This makes me think of the fad where people on youtube will hold a microphone up in frame, because it somehow connotes authenticity. I'm sure some people are already embracing a bit of sloppiness in their writing as a signal of humanity; I'm equally sure that future chatbots will learn to do the same.
- Customer: Excuse me, I'm looking for the Aunt Jemima maple syrup. Can you point me in the right direction?
- Employee: y u ask like chatbot
Got several comments saying they were "AI slop."
Even had a screen cap of my drawing process.
Kinda funny to think my drawings, which have likely "trained" AI image generators, are now getting accused of being AI.
Now you need a really big microphone, something that looks like it was built in 1952.
This applies not only work-stuff itself also to the job-applications/cv/resume and cover-letters.
Yes I enjoy lisp, how could you tell
<li> do this
<li> and this
and <img alt='this'> instead of <img ... />
You might like Lisp, but what you're saying reminds me of the late 00s/early 2010s xHTML2 vs. HTML5 debate :)
[0] :))
It's one of those things I think are worth putting some extra effort into, I'm glad to see at least one other person giving it some thought. Thx <3
I've also noticed an increase of this in myself and others, I used to edit a lot more before sending anything, but now it seems more authentic if you just hit send so it's more off the cuff with typos, broken sentences and all.
I'm sure an LLM could easily mimic this but it's not their default.
> I started making deliberate grammar and spelling mistakes in professional context[s]. Not like I have ~a~ perfect writing anyway, but at least I could prove that it was self-written, not an auto-generated slop. (Could be self-written slop though :)
> This applies not only [to] work-stuff itself also to the job-applications/cv/resume and cover-letters.
I conclude you are real.
My double-space-after-a-period though, I will keep that until the end. Even if it often doesn't even render in HTML output, I feel a nostalgic connection to my 1993 high school typing teacher's insistence that a sentence must be allowed to breathe.
If leaving out the Oxford comma here was an intentional joke I both commend and curse you!
• Like
• This
(option-8 on a Mac US keyboard layout). Now it looks like something only an LLM would do.
My phone lets me long-press the hyphen key to get an em-dash so sometimes I'll use it.
Probably the biggest tell that I'm not AI is that I'm probably not using it in the appropriate circumstances!
"respond like a twitter user", "pretend like we're texting", etc
> "respond like a twitter user", "pretend like we're texting", etc
+1 to it. I actually had given a response to the above parent comment itself using Kimi and I would've said that its (sort of) a good emulation fwiw.
(This above line itself was written by AI itself: https://www.kimi.com/share/19c96516-4032-8b73-8000-0000f45eb...)
I don't know if worse grammar could make a difference aside from removing false negatives (ie. nowadays people with good grammar are questioned if they are LLM's or not) but this itself doesn't mean that worse grammar itself means its written by a human. (This paragraph is written by me, a human, Hi :D)
Also adding better "context" into the discussion, than the usual claims/punchlines of marketing-speak.
Maybe it's not exactly the grammar itself but also overall structuring of the idea/thought into the process. The regular output sounds much more like marketing-piece or news-coverage than an individual anyway. I think, people wanna discuss things with people, not with a news-editor.
If I understand you correctly, then Yes I completely agree, but my worry is that this can also be "emulated" as shown by my comment by Models already available to us. My question is, technically there's nothing to stop new accounts from using say Kimi and to have a system prompt meant to not sound AI and I feel like it can be effective.
If that's the case, doesn't that raise the question of what we can detect as AI or not (which was my point), the grand parent comment suggests that they use intentionally bad human writing sometimes to not be detected as AI but what I am saying is that AI can do that thing too, so is intentionally bad writing itself a good indicator of being human?
And a bigger question is if bad writing isn't an indicator, then what is?
Or if there can even be an good indicator (if say the bot is cautious)? If there isn't, can we be sure if the comments we read are AI or not
Essentially the dead-internet-theory. I feel like most websites have bots but we know that they are bots and they still don't care but we are also in this misguided trust that if we see some comments which don't feel like obvious bots, then they must be humans.
My question is, what if that can be wrong? It feels to me definitely possible with current Tech/Models like say Kimi for example, Doesn't this lead to some big trust issues within the fabric of internet itself?
Personally, I don't feel like the whole website's AI but there are chances of some sneaky action happening at distance type of new accounts for sure which can be LLM's and we can be none the wiser.
