73 sites now and counting).
There are currently three stages - Private Beta - Public Beta - Public Launch. At each stage it is more and more difficult to have the points needed to access moderation-like functionalities. Prior to Public Launch, moderators are elected, and regularly renewed.
There is not a too strong separation on these systems between trusted user and moderators, more a continuum of collaborating users, with stage filters to avoid noise from the user base eating the time of moderators.
You can read a lot about his views on moderation in Jeff Atwood’s blog where this is a frequent topic. They also discuss and compare their problems with the ones faced by other sites such as reddit.
Hi ogerard (#53),
Thanks for the comment - very informative - and thanks also for the link to Jeff’s blog (although it seems he left the company two weeks ago).
I think this points idea is a very good one both for the discussion of papers and also for the more ambitious goal of peer review. However, some people have an aversion to a points system, so it would have to be rethought somewhat. Perhaps, the option to hide your points, i.e. make points hidden by default, might help so that it does not become a popularity contest.
I’ve used the phrase “points for grown ups”. Is there a site that allows you to hide points? Any site that takes points more seriously than MO? I’ve been thinking about how to quantify credibility both in the editorial process, but also quantifying credibility of an article (and ultimately the author). Any thoughts would be appreciated.
Thanks, ogerard and Ben.
A point system seems necessary to make a Selected Papers Network largely self-moderating. If they system actually becomes popular, relying on a few pre-chosen moderators won’t work. They’ll eventually burn out. A point system allows the system to scale up.
But I agree with Eric: the point system would need to be set up intelligently. We need to think carefully about this.
I’m not sure the point system at MathOverflow deserves the scorn some people seem to pour on it. There are some things people love to complain about that are nonetheless necessary, like taxes.
What do the MathOverflow bigshots think about this? Do you occasionally tweak the point system to improve it? Or: are you considering this?
It seems like another big decision involves how much the comments should be required to tightly focus on the paper at hand, versus letting them drift off into open-ended discussions. This is a ’slippery slope’ issue: it’ll take work to establish a clear policy.
One possibility is to allow different kinds of discussion, but have them marked as such. This is also a bit tricky.
Just some opinions: (1) I feel that we can improve on the present refereeing by facilitating: (a) Categorised responses: whether proofs are correct is independent of how interesting. It is often clear the result is interesting if correct. Further, originality, technical difficulty etc. are different aspects of ‘interesting’. (b) Reviewing a part of the manuscript: Someone may vouch for the correctness of a part of paper, but may not wish to do so for the rest.
I think in a grown up points system, an individual’s points should not be on display (why would it?) and merely determines the degree of editorial rights.
I also think points can be conferred onto others at the expense of reducing your own points. For example, if John racked up enough points to become “Senior Editor” ten times over and his friend, a well-known mathematician joins, John should be able to promote him immediately to “Senior Editor” without forcing the mathematician to start at zero. However, it would be important that points remained scarce (like a kind of currency) so John ’s ability to promote people is limited by the points he has.
Still thinking out loud during a coffee break…
I just wanted to point out a site called PaperRater.org in case people hadn’t seen it. The aim seems to be to allow people to rate and comment on papers, and it interfaces with the arXiv.
One thing that already exists is PLoS ONE: the refereeing only asserts the equivalent of correctness (PLoS is mainly run by biologist and medical researcher, as far as I know), and the importance of the results are left to the reader to assert, with a commenting system.
PLoS itself is quite a success (but the author-pay model does not suit the mathematical community, I think), but the comments are very rare. I feel like interfacing commenting with reviewing inside the current journal system could be a good starter.
As this will be a place to discuss mathematics, you’re likely to want to be able to display mathematics via the web. Have you thought about how you’re going to do that?
I’d like to direct your thinking a little on this by pointing you to a couple of answers of mine on http://tex.stackexchange.com: my main argument in favour of MathML, and a follow-up.
