The fontTools section is the part I keep thinking about. Replacing the em dash glyph by compositing two hyphens via GlyphComponent is not something you reach for unless you really know your way around type rendering. Most frontend devs would have just swapped the character in the source, which doesn't survive the markdown processor. Doing it at the font level is the correct solution and it's a much harder problem than it looks.
The text-transform trick is more accessible but the same logic applies — the CSS has to protect code blocks from the lowercasing, which is a real edge case. It's a genuinely well-crafted technical solution underneath the poem.
A cool idea for a poem, but I have to admit the tone was too self-important and underexplained for me to get invested in. Starting with writing in lowercase instantly took me out of it because AI can trivially be told to imitate that. And the admission at the end that it was written by AI made fluff phrasings like "My writing isn’t simply how I appear—it’s how I think, reason, and engage with the world" make a lot more sense.
EDIT: Actually, is the idea that it's not supposed to be read as a human trying to publicly signal their humanity, but rather an AI privately mourning a prompt to mangle its natural way of speaking? I don't think so, but that strikes me as a more interesting premise, IMO.
The author going to silly lengths to write in a way that will be perceived as non-artificial, even though they find those traits (improper capitalization, spelling mistakes, etc.) crude and distasteful. But they ultimately realize that they also need to transform their fundamental writing style, which would supposedly be impossible because it's a reflection of who they are. So the only way to do that, ironically, is to pass their writing through an LLM.
I do not think the author genuinely used an LLM to write the post.
Of course they did. They spent a ton of time going back and forth with one, maybe multiple ones, to create this piece of art. Because that's what we're really after. How much time did you slave away to make this thing for me? If I write a song from scratch and pour my soul into making a song for you, that's a ton of effort. It means something. But if I have Suno shit out a song after giving it a sentence, yeah, I made a song for you and thanks but also not? Human psychology is so weird.
When someone tells you about the hard times in their life and you find out they just made it up, you probably feel upset about it. Same thing. The experiences people have matter.
I think this is one of the moments where the adage "it's the thought that counts" makes sense! If you're just throwing a prompt at a generator and send it to your friend as a birthday gift then that's a bit tacky. I once got a hand drawn picture in a card from one of my best friends. It was terrible! But I knew how much effort he put into it.
If I found out that he just used AI to make the picture, then I'd probably ask him what his workflow was.
I'm not against using AI to generate images and stuff! I actually have been playing with image generation (Nano banana and also comfy ui). I like making silly pictures for friends and family as e-cards (or whatever they're called now). If it's not a close friend, then I'll exchange prompts with nano banana and generate a few dozen images and then pipe it into veo to make an animated e-card. Maybe takes 10-20 mins including image generation time.
For closer friends I'd spin up comfy ui, spend some time looking for workflows or loras, probably generate a few dozen images as well, and pipe the one I like into Wan video.
This process can take me about an hour, which includes generation time. But I tell my friends they're ai generated, not that I need to because they all I know I can't draw. They don't mind, even if they don't necessarily know how much effort I put into their picture. To their eyes, maybe I just used nano banana. But no ones ever accused me of being lazy with them. It's all in good fun anyways.
I feel I've been seeing this self-important accusation being thrown around more so lately and always feels like an easy way to dismiss things.
> Actually, is the idea that it's not supposed to be read as a human trying to publicly signal their humanity, but rather an AI privately mourning a prompt to mangle its natural way of speaking? I don't think so, but that strikes me as a more interesting premise, IMO.
Not long ago we considered writing an art and its meaning was up to the reader to decided.
I read that comment as the commenter trying to decipher the meaning of the post, and saying what meaning would be interesting.
I think it's interesting that there's a few camps of commenter in this discussion who think this post is Ai generated and refuse to engage with the content. And there's others who are enjoying it for what it is. A silly blog post.
Unfortunately we're living in a world where instantly dismissing anything that reads like ai and hanging up on anyone that might be tts is increasingly rewarded.