All the same time that real accounts are gonna get questioned if they are LLM or not if they are new (my account is almost 2 years old fwiw and I got questioned by people esentially if this account is AI or not)
But what this does do however, is make people definitely lose a bit of trust between each other and definitely a little cautious towards each message that they read.
(This comment's a little too conspiratorial for my liking but I can't help but shake this feeling sometimes)
It just is all so weird for me sometimes, Idk but I guess that there's still an intuition between whose human and not and actually the HN link/article iteslf shows that most people who deploy AI on HN in newer accounts use standard models without much care which is the reason why em-dashes get detected and maybe are good detector for sometime/some-people and this could make the original OP's comment of intentionally having bad grammar to sound more human make sense too because em-dashes do have more probability of sounding AI than not :/
It's just this very weird situation and I am not sure how to explain where depending on from whatever situation you look at, you can be right.
You can try to hurt your grammar to sound more human and that would still be right
and you can try to be the way you are because you think that models can already have intentionally bad grammar too/capable of it and to have bad grammar isn't a benchmark itself for AI/not so you are gonna keep using good grammar and you are gonna be right too.
It's sort of like a paradox and I don't have any answers :/ Perhaps my suggestion right now feels to me to not overthink about it.
Because if both situations are right, then do whatever imo. Just be human yourself and then you can back down this statement with well truth that you are human even if you get called AI.
So I guess, TLDR: Speak good grammar or not intentionally, just write human and that's enough or that should be enough I guess.
I've noticed a habit of late of people accusing a comment of being LLM generated if they disagree with it. It was getting quite tiresome a few weeks ago but seems to have died down.
I suppose it is possible that they are actually LLMs making the accusations? :-)
(I'm one of those weirdos that try to use proper grammar and complete sentences in text messages and instant messages.)
I use em dashes, and I don't care whether or not someone assumes I'm an LLM. Typography exists for a reason.
>I put the em dash on modifier+dash
This is the default on Macs
I wonder how much crossover there would be between a trained text analysis model looking for Gen-X authors and another looking for LLM's.
But that's a different issue.
Entire sentence structures have been effectively blacklisted from use. It's repulsive.
Speaking of overusing something until it becomes cringe, has anyone shown their kids Firefly? Does it still hold up after the Joss Whedon signature bathos (and other tics) became a tentpole of the Marvel Cinematic Universe and created an abundance of cultural antibodies?
There were a few times we cringed a bit (with both shows) but overall stood the test of time. I didn't watch Buffy & Angel first time around, so it was a bit of a cultural moment I got caught up on. And it was nice to revisit Firefly, the little bit of it we got.
There is no such thing as blacklisted by other commenters.
You’d think ethically leaving it in would be better. But we’re talking about big tech companies here.
Well, to be fair Gen-z slangs also have a massive impact. My generation sometimes point blank said to me that they didn't have the attention span to read my sentence :/
Definitely picked up a few slangs along the way now. I had to somehow toggle a switch between how I write on HN/how I write with my friends the first few times and I write pretty informally in HN, but its that you got to be saying lowk bussin rizz 67 to make sense.
My friends who use insta literally had Abbreivations which were of 9 letter words in my own language that the insta community of my nation's gen-z sort of made.
Although I would agree that we haven't seen a whole unicode being thrown this way in ALL generations (I feel like universally everyone treats em-dashes as something written by AI or definitely get an AI alert)
But I think that 67 is something that atp maybe even most adults might have gotten exposed to which has probably changed the meaning of number.
I’m waiting for a Philip K. Dick bot to declare me non-human.
Am I the only one who in a Captcha test sometimes wants a different option for the “I am Human” check box? Ironically really since to prove we’re human we have to check the boxes with a crossing in them, no account to be made of people who call them zebra crossings.
Now I find myself deliberately making things worse to avoid being accused of not being human! Bah!
Tip: Patterns like “It’s not just X, it’s Y” are a more telltale sign of LLM slop. I assume they probably trained on too much marketing blurb at some point and now it’s stuck.
I dunno this en versus em dash stuff, I just use the minus sign on my keyboard.
> A new way, a better way.
The autumn winds blow.
I also like …
This is like ruining swastikas and loading rainbows
That's one of the signals I use to detect if YouTube videos are AI slop. If it's narrated by a non-native speaker, it's much more likely to be high quality. If it's narrated by a British voice with a deep timber, it's 100% AI.