WCSTombs

This was a long rant that I pretty much agree with, except at the conclusion I still don't know what the author is trying to propose. What is this supposed new "WYGIWYM" paradigm?

Anyway, every recent time I've had to use a word processor, what I wished I could have done instead is write the whole thing simply in HTML or Markdown, and have a separate style sheet to choose all the fonts, font sizes, layout, etc. I pretty strongly feel this type of workflow should completely replace document editing, and should have done long ago. I know a lot of people prefer a more WYSIWYG style for writing, but that's fine: you can provide a word-processor-like interface, but instead of manually styling the text piece by piece, you assign from predefined styles, which of course you also have control over, but separately from the content. I think Libreoffice allows you work somewhat like this if you know what you're doing, but it's pretty clunky and certainly not obvious, and I haven't used it enough to say for sure.

By the way, IMO there's nothing wrong with obsessing over the minutiae of how content looks. However, I would never want to prevent a reader from changing literally anything about my site's styling to suit their own needs or desires. My carefully chosen styling is meant for the 99% of cases where people just load the site in a stock browser, and for the other 1% who choose to override the styles, it should also "just work."

That Lean Rada article is really fun!

I can't help but agree, and sometimes I think about the alternate version of the web where XML realized its full potential as the dominant technology. The obvious problem with XML is that the syntax is so verbose and redundant that it's only barely human-writable, but this is 100% a solvable problem.

Yeah, XSLT 1.0 has some really obvious missteps, so much so that to my knowledge basically nobody just implements XSLT 1.0. What they actually support is (most of) XSLT 1.0 plus a handful of EXSLT extensions (which are completely unofficial and not part of any XSLT spec, by the way). In my opinion the EXSLT stuff is the bare minimum needed to make XSLT 1.0 at least a passable language. Meanwhile, the XSLT spec evolved into 2.0 and 3.0, which as far as I'm aware, even to this day, has no complete open-source implementation.

I've read a few articles by Chris Lee, and I normally like them quite a bit. However, IMO there's something missing from this one, like how does this experiment push the frontier of quantum entanglement research? Is this the most complex entangled state ever created in a lab, or something like that?

Pretty poor reporting here, but let me see if I can get this straight: Georgia ballots weren't truly anonymous, but contained enough identifying information (through these ID numbers) to be correlated to individual voters.

Here's the most important thing:

> Georgia’s outgoing secretary of state, Brad Raffensperger, locked down the tabulation data that someone might use to identify a voter, ordering any public release to redact the ID numbers after an election, which “substantially address[es] this flaw”, said Ben Adida, an MIT-trained cryptographer and the founder of the election technology non-profit VotingWorks, as he addressed the board on Thursday.

But if the unredacted data still exists anywhere, it's still a massive liability. It can still be leaked by unscrupulous people who have access to it, or disclosed to other unscrupulous actors via coercion. The flaw hasn't been fully addressed until the data is expunged and the system is updated to avoid creating it in the first place.

By the way, if I'm reading between the lines correctly...it appears the flaw has nothing to do with "AI," which is why I've called this "poor reporting." The academic who discovered the exploit either used an LLM in their research, or simply demonstrated that AI models have enough knowledge to give out the steps to do the exploit. If the LLM is somehow integral here, that is very poorly explained. My guess is either the author or editor decided to make the story about AI, to ride the current wave of "dangerous AI" hysteria, and then stuck it into the story everywhere they could.

This is a cool concept and something I was looking into recently!

IMHO, XML as a concept is a bit underrated and misunderstood nowadays. Unfortunately the way the culture of XML has evolved over the past decades probably has a lot to do with that, but that shouldn't mean we have to throw out the baby and the bathwater.

For one small example of what I mean, Jupyter notebooks are JSON, while it certainly seems to me that XML would be a better fit, minus any concerns about tooling. A cell's content is represented in the file as an array of strings, while in XML it would just be a text node with the literal contents. But more generally, I increasingly feel that XML should have become the lingua franca for abstract syntax trees of all types. Hopefully it can still become that, and Invisible XML seems like a good step in that direction.

By the way, I should point out that the CommonMark AST has an official XML representation, but as far as I know that's the only Markdown dialect that does, and I also don't know of an ixml grammar. Wouldn't it be great if every Markdown dialect had an ixml grammar though? If Invisible XML has any limitations that prevent this from happening, then IMHO those really should be addressed, since if it can't represent Markdown, then it's probably just not very useful after all.

Yes, and I think that's more likely to be the issue with nytimes.com (from my completely unscientific investigation).