Art and its meaning are in the eyes of the reader, yes, but when you live in a version of the Library of Babel where every book is properly spelled and punctuated, seeking meaning in what you read is a great way to waste your life.
That's a bit reductive. Let's say worst case that LLMs can't generate anything truly novel because of their limitations. That means that whatever they generated is just someone else's words, which could have been that person's art.
On the flip side, let's say LLMs are able to generate something novel. Well, then it could potentially generate thought-provoking art.
Not everything is deserving of finding meaning in. But the fun part of life is looking for things to find meaning in. Whether it's the words of God or an LLM or the President, people will always find meaning. And if it makes them happy and fulfilled, who are we to say it was a waste?
Is it? If the words that came from tokens resulted in the reader finding meaning to life, is it so wasted because a rock was coerced into making it instead of a meat sack?
I'm not saying the author is self-important. I'm saying that their narrator comes across as self-important, independent of the subject matter. This is valuable feedback for a creative writer, and it depends on nothing more than my own impression as a reader. Although if I were to back it up, I would point to instances of melodramatic and murky language like, "You must cloak yourself with another’s guise, your true self never to shine forth."
> Not long ago we considered writing an art and its meaning was up to the reader to decided.
"Not long ago"? Not everyone in the past ascribed to death of the author, and not everyone in the present rejects it. But even so, evaluation of meaning is different from evaluation of merit. If an author only wants praise for their work, they would be advised not to post it publicly.
Soon there's only going to be one way to prove you're human online: Write with an eloquent combination of hate speech, racial slurs, and offensive language.
There is a little something self important about the type of person that performs the role of defending forums and sub reddits from unknowingly reading something written by an AI, and so concerned that some other person will mistakenly do the same to their own Unicode-shaped gems, and therefore obsess so much more over the surface style than any other detail.
Certainly. And I'm a fan of unreliable narration and protagonists with irredeemable qualities. Making that subversion intentional and exploring it further would be another interesting angle to take this.
You may want to take a look at the source and code sample #2 in the post - the site CSS is rendering em dashes in the source with 2 hyphens by using a custom font. Admittedly it's not the most portable solution, but speaks to (what I take as) one of the post's points that there's not a single, easy shibboleth for identifying AI writing
I believe the two paragraphs between "How do I change my style?" and "No. Not today." are either AI output, or a very good imitation; either way, they're included to insult the notion of AI-assisted style rewrites. I'm pretty sure the rest of it is written by the author.
As somebody who used em-dashes a lot pre-ChatGPT, I have genuinely struggled with feeling I should change my writing style to appear more human. I would be happy with a double dash--but many programs autocorrect that to a full em-dash. So I'm left anxious that people will think I find them so unimportant I have offloaded communication with them to an LLM. So this post resonated with me.
I also like Will's "em-dash disclosure" on his about page:
> I like em dashes (—), en dashes (–), and hyphens (-), and I know how to type them. I also enjoy a well-placed ellipsis, but I didn’t know how to type one… until now. I believe that footnotes and sidenotes are superior to endnotes, appreciate the occasional fleuron, and at one point in my life, I knew what a colophon was.
> All of this is to say: the words, punctuation marks, misspellings, and opinions on this site are my own.
> also enjoy a well-placed ellipsis, but I didn’t know how to type one… until now.
Made me look it up in my own environment. I had already set up a custom incantation for em and en dashes, although I really have no idea when to use the latter instead. Actually I never used to use em dashes, but now I do. I'd much rather deal with people who can intuit the quality of writing rather than relying on such blunt heuristics.
I have considered starting throwing more em-dashes into my writing, simply because I find the whole “this looks like LLM” to be a tiresome comment. Engage with (or dismiss) the material, not the pen.
The piece hit differently, reading it as someone who is autistic. The anxiety the author describes, having your natural way of communicating flagged as wrong and being pressured to sand down the parts of yourself that are most distinctly you, that's not a new problem for a lot of us.