This forum is fully maths-enabled. The input format is iTeX which gets converted to MathML. This can either be displayed as-is (if the browser can cope with it), converted to an image, use MathJaX to convert the MathML to HTML+CSS (note: it converts the MathML not the original input), or the plain source can be displayed. This optimises speed (as the conversion is done server-side), flexibility (as the actual output format is customisable by the user), and accessibility (both MathML and the original source are available).
Do not try to convert LaTeX to anything. One simple \newcommand{\a}{\a}\a
would kill your server!
iTeX is a C library that converts a strict subset of LaTeX mathematics to MathML. Once it is encoded as MathML then you can use MathJaX for broken browsers, and compliant browsers work just fine. There are no security implications, unlike a real LaTeX converter.
This site uses Markdown PHP Extra for basic syntax and iTeX for mathematics. It took a bit of work to get them to play nicely together, but I have it all in place. If you’re using PHP I can just give you the files as they are. If you’re using a different language, I’ll need to extract the pieces to be translated but it wouldn’t be hard.
It is a solved problem, that’s what I’m trying to say. Depending on the language you’re using, it might need a little work but only in the mechanics. We use it here, on the nForum, and on the nLab (about 6000 pages of mathematics).
I agree with Eric in 60: a system where using the site eventually confers editorial privileges is good, but at maximum the level of editorial privilege should be publicly shown, not any kind of “score.” The name “reputation” on MO was genuinely a mistake, but I think no name is the best name in this case. There’s a non-insane argument that not even that should be shown (though it probably should be publicly viewable who “voted” to shut down a conversation as on MO).
There’s good news and bad news. The good news is that you can get iTeX to work as a python extension via swig. The bad news is that python is one (of the many) languages that I’ve not used so I can only advise how to integrate it with markdown in generic terms.
You can get the original source for iTeX2MML from Jacques Distler’s web pages. It is open source. I’ve added the swig bindings for python, and you can get them from my web pages (it’s a BZR repository, though browsable as well). Mine might be a bit out of date wrt to Jacques’ version - I ought to update it but in the meantime, the Makefile from mine ought to work with his.
Then you’ll need python markdown.
Integrating iTeX with Markdown is a little work to get right, but not too much. I know how to do it for PHP, but I’ve no experience of python so although I can explain what needs doing, I don’t know exactly how to do it. Markdown works by calling a set of parsers in turn. So you need to define two new ones: an inline parser and a block parser. I put these at levels -20 for inline maths and 62 for block maths (this are for PHP Markdown, python might use a different system of figuring out what order to do the parsing). There’s one extra step: Markdown escapes certain characters, but most of these shouldn’t be escaped for iTeX. So you need to run the escaping once with a limited set (namely, to escape \$
), then run the maths parsers, and then run the full escaping system afterwards. Again, not difficult to do.
I think that Yla has a point (#66) that this discussion is fragmenting slightly and it’s getting tricky to follow the subthreads. Remember that anyone can start a discussion here - it is a forum, after all! - and if we get quite a few on this I could create a new category for discussions related directly to this project. For now, maybe discussions relating specifically to this could have SPN in their title, or be tagged SPN (though the tags aren’t visible from the main discussion list page).
Did you read the links to tex.stackexchange that I gave above? For example
[MathJaX] eliminates the need for users to install math fonts
by forcing them to download them every time! (This is a major myth about MathML. The STIX fonts are now released so no mathematician has any excuse for not having a proper font on their computer that supports all the mathematical parts of unicode.)
When I’ve used, say, MathOverflow on an old laptop on a dodgy connection then MathJaX makes it unusable because it takes so long to download everything. It’s okay for a site that has only a few bits of mathematics, but anything with a large amount (say a typical MO question page) rapidly becomes too slow to work with.
The solution I outlined above: using iTeX server-side and then MathJaX client-side to convert the MathML if needed is the best of all possible worlds.