The difference between big-company and FLOSS UX design, including profit-motivated behaviors (a bit indirectly, but it's definitely there), is specifically addressed in the article, so I don't think that's a very fair characterization.

I actually agree about customization, but that's only part of it, and seems to be pretty orthogonal to what the article is talking about. You can only customize to features that are actually implemented by the system, and coming up with those seems like the harder problem.

> The final, particularly strong, objection from the tech community is "don't touch my stuff". People get angry when they have worked hard to get their desktop environment just as they like it and something changes to disrupt that.

Yeah, this can be a little extreme, and I feel we can benefit from being a little more open to trying new things.

The part about designing specifically for large displays and widescreen monitors was interesting. We can also reverse it, though: right now I'm away from home, and I only have this old laptop which is mostly great but has a shitty touchpad as the mouse, and I have to think this must be really common. However, this situation seems to be completely ignored by all desktop UX design. We seem to be pretty good at coming up with ways to use the mouse in our programs, but when the mouse is hard to control and unreliable, it causes pain. I guess I can use ratpoison or something, but that seems to be giving up too much.

Regarding the AI going deep into the filesystem and organizing my data...I'm extremely skeptical of inserting any kind of inherent unpredictability into how I interact with the computer. Giving a command or input should produce a well-defined and predictable (in principle) action. Getting rid of that paradigm fundamentally changes the relationship between the user and the computer in a way that, in my opinion at least, greatly diminishes its usefulness as a tool.

> Capable people with decades of experience can successfully automate away a lot of tedious work, and still end up with a final product identical to what they would have done by hand.

Come on, I don't think the last part is remotely true. It's certainly an extraordinary claim that would need to be backed by research. If it were really true that the final product was identical to what the developer would have written by hand, that alone would solve most of the deeply problematic ethical issues that seem inherent to LLMs, but I think it's bullshit, and moreover there's no way to accomplish what's claimed (except for simple autocompletion, which is a very special case).

There's also this:

> The license of source code they produce is legally unclear.

If you really think that, how is it not an immediate blocker for LLM-generated contributions? Aren't you putting the whole project in legal jeopardy?

I completely agree. When the docs are in the same repository as the source code, at least you can keep the documentation atomically synced to the code. No other method allows this (not counting setups with Git submodules, which is the same thing with more steps), and that should be an immediate dealbreaker.

Docs being updated to reflect the code can then (and must, IMO) be a blocking constraint on all pull requests.

> Or is it because they then publish those docs somewhere else?

Exactly, I use Sphinx to make a nice website from the markdown sources (although it doesn't have to be markdown). That includes generating a full API reference page from all the docstrings.

There was another article submitted today lamenting the slowness of the modern internet, using nytimes.com as an example. Comparing that to this site gives, IMHO, a big part of the answer: don't try to fix what isn't broken. (I also disagree about the UI being horrible.)

> U.S. download speeds increased 174% from 2020 to 2024. 5 Gbps is almost 90,000x the bandwidth I had in 1996. So why are we’re still dealing with slow, bloated websites?

Yes, because the web is more bloated now than it ever was, but also because you're conflating latency and throughput. If the server takes multiple seconds to send the first bit of data, no amount of bandwidth will ever get you within the 0.1-second deadline.

Conflating latency and throughput is a surprisingly common error, and at my last job processing sensor data in real time, I had to make a huge effort to convince other people that latency was the performance metric that mattered, much more than throughput.

In case you feel like branching out beyond the digital completely, I'll leave this here (no affiliation): https://inktober.com/rules

> and almost zero explanation as to how his approach was actually different from any other tech company's.

> Except the "free classes" part: those were cool, ...

Seems like you answered your own question there. As far as I'm aware that practice is pretty much unheard-of, and you're dismissing it pretty casually...

I don't understand the Ong quote:

> Most persons are surprised, and many distressed, to learn that essentially the same objections commonly urged today against computers were urged by Plato in the Phaedrus (274–7) and in the Seventh Letter against writing. Writing, Plato has Socrates say in the Phaedrus, is inhuman, pretending to establish outside the mind what in reality can be only in the mind. It is a thing, a manufactured product. The same of course is said of computers. Secondly, Plato's Socrates urges, writing destroys memory. Those who use writing will become forgetful, relying on an external resource for what they lack in internal resources. Writing weakens the mind.

Wasn't Plato writing this stuff down? Why would he argue against what he was literally doing at that moment?

I think I'm also not clear on who "Socrates" is supposed to be here and his relation to Plato. The article attributes the argument to the actual person, while Ong seems to consider him not only as one of Plato's characters, but as some sort of self-insert for Plato, to ascribe the argument to Plato himself.