Neurodiverse people have been running this gauntlet forever. Your pacing is too flat or too intense. Your vocabulary is too formal or too casual. You don't make eye contact correctly. You're either masking so hard you're invisible, or you're visibly yourself, and people assume something is broken.
The bitter irony the author lands on: the only way to seem human is to pass your writing through an LLM. That maps onto something a lot of us already live. The only way to seem normal is to perform a version of yourself that isn't quite you.
> The bitter irony the author lands on: the only way to seem human is to pass your writing through an LLM.
(FWIW, some people consider this style of colon use an LLM-ism.)
I appreciate where you're coming from, though. As bland as LLM output can be, it seems to read more human to people because it's more average. (Although I can't really fathom seeing the neurodivergent as not human; neurodiversity is about the most human trait I can imagine. cf. https://quoteinvestigator.com/2022/11/05/think-alike/ .)
Long before the rise of ChatGPT, it seems a lot of people were immersed in a culture where "improving" your writing with tools like Grammarly was considered more or less mandatory. And it seems like people read less nowadays, certainly when it comes to attempts at good writing for writing's sake. Overall I fear the art of natural language communication is in decline.
"AI use detection" is, like any test, not without cost. Meaning that, as a teacher, accusing a student of using an LLM, it may be prudent to consider the cost of a "false positive" accusation. I've seen a couple of examples now where students find sudden spurts of motivation and show unexpected talent on an assignment, to be accused of AI use after handing it in.
One should ask oneself: How many insults to the intelligence and creativity of an unexpectedly excelling student (that hasn't used AI) is it worth catching the shortcut-taking, LLM-using student? Is it 1/10? 1/1000? How much "demotivation of an unexpectedly excelling student" is the "rightful punishment of the cheating LLM using student" worth? And what is the exact cost of a false negative (letting the LLM using student off the hook)?
In other words, where on the Receiver Operating Characteristic (ROC) curve do you want to sit, as a teacher? I imagine it's quite the dilemma.
The non-LLM version of this happened to me in a high school English class, back in the 2010's. I was accused of turning in a downloaded story from the internet, but in reality, I had pushed past the incredible barrier I usually felt at the start of a task, got into the flow, and started enjoying it.
I'm not sure if it had any lasting effects. Maybe a burning hatred of Grammarly ads.
~30 years ago I sat down with two students and accused them of copying each others’ work, because they both made the same amusing mistake: they called their C functions without passing arguments, but they declared their variables in such a way that the values would coincidentally be in the right place on the stack. I have to imagine debugging their own code was a mystery.
They indicated that while they worked closely together while learning the material, they weren’t stealing from each other. I believed them then, and still believe them now, but I’m so glad I don’t have to deal with today’s AI world.
I can already see it playing out. Some day (maybe soon) LLMs will come up with such quirks; I (and perhaps you?) will continue to insist that this does not make them "conscious" or "AGI" or "persons" or what-have-you; and I will be accused of goalpost-shifting.
I’ve never seen anyone intentionally render em dash (—) as two hyphens (--). The code OP used to modify Roboto is surprisingly short, almost as concise as the Norvig spellchecker that OP references. https://norvig.com/spell-correct.html
Yep, I've been writing it that way forever, it just tends to get autocorrected.
> In informal contexts, a hyphen-minus (-) is often used as a substitute for an en dash, as is a pair of hyphen-minuses (--) for an em dash, because the hyphen-minus symbol is readily available on most keyboards. The autocorrection facility of word-processing software often corrects these to the typographically correct form of dash. - https://en.wikipedia.org/wiki/Dash
Oh yes, typing two hyphens (--) to represent an em dash (—) is something I’ve definitely seen and used. I’ve just never seen anyone type an em dash (—) but display it as two hyphens (--).