Lastly, this isn’t something that you can change half-way through without a lot of bother. You need to get it right from the outset. The reason being that MathJaX’s TeX-like syntax and iTeX’s TeX-like syntax are not the same (and note that both are TeX-like. Neither is true TeX - and you don’t want true TeX for lots of reasons - so going from one to the other is going to be a nightmare.
But it’s not just a benefit for Firefox users (and you’re ignoring the fact that there’s an IE plugin which makes it possible for IE to render MathML). Server-side conversion is a benefit for all users as MathJaX is faster when converting MathML than when converting TeX. Accessibility is also a major consideration here: don’t assume that all mathematicians have normal sight.
By “incompatible syntax” I did not mean just the delimiters. Neither iTeX nor MathJaX is real TeX and so both just support a subset of (La)TeX commands. The lists are not the same. So if you switched, you would have to translate everything from one format to another.
I notice you are making commenters on this forum explicitly turn on markdown / iTex for each comment rather than simply making it the default. I’d be interested to hear your thinking behind that design decision…
Laziness? Stuipidity? These should not be discounted.
Anyway, it isn’t for each comment. I don’t remember exactly how it remembers it, but the system does remember your default. Or at least, it ought to. If it isn’t then something is broken.
As for which is presented to new users, then Text makes more sense because it does what it says on the tin. If someone has never heard of Markdown then they may be taken by surprise when their carefully written text gets all mangled (I recently edited someone’s post who’d written about some costs in dollars. It looked a little odd because the dollars weren’t escaped and so the bit between got rendered as mathematics.).
While we’re at it, here’s another tip: allow users to write pure XHTML in their posts. Then you don’t have to worry about the fact that you’ve got to anticipate everything someone wants to do. We don’t often have diagrams on the nForum, so we don’t have a syntax for it. But it’s possible to cut-and-paste some SVG code directly in and have it render correctly. Of course, this leaves you open to certain types of attack, but then you pass the output through a validator and sanitiser which makes sure that no dangerous code gets through. Doing it on the output is more reliable (the output has to be XHTML so you only need to understand that), and leaves you more flexibility in input formats.
Andrew: you wrote (75):
no mathematician has any excuse for not having a proper font on their computer that supports all the mathematical parts of unicode
I hope your tongue was in your cheek.
Most mathematicians have no idea how this stuff works. Most mathematicians don’t know what unicode is. What percentage of mathematicians do you think have downloaded the STIX fonts? I have, but only because I’ve seen your encouragements to do so. I’m grateful to you for suggesting so forcefully that everyone does it, and for telling us how easy it is, but most mathematicians don’t read the nForum.
I don’t know whether you count “having no idea how this stuff works” as an excuse. I’d call it more than an excuse; I’d call it a reason.
But I suspect your tongue was, indeed, partly in your cheek.
But only partly.
I like your distinction between an excuse and a reason. If someone has a reason why they haven’t installed the STIX fonts then I feel that I can gently explain just how easy it is, point them to simple instructions (for example, it’s in all the major Linux distributions now - even Debian, which isn’t known for having the most up to date stuff), and maybe they’ll do it. An excuse is just that: an excuse, an “I can’t be bothered, why doesn’t everything just happen the way I want it to?”. The STIX fonts are also included in the latest TeX distributions - did you know that? I use them in all of my lectures now so that I don’t have to keep loading obscure packages just to get some strange symbol.
My underlying gripe - and why my tongue is only partly in my cheek - is that we’re meant to be professional mathematicians, and yet we often behave like a bunch of amateurs. Whinging about downloading some fonts is amateur behaviour.
It’s not just us mathematicians, it’s academics in general. Take job applications: “We want the best candidates”. Well, of course! But what exactly is meant by “best”? No-one really knows, and of course it differs by institution and country. But no university is ever going to say anything other than “We want the best candidates”. It’s an enormous game of bluff. One discussion that I really want to have here is about how as a mathematician I’d like to be evaluated. Rather than just complaining (whinging) about impact factors, I’d like to figure out just how we decide, “Am I actually making a positive contribution, or should I get out of academia and go and do something more useful with my life?”.
