Comment by chrismorgan | Hacker News Reader

Comment by chrismorgan | Hacker News Reader

> “This is one of the first studies, if not the first, to show that the use of AI in writing could lead to cultural stereotyping and language homogenization,”

I just want to make sure others agree and it wasn’t just me (or perhaps non-Americans in general)—it was blindingly obvious this would be, must be, the case, right? That although this might be the first formal study of it, there would have been literally no doubts as to what the outcome of such a study might be? That at least some degree of language homogenisation will be quite inescapable if you do LLMs the way we have?

On the cultural aspects, it’s well-documented and -understood what effects US TV and movies have had on other countries. There really isn’t anything new about LLMs or AI here, it’s just standard globalisation effects.

(I also just now learned what a crazy term “Global South” is <https://en.m.wikipedia.org/wiki/Global_North_and_Global_Sout...>, and how it does not mean at all what I thought it meant or what any sane person would expect. Was it not enough that “Western” bears no strong correlation to geography, that we need more terms that utterly abuse geographical references when they’re actually about socioeconomic characteristics? Apparently I have moved south by migrating from Australia to India.)

'Global south' is just the latest in a fraught space where every term seems to eventually get dragged down. Originally it was 'third world' that started off as a positive, aspirational term but became a derisive term. Then it was 'developing country' that was supposed to be not judgmental but became so. I'm sure 'global south' will go the same way.

It’s funny, because on the face of it, it seems the worst of all. You might as well just say “dark skinned countries,” since that’s what the term is effectively getting at.

>You might as well just say “dark skinned countries,”

Argentine is 97% European descent. Where is non-white countries like Japan and South Korea are in the Global North.

It isn't about race/skin color. The "Third World" was well suitable until Soviet Block (Second World) collapsed with most of its components going either into the First/Western World or into the Third World thus resulting in the world being partitioned just into 2 large parts with Russia+Belarus being the distant small 3rd, so small that it is just easier to count them into the First/Western style developed world especially considering that they both moved into capitalism losing that major trait - socialism - of the Second World. (Though 30+ years later Russia does more and more drift toward the former Third World, i.e Global South).

We all known Japanese and Korean people are white by proxy in a lot of people's minds.

My point wasn't about the grouping itself though, just the term global south effectively being "worse" than the previous terms. For the northern hemisphere, where most people live, south = darker skin.

The "global south" is a term which originates in writings about the vietnam war in the 60s. It's by no means a new term.

Use third world country if you want, but at least in my mind, the terms "first/second/third world" are more tied to which hegemony you fall under, where the first world is the US hegemony, and the third world is without a hegemon. "First world" is kinda synonymous with "the west". To me, the term "global south" communicates that it's being contrasted against all of "the north", both east and west, while using the term "third world" communicates that the context makes a distinction between "the west" ("first world") and the old eastern bloc ("second world") somehow relevant. Some "second world" countries such as China are also part of the "global south".

I'm sure there are people who use the term "global south" simply because they perceive it to be less judgemental somehow, but I might've used it in this context because "third world" communicates something different.

Honestly though, I would've probably just used the term "non-western" (maybe "non-first-world"?), since that's the distinction the article actually draws. Eastern Europe is also affected here, after all. Or maybe I would've drawn the distinction at "US" and "non-US", since Europeans don't necessarily want their writing to sound American either and the old US hegemony seems to be on its last legs.

Correct but besides the point. The question of the study is "how much and how do you quantify this?".

There are lots of blindingly obvious qualitative statements, where the quantitative parts is far from obvious. That makes them a good starting point for research.

This is like writing about Newton's theory of gravity as "scientist finds out apples fall downwards. Wasn't that already obvious?".

A few months ago, there were articles about English having more a Nigerian (I believe) dialect because that's where the labelling of the supervised learning happened in the early days.

If that had continued, combined with where actual users are, perhaps it would have broadened English instead?

I agree that homogenization is the likely default outcome. ML models do tend to have a strong tendency to prioritize modelling the most common things well, which makes output in a generative models also biased towards mean/mode. And there are the limitations in representation in the datasets themselves. And also bias in that the developers are primarily English speaking, so that language gets priority. But I do no see homogenization as inevitable. LLMs do pick up a wide range of speech patterns, both regional, different periods of time , sociolects and styles. And they are pretty darn good at "role-playing", outputting language tailored to a particular role. And this can be configured rather effectively using a session or system prompt. So if everybody picked their voice, we could perhaps get a broadening of outcomes. Maybe I should add "always speak like a refined scholar from the Victorian era, spiced up with 80ies British goth"...

You know what, actually if a non-American movie non-ironically tries to be like American it instantly becomes kitsch. Even British movies are quite distinct from the from American.

Anything AI writes is dull anyway, it writes stuff that nobody wants read beyond getting some information. Maybe if you are learning English you may pick up something from it though.

Also, I recall something about AI English actually being Nigerian English because those companies used a lot of Nigerians in training.

> Apparently I have moved south by migrating from Australia to India.

Yeah, I we understand "south" to mean "closer to the equator". (That's kind of how it works in the popular imagination. E.g., southern Brazil is more "nordic" than northern Brazil.)

That’s definitely not how I’ve ever seen it used in Australia or India. North and south are fairly strictly geographic terms. South India is roughly the southern half of the country—following a cultural and geographical divide, but pretty neat overall. Southern Australia follows the southern edge of the country (not to be confused with South Australia, a state in the south). One is in the northern hemisphere and one in the southern, but they use the terms the same way, pointing to the poles.

> That's kind of how it works in the popular imagination

It absolutely does not. South means, well, South.

> E.g., southern Brazil is more "nordic" than northern Brazil.)

How so? Southern Brazil is clearly closer to the South pole than Nothern Brazil.

It's a racial term. "South" as in "dusky" or "brown", etc.

Reply on news.ycombinator.com