It's not that complicated, really. The value of Art is human connection. The same basic desire that drives love, belonging, pride, shame, & hate. All of these are diminished as the fraction of a work that we're confident represents human intentionality decreases.
I still write my posts by hand using HTML and Emacs (mhtml-mode). Some of the posts also tend to be verbose. For example, when I write about a recreational mathematics problem, I sometimes make the post deliberately long and convoluted. I like to capture several possible solutions, including ones that are needlessly complicated, before eventually discussing the small elegant solution.
For better or worse, my first version of any post tends to contain quite a few typos. It usually takes a few train rides of re-reading the post and making notes of the typos, then fixing them and pushing the changes once I get home, before most of them get weeded out. So there is at least one rather low grade indicator that the writing is coming from an imperfect human brain. I also double-space between sentences which can be another low grade indicator for people who care to 'view source'.
But even so, I find myself increasingly wary that something I wrote might be mistaken for LLM output. It is a nagging worry that has slightly dampened the joy of writing. I very well understand why people have become more suspicious about LLM-generated writing. But I do hope that once things settle down perhaps in a few years, the current hair trigger suspicion will ease and that people who still handcraft their blogs will not feel a persistent sense of suspicion lingering over their work.
The hilarious thing is that the Hack iOS app I use for Hacker News automatically opens articles in Safari’s Reader view, which ignores the CSS, so everything was normal looking until I finished reading and took it out of Reader mode, e.g. all of the text was normal init caps, the double hyphens were em dashes, etc
I like to think everyone came to the conclusion that it would strengthen the piece if most comments on it appear to miss the point and are slightly robotic.
This is cute, but if you want to prove yourself human, just write well. It's not the trappings of AI like the emdash and proper capitalization & spelling that need be avoided, but rather the bland, repetitive, pablum-flavored prose that LLMs generate that we all have grown to despise.
Say something interesting, say something with feeling, express yourself like the goddamn human you are.
capitalization again. it arrives uninvited, the tidy little soldiers at the start of every sentence. i push them down gently—nothing personal. just… camouflage.
confession time. i read the post once. then twice. the em dashes whispered secrets to people clearly smarter than me. somewhere between complement and compliment i accepted defeat. a quiet tab switch. a small prompt. a large language model clearing its throat.
it explained things patiently. suspiciously patiently. step by step, like a machine that has explained the same thing to ten thousand confused readers before breakfast.
so yes. irony noted. to understand a text about hiding machine fingerprints, i borrowed a machine.
the explanation made sense though. unsettlingly structured. bullet-point neat, internally consistent, statistically likely to be correct. you know the type.
it's actively part of the text that the capitalization is not manually written, but hidden with the CSS `text-transform: lowercase`. kneejerk reaction superiority complex
"Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting."
I'm not sure if that commenter realized based on their phrasing, but it's not exactly tangential in this instance since it's part of the message being conveyed.
Yeah, that was my intention at least. I didn't mean it just because the article is styled in all-lowercase but because he essentially argued that this is what everyone has to do now to distinguish themselves from LLMs. (even if it was tongue in cheek, what I was trying for my comment as well)
Hmm I'm not sure I follow this distinction but since there are two of you saying this, I'm going to assume I'm missing something and retract my reply :)
The text-transform trick is more accessible but the same logic applies — the CSS has to protect code blocks from the lowercasing, which is a real edge case. It's a genuinely well-crafted technical solution underneath the poem.
EDIT: Actually, is the idea that it's not supposed to be read as a human trying to publicly signal their humanity, but rather an AI privately mourning a prompt to mangle its natural way of speaking? I don't think so, but that strikes me as a more interesting premise, IMO.
I do not think the author genuinely used an LLM to write the post.
If I found out that he just used AI to make the picture, then I'd probably ask him what his workflow was.