Andrew wrote:
This is a major myth about MathML. The STIX fonts are now released so no mathematician has any excuse for not having a proper font on their computer that supports all the mathematical parts of unicode.
Does this mean that it is necessary to install fonts in order for your preferred solution to look right? If so, this seems like a fatal objection to me. A significant fraction of mathematicians are not sufficiently computer-savvy to do this. The same goes for installing a browser plug-in.
In my experience, the TeX on MathOverflow (using MathJax) has always looked pretty good and loaded reasonably fast, while the TeX on the nForum and n-Category Cafe often looks weird and sometimes loads slowly. Most people seem happy with the way TeX works on MathOverflow.
I just tried a highly unscientific experiment. I went to the nLab home page and timed how long it took to download the page (I have a plugin that measures it accurately). I have a plugin that allows me to change the user agent so I did it once as firefox and once as IE (without MathPlayer). As Firefox, it does not use MathJax, as IE then it does. The download time doubled, and the page size went up by about a third.
Does this mean that it is necessary to install fonts in order for your preferred solution to look right?
Only when glyphs are used that you don’t already have on your computer. Then it’s a case of either you install them yourself, or the browser downloads then every single time.
In my experience, the TeX on MathOverflow (using MathJax) has always looked pretty good and loaded reasonably fast, while the TeX on the nForum and n-Category Cafe often looks weird and sometimes loads slowly. Most people seem happy with the way TeX works on MathOverflow.
I just looked up where you are. Assuming that I’ve found the right Kevin Walker, then I’m not surprised about this. MO is hosted in the US on the back of a major company (stackexchange) with huge budgets and large amounts of money from venture capital. The nLab is also hosted in the US, but on a VPS that I pay for. The nForum is hosted in Norway on my University account. So the expectation should be that MO is quicker than the nLab. But my experience is otherwise.
As for the look, well, again there’s the next strength of MathML. Because it is part of the XHTML specification, it is completely under the control of CSS and so is very easy to style to make it look nice (for why the nLab doesn’t look as polished as MO, then I refer the honourable gentleman to my previous remark about who pays for what).
But let me point out, again, that my preferred solution actually gets the best of both worlds. Those who want MathJaX to control what their output looks like can do so. Those that don’t, don’t. The thing is that no-one is forced either way and those that can cope with downloading fonts and using a compliant browser get the speed advantage and other benefits without taking anything away from the others. One could even have an option in your user account that said “Even when I’m using Firefox, use MathJaX to render the maths”. Hmm, maybe I’ll do that on this forum - then you can test it out properly.
Let me end by twisting your words a bit!
“Does this mean that it is necessary to have javascript enabled in order for your preferred solution to look right? If so, this seems like a fatal objection to me. A significant fraction of mathematicians are sufficiently scared of malware to do this.”
In my solution, you can look at the page with javascript or without, with downloading fonts or without, with your own CSS applied or without. By doing server side conversion to a known format, all of this is possible. By shoving the conversion to the browser, it all gets complicated.
Hmm, I’m beginning to feel as though I’m getting a bit belligerent about this issue! It is something I’ve thought about a lot, and it’s something where I feel that lots of people have opinions without having thought too much about it (I’m not casting aspersions here), and that tends to make me a bit more forceful than I ought to be. I’ve gotten into arguments elsewhere about this that I wish I hadn’t.
So I’ll shut up now, with apologies to anyone who’s found me too argumentative on this.
@Andrew #83: From my point of view you come across as enthusiastic, not belligerent.
My main point is that the best system is the one which works best on widely deployed web browsers in their default configurations. In my experience, TeX rendering on nLab has more formatting glitches that TeX rendering on MathOverflow. I have some screen shots illustrating this, but I’m traveling today and so won’t have a chance to upload them until late today or maybe tomorrow.
