Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

marcusolang.substack.com

Post reply on HN
Print view View original

301–310 of 533 posts

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#301

Post by SoftTalker »

phantasmish wrote: post #240
echelon_musk wrote: post #221

Earlier quoted context omitted.

Are you an English speaking American? Because being a native English speaker and actually being English, or from a former English colony will differ. I'd characterise Americans as less pretentious and more straight talking. This kind flowery language is typical (or symptomatic depending on diagnosis) of how English people actually used to speak and write. The average English vocabulary has dwindled noticeably in my l…

> I'd characterise Americans as less pretentious and more straight talking. Various registers representing a huge proportion of US English we see and hear day-to-day are terrible. American “Business English” is notably bad, and is marked by this sort of fake-fancy language. The dialect our cops use is perhaps even worse, but at least most of us don’t have to read or hear it as much as the business variety.

Most writing is intended to communicate. Business writing is intended to create an impression.

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#302

Post by heavyset_go »

bpodgursky wrote: post #273
Timpy wrote: post #25

A lot of training data was curated in Kenya[0]. I would imagine if LLM data was curated in Japan our LLMs would sound a lot like the authors of their most popular English text books. Maybe other common Japanese idioms would leak in to the training data, like "ね" or "でしょう", ChatGPT would say "Don't you agree?" at the end of every message. [0] https://www.theverge.com/features/23764584/ai-artificial-int...

This is a wild misunderstanding of LLMs. Data labeling has nothing to do with generating the astronomical text corpus used to train modern LLMs.

The HF part of RLHF to refine the output of LLMs also happens in these places

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#303

Post by Uehreka »

nitwit005 wrote: post #193

This is also happening to artists, people who make YouTube shorts, and similar. Everyone gets accused of being AI if the feel happens to match. I'm sure there's some voice actor out there who can't get work because they sound too similar to the generated voices that appear in TikTok videos.

I did a video a couple years ago about a thing I did in Factorio[0] and got a couple comments who didn’t ask if I had used an AI voice, they just straight up told me that the AI voice I used was off putting. I didn’t use an AI voice, in fact I appeared on camera at the end of the video in part so that people wouldn’t have to guess, but I guess people who thought I was AI didn’t feel like watching the whole video.

I suppose I don’t mind people using AI voices if they have a thick accent or are shy about their voice, but if I’m watching a video and clock the voice as AI (usually because the tone is professional but has no expression and then the speaker mispronounces a common word or acronym) it does make me start to wonder if the script is AI. There are a lot of people churning out tutorials that seem useful at first but turn out to have no content (“draw the rest of the owl” type stuff) because they asked AI to create a tutorial for something and didn’t edit or reprompt based on the output. The video essay world is also starting to get hit pretty hard, to the point that I’m less willing than ever to watch content unless I already know the creator’s work.

[0] Shameless plug: https://youtu.be/PGiTkkMOfiw

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#304

Post by zahlman »

lynndotpy wrote: post #164
rukshn wrote: post #28

I had a similar experience. We were talking about a colleague for using ChatGPT in our WhatsApp group chat to sound smart and coming up with interesting points. The talk sounds so mechanical and sounds exactly as ChatGPT. His responses in Zoom Calls were the same mechanical and sounds like AI generated. I even checked one of his responses in WhatsApp if it's AI by asking the Meta AI whether it's AI written, and Meta…

I'm definitely in the "ChatGPT writes like me" experience. I am a big fan of lists, and of using formatting to make it all legible on a short skim. I'm a big fan of dyslexia-friendly writing too, even though I am not dyslexic myslef. I can't blame others though- I was looking at notes I wrote in 2019 and even that gave me a flavor of looking like a ChatGPT wrote it. I use the word "delve" and "not just X but also Y o…

> dyslexia-friendly writing

... How does that work, exactly?

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#305

Post by heavyset_go »

buu700 wrote: post #256
bakugo wrote: post #236

Earlier quoted context omitted.

> Why can't the LLM refrain from improving a sentence that's already really good? Because you told it to improve it. Modern LLMs are trained to follow instructions unquestioningly, they will never tell you "you told me to do X but I don't think I should", they'll just do it even if it's unnecessary. If you want the LLM to avoid making changes that it thinks are unnecessary, you need to explicitly give it the option t…

That may be what most or all current LLMs do by default, but it isn't self-evident that it's what LLMs inherently must do. A reasonable human, given the same task, wouldn't just make arbitrary changes to an already-well-composed sentence with no identified typos and hope for the best. They would clarify that the sentence is already generally high-quality, then ask probing questions about any perceived issues and the…

Reasonable humans understand the request at hand. LLMs just output something that looks like it will satisfy the user. It's a happy accident when the output is useful.

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#306

Post by overfeed »

unsupp0rted wrote: post #285
overfeed wrote: post #263

Earlier quoted context omitted.

Perhaps yet another American cultural artifact. One that - if I were to guess - originated from the Calvinist disdain for ostentiousness.

Yes yes, anybody who prefers plain, easily parsed wording is American. Wording? Don't you mean diction?

A -> B =/= B -> A.

I didn't claim that this was exclusively American. Though I'd have to admit that one doesn't have to be American to adopt Ameracanisms: rhotic Rs, Netflix color-grading, and copy-cat political movements are other American cultural artifacts showing up across the world due to America's dominance of the zeitgeist.

Rap verses in pop songs wasn't a spontaneously phenomenon across the globe, the origins are tracably American - but that doesn't make all rappers American.

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#307

Post by buu700 »

heavyset_go wrote: post #305
buu700 wrote: post #256

Earlier quoted context omitted.

That may be what most or all current LLMs do by default, but it isn't self-evident that it's what LLMs inherently must do. A reasonable human, given the same task, wouldn't just make arbitrary changes to an already-well-composed sentence with no identified typos and hope for the best. They would clarify that the sentence is already generally high-quality, then ask probing questions about any perceived issues and the…

Reasonable humans understand the request at hand. LLMs just output something that looks like it will satisfy the user. It's a happy accident when the output is useful.

Sure, but that doesn't prove anything about the properties of the output. Change a few words, and this could be an argument against the possibility of what we now refer to as LLMs (which do, of course, exist).

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#308

Post by tom_ »

shadow28 wrote: post #283
OptionOfT wrote: post #250

Earlier quoted context omitted.

Yea. I've been in the States for 8 years, yet sometimes my brain thinks about a word and writes it down phonetically. But hey, at least you know I didn't use ChatGPT to conjure that comment.

Sorry for the pedantry, but "thought" and "taught" are actually phonetically different (/θɑːt/ vs /tɑːt/).

That may not be true if you struggle with "th"? Some ESL speakers do.

Re: I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me

#309

Post by zahlman »

Yizahi wrote: post #117

While author is correct in general, I would like to add a counter-point regarding em-dashes specifically. Yes, many people use them like this - and many website frameworks will automatically replace a keyboard not-really-a-minus symbol with em-dash. So that is not a sign of the LLM generated slop. What LLMs also do though, is use em-dashes like this (imagine that "--" is an em-dash here): "So, when you read my work--…

> What LLMs also do though, is use em-dashes like this (imagine that "--" is an em-dash here): "So, when you read my work--when you see our work--what are you really seeing?"

>You see? LLMs often use em-dashes without spaces before and after, as a period replacement.

It would not make any sense at all to use periods in the places where those em-dashes are supposedly "replacing" periods in the example.

Post reply on HN