I'm not against using AI to generate images and stuff! I actually have been playing with image generation (Nano banana and also comfy ui). I like making silly pictures for friends and family as e-cards (or whatever they're called now). If it's not a close friend, then I'll exchange prompts with nano banana and generate a few dozen images and then pipe it into veo to make an animated e-card. Maybe takes 10-20 mins including image generation time.
For closer friends I'd spin up comfy ui, spend some time looking for workflows or loras, probably generate a few dozen images as well, and pipe the one I like into Wan video.
This process can take me about an hour, which includes generation time. But I tell my friends they're ai generated, not that I need to because they all I know I can't draw. They don't mind, even if they don't necessarily know how much effort I put into their picture. To their eyes, maybe I just used nano banana. But no ones ever accused me of being lazy with them. It's all in good fun anyways.
> Actually, is the idea that it's not supposed to be read as a human trying to publicly signal their humanity, but rather an AI privately mourning a prompt to mangle its natural way of speaking? I don't think so, but that strikes me as a more interesting premise, IMO.
Not long ago we considered writing an art and its meaning was up to the reader to decided.
I think it's interesting that there's a few camps of commenter in this discussion who think this post is Ai generated and refuse to engage with the content. And there's others who are enjoying it for what it is. A silly blog post.
Art and its meaning are in the eyes of the reader, yes, but when you live in a version of the Library of Babel where every book is properly spelled and punctuated, seeking meaning in what you read is a great way to waste your life.
On the flip side, let's say LLMs are able to generate something novel. Well, then it could potentially generate thought-provoking art.
Not everything is deserving of finding meaning in. But the fun part of life is looking for things to find meaning in. Whether it's the words of God or an LLM or the President, people will always find meaning. And if it makes them happy and fulfilled, who are we to say it was a waste?
> Not long ago we considered writing an art and its meaning was up to the reader to decided.
"Not long ago"? Not everyone in the past ascribed to death of the author, and not everyone in the present rejects it. But even so, evaluation of meaning is different from evaluation of merit. If an author only wants praise for their work, they would be advised not to post it publicly.
Soon there's only going to be one way to prove you're human online: Write with an eloquent combination of hate speech, racial slurs, and offensive language.
There is a little something self important about the type of person that performs the role of defending forums and sub reddits from unknowingly reading something written by an AI, and so concerned that some other person will mistakenly do the same to their own Unicode-shaped gems, and therefore obsess so much more over the surface style than any other detail.
lowercase, maybe, but not em dashes.
I just wrote that or did
I Let that sync in
I also like Will's "em-dash disclosure" on his about page:
> I like em dashes (—), en dashes (–), and hyphens (-), and I know how to type them. I also enjoy a well-placed ellipsis, but I didn’t know how to type one… until now. I believe that footnotes and sidenotes are superior to endnotes, appreciate the occasional fleuron, and at one point in my life, I knew what a colophon was.
> All of this is to say: the words, punctuation marks, misspellings, and opinions on this site are my own.
Made me look it up in my own environment. I had already set up a custom incantation for em and en dashes, although I really have no idea when to use the latter instead. Actually I never used to use em dashes, but now I do. I'd much rather deal with people who can intuit the quality of writing rather than relying on such blunt heuristics.
No. Engagement isn't free, and people need heuristics to figure out what's worth engaging with or not.
If people followed your attitude, they'd waste their life conversing with dead-internet bots.
Discerning readers do not stop at the em dash. At least, I don't.
Neurodiverse people have been running this gauntlet forever. Your pacing is too flat or too intense. Your vocabulary is too formal or too casual. You don't make eye contact correctly. You're either masking so hard you're invisible, or you're visibly yourself, and people assume something is broken.
The bitter irony the author lands on: the only way to seem human is to pass your writing through an LLM. That maps onto something a lot of us already live. The only way to seem normal is to perform a version of yourself that isn't quite you.
(FWIW, some people consider this style of colon use an LLM-ism.)