Here are the screen shots I promised in the previous comment. I made a test page on nLab, then entered the same source as a MathOverflow question. I did this with Firefox, Safari and Google Chrome, all on a Mac.
The grey backgrounds for the MO images are due to the “preview question” feature – I didn’t actually ask an MO question with that text. (Though it is fun to imagine the lengthy meta.MO discussions that might have resulted if I did post a question like that, perhaps prefaced by “Please ignore, just testing” or “How does this look on your browser?”)
For all three browsers, I think the MO rendering looks significantly better than the nLab rendering. In my opinion, none of the arguments Andrew gives above outweigh this. (I’m assuming that what Andrew proposes for the SPN site is the same as what nLab actually does now. If this is not the case, I would be interested to know what the differences are.)
Firefox, nLab:
Firefox, MO:
Safari, nLab:
Safari, MO:
Chrome, nLab:
Chrome, MO:
The Safari and Chrome versions are identical, so far as I can tell, which is not surprising given their shared genealogy.
Very interesting (incidentally, the Sandbox is the usual place for experimenting).
What is of particular note there is that the biggest differences are with Safari and Chrome and … *drum roll* … both are rendered using MathJaX. This suggests that the difference is more about configuration than anything else, and that anyone who really wants to read mathematics on the web should use Firefox (no, they didn’t pay me to say that!). In all honesty, I suspect it’s a mix of several things one of which is that the nLab is optimised for Firefox.
As for outweighing. Hmm. Not so sure. CSS and Javascript are Dark Arts and much can be accomplished by them. I’m not sure what are the outweighing differences that you see. The majority of the differences that I see are down to fonts, and whilst the computer modern fonts are traditional in TeX, they were designed for the printed page: reading stuff on the screen is different to reading stuff on paper. For example, that hat is almost invisible in the MO pages and I find that when reading mathematics on the web then I’m more likely to miss things like that than when reading them on paper.
Here is what MathJaX is doing to produce the $f(a) = a'$ on MO:
<span class="math" id="MathJax-Span-34"><span style="display: inline-block; position: relative; width: 3.592em; height: 0pt; font-size: 124%;"><span style="position: absolute; top: -2.59em; left: 0em; clip: rect(1.651em, 1000em, 2.973em, -0.677em);"><span class="mrow" id="MathJax-Span-35"><span class="mi" id="MathJax-Span-36" style="font-family: STIXGeneral; font-style: italic;">f</span><span class="mo" id="MathJax-Span-37" style="font-family: STIXGeneral;">(</span><span class="mi" id="MathJax-Span-38" style="font-family: STIXGeneral; font-style: italic;">a</span><span class="mo" id="MathJax-Span-39" style="font-family: STIXGeneral;">)</span><span class="mo" id="MathJax-Span-40" style="font-family: STIXGeneral; padding-left: 0.313em;">=</span><span class="msubsup" id="MathJax-Span-41" style="padding-left: 0.313em;"><span style="display: inline-block; position: relative; width: 0.857em; height: 0pt;"><span style="position: absolute; top: -2.59em; left: 0em; clip: rect(1.972em, 1000em, 2.777em, -0.513em);"><span class="mi" id="MathJax-Span-42" style="font-family: STIXGeneral; font-style: italic;">a</span><span style="display: inline-block; width: 0pt; height: 2.59em;"></span></span><span style="position: absolute; top: -2.953em; left: 0.546em;"><span class="mo" id="MathJax-Span-43" style="font-size: 70.7%; font-family: STIXVariants;">′</span><span style="display: inline-block; width: 0pt; height: 2.59em;"></span></span></span></span></span><span style="display: inline-block; width: 0pt; height: 2.59em;"></span></span></span><span style="border-left: 0em solid; display: inline-block; overflow: hidden; width: 0pt; height: 1.202em; vertical-align: -0.257em;"></span></span>
Here is what the nLab is doing to produce the same:
<math class='maruku-mathml' display='inline' xmlns='http://www.w3.org/1998/Math/MathML'><mi>f</mi><mo stretchy='false'>(</mo><mi>a</mi><mo stretchy='false'>)</mo><mo>=</mo><mi>a</mi><mo>′</mo></math>
… and people complain that MathML is too verbose! Moreover, I can easily style the nLab’s MathML in my browser. There are extensions that let me add CSS rules to change how it looks. I don’t know how those interact with MathJaX - I just did a search and couldn’t come up with anything definite. And that really matters. People who have visual impairments can use CSS to modify a page to make it more readable to them. Some find it easier to read text when it is a particular colour: client-side CSS can do this. Here’s an example: http://mathsnotes.math.ntnu.no/mathsnotes/show/homework+2011+8. This is a homework I set. I like to give hints on difficult questions, but to enable people to have a go without using the hints then I style the hint so that it only shows when the mouse hovers over it. This is done by pure CSS - no javascript. With Firefox, this works perfectly. With browsers using MathJaX then the maths is displayed whatever.