You had me with that one. (I don't mean had...you know what I mean.)

> When you write something down and it is read thousands of years later, interpreters are doing a lot of work to try to figure out what was meant and the thought the author was trying to get across. I fully believe that there are entire, complex thought processes and ideas which were lost in translation due to contextual differences.

True, but I don't think leaving it unwritten makes it better.

`git add --resolved` is a wonderful idea, and definitely something I would start using.

It's very hit or miss. Abylight Studios (Super Hydorah, Cursed Castilla) seems to do them pretty well, but then I've seen some that left me wondering how they ever made it past QC. (I won't name them because it's been a while and I don't want to name the wrong game.)

I want to point out though that it's impossible for any CRT filter to have a peak brightness that's not significantly lower than without the filter, since it can darken pixels but not brighten them past what the maximum already is.

While I pretty much agree with what's being said here, and I certainly agree with the conclusion, I disagree with this point of view:

> It follows, Sartre argues, that the act of writing, as a means to convey a thought to a reader, may only be completed through the act of reading. Reading can only be done by others; writing, therefore, only exists to be read – it can serve no other purpose.

That seems like a pretty extreme view, and I find the act of writing can also be useful to the writer. But I think this actually supports the author's conclusion more than Sartre's claim.

That is not what I said at all. For research whose ostensible purpose is to increase human understanding, the absolutely bare minimum we could ask for is for the authors to understand their own work well enough to write their research paper themselves, without having an LLM crank it out. I'm not saying anything about LLMs at other points during the research. Unless I'm missing something, the TMLR policy does not forbid using LLMs to generate entire research papers. It only says the authors are "responsible for the content" and warns against "low-quality contributions."

This is the journal's policy on LLM use by authors [1]:

> LLMs may be used as general-purpose assistive tools. Whichever tools are used, authors are fully responsible for content on which they are listed as (co-) authors. This includes, but is not limited to, content generated by LLMs that could be construed as plagiarism or scientific misconduct (e.g., fabrication of facts). Low-quality contributions (be they submissions or reviews) that appear to be largely LLM-generated will be closely examined for evidence of the issues mentioned previously, such as scientific misconduct. LLMs are not eligible for authorship. We will periodically revise this policy as new information about the use of LLMs in the scientific process becomes available.

While it doesn't outright encourage using LLMs, it's right at the door, and IMO a policy this weak is actively contributing to the problem the article's author is complaining about. In my opinion any policy weaker than "using LLMs to generate any part of your submission is not allowed and considered a serious breach of ethics" is insane. People like to say that such policies are unenforceable, but that's really not the point (at first), since there are other things like (somewhat ironically) p-hacking that are pretty hard to detect but still widely recognized as unethical. We haven't exactly solved p-hacking either, but at least most of us can agree that p-hacking should be eliminated.

It's hard for me not to read between the lines here. Maybe it's the tinfoil talking, but it being a machine-learning journal, it probably embodies a generally pro-AI philosophy, and thus may not want to discourage too much of it...

It may also be worth noting that this journal apparently uses AI itself on the reviewing side [2]. I'm not claiming this is super unethical or anything as long as the main review is human (although I have concerns), it probably should be part of the conversation.

[1]: https://jmlr.org/tmlr/editorial-policies.html

[2]: https://medium.com/@TmlrOrg/ai-reviews-at-tmlr-for-assessing...

Apparently I'm a bit slow today, because my brain started completing the headline as "Diplodocus, Long Thought Extinct, ..." before my eyes could finish reading it.

> The animal’s fossilized remains — 14 well-preserved tail vertebrae and several chevron bones — were unearthed at the La Tejería site near the city of El Castellar in the Spanish province of Teruel.

According to Wikipedia (if I'm reading correctly), the chevron bones are also part of the tail [1]. It's pretty cool how such a small part of the skeleton can carry so much information!

[1]: https://en.wikipedia.org/wiki/Haemal_arch

Please just write in your own words, and don't filter your writing through AI. Doing the latter will make you a worse writer.

It's not unheard of, like when the South Korean government lost a bunch of its data in a data center fire. Don't have the only copy of your data in a data center, and you won't have to worry about such a thing happening to you.

It's a clumsily made point, but I think the gist is that the letter asks AI to align to a certain ideal vision of the field of math, that the math community itself is not well aligned to. I don't think the article makes this point well, though, just by listing these incidents. Don't get me wrong - each incident is important and worthy of attention in its own right, but I find the overall argument to be pretty weak and strongly reeking of whataboutism.

(2025).