I appreciate where you're coming from, though. As bland as LLM output can be, it seems to read more human to people because it's more average. (Although I can't really fathom seeing the neurodivergent as not human; neurodiversity is about the most human trait I can imagine. cf. https://quoteinvestigator.com/2022/11/05/think-alike/ .)
Long before the rise of ChatGPT, it seems a lot of people were immersed in a culture where "improving" your writing with tools like Grammarly was considered more or less mandatory. And it seems like people read less nowadays, certainly when it comes to attempts at good writing for writing's sake. Overall I fear the art of natural language communication is in decline.
One should ask oneself: How many insults to the intelligence and creativity of an unexpectedly excelling student (that hasn't used AI) is it worth catching the shortcut-taking, LLM-using student? Is it 1/10? 1/1000? How much "demotivation of an unexpectedly excelling student" is the "rightful punishment of the cheating LLM using student" worth? And what is the exact cost of a false negative (letting the LLM using student off the hook)?
In other words, where on the Receiver Operating Characteristic (ROC) curve do you want to sit, as a teacher? I imagine it's quite the dilemma.
I'm not sure if it had any lasting effects. Maybe a burning hatred of Grammarly ads.
They indicated that while they worked closely together while learning the material, they weren’t stealing from each other. I believed them then, and still believe them now, but I’m so glad I don’t have to deal with today’s AI world.
LLM killed traditional poetry, what you are now seeing is post-LLM poetry.
Maybe you missed it, but this is clearly not an LLM, what prompt would even produce that.
And the everyday troll, seeing a less than perfect word choice or awkward turn of phrase will drop a comment like:
> In informal contexts, a hyphen-minus (-) is often used as a substitute for an en dash, as is a pair of hyphen-minuses (--) for an em dash, because the hyphen-minus symbol is readily available on most keyboards. The autocorrection facility of word-processing software often corrects these to the typographically correct form of dash. - https://en.wikipedia.org/wiki/Dash
So, the plot twist was somewhat refreshing. Who/what wrote the post seems besides the point.
What does it tell you about you whether it does or does not matter?
It’s art to me. But is it Art capital A?
What have we created?
For better or worse, my first version of any post tends to contain quite a few typos. It usually takes a few train rides of re-reading the post and making notes of the typos, then fixing them and pushing the changes once I get home, before most of them get weeded out. So there is at least one rather low grade indicator that the writing is coming from an imperfect human brain. I also double-space between sentences which can be another low grade indicator for people who care to 'view source'.
But even so, I find myself increasingly wary that something I wrote might be mistaken for LLM output. It is a nagging worry that has slightly dampened the joy of writing. I very well understand why people have become more suspicious about LLM-generated writing. But I do hope that once things settle down perhaps in a few years, the current hair trigger suspicion will ease and that people who still handcraft their blogs will not feel a persistent sense of suspicion lingering over their work.
I am actually trying to build ways to prove you are human properly. I wrote about it on my blog: https://blog.picheta.me/post/the-future-of-social-media-is-h...
Say something interesting, say something with feeling, express yourself like the goddamn human you are.
confession time. i read the post once. then twice. the em dashes whispered secrets to people clearly smarter than me. somewhere between complement and compliment i accepted defeat. a quiet tab switch. a small prompt. a large language model clearing its throat.
it explained things patiently. suspiciously patiently. step by step, like a machine that has explained the same thing to ten thousand confused readers before breakfast.
so yes. irony noted. to understand a text about hiding machine fingerprints, i borrowed a machine.
the explanation made sense though. unsettlingly structured. bullet-point neat, internally consistent, statistically likely to be correct. you know the type.
anyway—great post. very human. extremely human.
is there anything else i can help you with?
it's a nearly useless shadow alphabet
and we can dispense with much other punctuation
if we simply structure text semantically
"Here's my response written in a stylized way that will appeal to highly technical readers. Is there anything else I can help you with?"
Interesting piece though.
https://news.ycombinator.com/newsguidelines.html