I am not saying that MathML is perfect, nor that its rendering in Firefox is perfect. It’s a chicken-and-egg situation: the more people use MathML, the more effort the Firefox team will put into getting it right, but until it looks perfect, people are reluctant to use it.
I’m also not trying to do down MathJaX. I think that MathJaX is the bridge that will get us from where we are now (really difficult to get maths on the web) to where we want to be (native rendering in the browser). But I don’t think that it is the destination.
@Andrews #86 This is just a quick comment at some of the technical details of MathJax that Andrew mentioned. Disclaimer: a) I haven’t had the time to join the discussion recently and b) I don’t want to get into a shouting match with Andrew.
Let me stress that I think you’re absolutely right. MathML is the future of representing mathematics on the web. MathJax is here to help this development, not hinder it.
Peter, thanks for joining the discussion. My impression is that you know far more about MathJaX than I do (I have a vague feeling that you are actually part of the team). Certainly with regard to:
I don’t want to get into a shouting match with Andrew.
Nor do I! And I don’t think that this is going to happen. It is true that I used to be very anti-MathJaX, but now I realise that, as you say,
MathJax is here to help [MathML’s] development, not hinder it.
Indeed, I read today the article in the Notices of the AMS about MathJaX. I may well have been reading into it what I wanted to hear, but it did seem that Davide has this goal.
But to realise that goal, it has to be that wherever possible a site using MathJaX will find it easy to switch to MathML (or to allow users to switch to MathML) with the full benefits of MathML - part of which is to be able to not use MathJaX. Therefore so long as it is possible a site should serve its maths as MathML and use MathJaX to render it if needed, rather than serving its maths as some-variant-of-TeX.
My fear is that if MathJaX is seen as the be-all and end-all of putting maths on the web, then people will render their pages with the maths still in TeX, which means that for those of us who prefer MathML we get absolutely no benefit whatsoever.
@Andrew, @Kevin,
strange, I thought the biggest differences were actually in Firefox in Kevin’s examples. The hat over the sigma on nLab is pretty unusual, and the CP is rendered quite poorly as well. Looking at these outputs, I’d agree with Kevin that the poor rendering in nLab’s system currently outweighs its advantages, and would have to recommend just use MathJax rendering TeX in the front end for now.
I’m hesitant at reopening the MathML/MathJaX issue … but if it truly is a beauty contest and there are no other considerations than merely what it looks like (I hope my tone conveys what I think of that!) then using the nLab as the representative of MathML is not fair on MathML as the amount of effort that I’ve gone into to get it to “look nice” is … nothing (I know that Jacques Distler - the maintainer of Instiki, the underlying software - does put in a good deal of work, but I don’t know how high “looks nice” is on his list and his definition of “looks nice” might not be the same as other peoples (which, as I’ve said, is a point in favour of MathML as it is easily styleable)).
The problem with statements like:
would have to recommend just use MathJax rendering TeX in the front end for now.
Is that the “for now” implies that it would be possible to switch later. It won’t, without some huge effort to translate between syntaxes. It’s also incorrect in that it implies that MathJaX renders TeX. It doesn’t. It renders a distinct subset thereof and it always will. Implementing full TeX in javascript would be a nightmare and I hope that no-one ever tries.
I’ve also just realised that there’s a misleading statement in Kevin’s post with the screenshots:
My main point is that the best system is the one which works best on widely deployed web browsers in their default configurations.
My point is that the server has a lot more to do than is being taken into account here. None of the screenshots is really representing anything in its default state. If you really want the “default state” then you need to compare:
blah $f(a) = a'$ more blah blah $\hat{\Sigma}_\infty$ blah $\mathbb{C} P^\infty$
with
blah $f(a) = a’$ more blah blah $\hat{\Sigma}_\infty$ blah $\mathbb{C} P^\infty$
since those are what the browsers actually get.
In all of them, the web server has sent instructions on how to interpret the raw document. For the MathML, it sends some CSS rules. It could send different ones. For the MathJaX, it sends a heck of a lot of javascript. It could send different stuff. So there is a lot of scope for changing the output without requiring the user to do anything. That’s why the “beauty contest” is the wrong way to decide what to do^{1}. The other issues really do outweigh this and there, server-side conversion to MathML beats everything else, hands down.
And even if it does come down to a beauty contest, I’d like to see some more rigorous standard than “looks nice”. We get a some of questions on tex.stackexchange.com which basically say “How do I get my webpage to look like my TeX documents?” and the answer is “You can’t, and you don’t want to”. Reading a document on a screen is very different to reading one on paper and there are a whole different set of typographical considerations - such as using less fancy fonts: the $\mathbb{C}$ in the MO samples is all very nicely seriffed, which lessens its readability on the screen. ↩
For whatever it’s worth, I agree with Andrew about MathML. We should be looking forwards, not just picking whatever is easier and looks a little bit better now. Isn’t that what this whole thing is about, on a different plane? Publishing in an established and high-prestige Elsevier journal is easier and looks a little bit better on our CVs now; putting together a better solution for the future takes some more work and a bit of sacrifice in the short term.
One thing which bothers me about MathJaX is that the rendering of math happens after the rest of the page is rendered. The page comes up with the pseudo-TeX code displayed (maybe grayed out), then the javascript goes through and fills it in with math, with a progress display at the bottom of the screen. On a page with a lot of math, and a slow computer or a slow connection, this can take a long time. Moreover, not infrequently the math is larger than the pseudo-TeX code that produced it, which causes the page to continually “scroll” itself downwards as the math is processed. This constant motion can make the page essentially unreadable while I wait for the math to finish loading. Furthermore, if I followed a link to an anchor in the middle of a page, then after all the math is displayed (even if it happened near-instantaneously) I am no longer looking at that anchor but have been scrolled away from it, sometimes very far away from it, by the changing heights of math objects.
This is why I always only read the nLab in Firefox rather than Chrome, even though I personally do not really care about minor differences in appearance like those shown by Kevin.
Just a note that it is actually working and being used : https://selectedpapers.net/
I hope this enterprise takes off. One potential problem is that few comments will be attracted by almost all papers. One reason for this may be that many mathematicians no longer work in areas where a single paper is an important unit. The older Fields-medally areas often have a high barrier for new papers, and so a new paper may be interesting. Most of combinatorics and theoretical CS allows for incremental work where several papers are needed before they are worth commenting on in detail. Can SPN deal with this kind of commenting?
Of course, there are papers in all areas that are not worthy of much comment. But my feeling has been over the last couple of years that active internet discussion of mathematics is rather skewed towards “big results”. It isn’t actually clear that such a focus is optimal for overall scientific advancement.