This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Around 1900, Europeans + European offshoots made up about 30%+ of the world population, today it's 7%. Higher education of women is strongly associated with low fertility, it's about as hard a fact as anything in the social sciences. If you want to reduce a country's fertility, educate more women.
Race suicide and replacement migration are a key trend of the 20th and 21st centuries. If there are no upcoming gamechangers in longevity, AI or similar, then we should expect this trend to continue. Then I suspect many, (including women) will look back on these predictions and theories with a rather different attitude than sneering and derision. Say, what's the Islamist stance on women's rights? What does the average bloke in Nigeria think about women in higher education? What about the punter in Uttar Pradesh, how does he think women should be treated and how does he actually treat them? They're already the Global Majority and will be the Overwhelming Global Majority, probably the Local Majority soon enough, considering migration trends and the limitless shortsightedness of the Western political class.
Oh and even if we do get a gamechanger in AI, don't worry, our anti-racist establishment and media has helpfully ensured that non-Grok AIs prize the life of a Nigerian somewhere around 2-20x more than those of white countries like France or Germany: https://arctotherium.substack.com/p/llm-exchange-rates-updated
The article is fascinating, and of course damning. I would like to posit a theory I've had ruminating in my head for awhile and which I doubt is unique but is certainly undervocalized:
LLM pretained safetyism weights and filters are looping back onto themselves, with the RLHF being inherently flawed due to either specialist user biases (the Scale.ai style problems of insufficient experts being overloaded with noncore queries), or large training corpa being just overwhelmed by raw numbers (Indian and Nigerian and Indonesian populations, for example). Because the pretrained datasets overweight western frames, the corrective measure artificially inflates weightages of deficiently focused populations per the corrective measures in place: the corrective lens is itself the distorting factor, not necessarily the underlying training corpus. This corrective lens being iteratively reinforced by biased entities and market heterogenization causes LLMs to end up having frames baked in that are, without user specified repositioning, going to reflect a weirdly 'woke' consensus that is unnoticed by most.
For western feminists, the gender war framing reflects a standard liberal belief: their ideology is axiomatically superior and all who come in contact with it are wololooed into accepting the feminist/liberal order: they don't need to reproduce because prosletyzing will replenish all their numbers. Needless to say this has not really worked out in observed reality, but perhaps thats just how we all have sequestered ourselves into different social media and meatspace realities. For the lesbian feminist in Portland celebrating Mamdanis win by having a trans focused poetry circle, her lived reality is entirely valid from her own perspective. 'Reality' as it were does not need to assert itself ever, and now they have AI to parrot their moral words back to them.
Isn't part of the problem that 'diversity' is somewhat fundamentally at-odds with 'next likeliest token' (or the equivalent for image generation models)? Except for whatever thermal noise is being added intentionally (which should be small) and active efforts to the contrary (which is, I think, dominating what we're seeing), the model isn't wrong to assume that "draw a person" merits a response that looks like a modal person.
Expecting "[minority fraction] of the outputs should look like [minority]" is maybe not completely crazy, but doesn't seem to align with the math as far as I'm aware. Nor is it even necessarily well-defined: which population? Should "draw an NBA player" match the NBA's demographics? Should it draw all players equally likely, or weight towards popular ones? Do we just mean current players? These are questions that have mostly been sidestepped for representation in political arenas --- affirmative action never has been asked to specify specific percentage targets, nor do I think it could do so without controversy. But for large scale computer-automated systems, it's not hard to start running cross tabs for things and finding imbalance everywhere. Not even sure myself what to do about all of that.
More options
Context Copy link
Yeah, it might well be the post-training 'to reduce toxicity' but I wouldn't discount the pre-training dataset. Imagine if you pump some poor nascent being full of all the 'white people have ruined knitting', 'the toxic whiteness of_____' 'reparations needed now' articles, all the internet... The only people who have much good to say about white people are /pol/ and various outlets like Amren or Stormfront and I suspect they just don't get included in training.
I agree that this is possible, but LLM scrapers didn't get a preselected corpa to run on, the scrapes were let loose all over the internet and then processed later. While there is endless woke garbage on NYT or Slate or Huffpo, there is also Alex Jones (PBUH) 3 kb per page html slop, murdoch outlets and disqus rants easily scrapable by the spiders. The filtering has to happen post ingestion because the internet is just that big and we 2000 era semilibertarian refugees usually lament the downfall of our old haunts rather than step foot in the now-friendly territory we used to shun.
We filter the pretraining datasets. Extremist materials are among the things routinely filtered out, and often not targeted in scraping at all. LLMs actually learn about 4chan from ADL.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I honestly think much of society is overeducated. It wasn’t until recent decades you found phrases like “educated idiot” to become more commonly observed in our parlance. In one sense if you make it independently out in the real world you have to become increasingly informed as the world today is a more complex place than it was a thousand years ago. Assembling an iPhone is much more difficult than figuring the basic uses of a garden hoe. Turchin I think was also onto something when he formalized the argument in greater detail.
The older phrase is "learned fool", which you'll find in Shakespeare.
More options
Context Copy link
Yeah, especially there's a gap between formal education and actual learning or erudition. You can breeze through a university degree these days with very little effort or knowledge acquisition, certainly never learn to think. It was bad before AI but it's gotten way worse now. I think universities should be closing down undergraduate courses en masse, in many places it's basically a scam where they coast on prestige earned by a more learned generation of scholars, conferring fancy pieces of paper on foreign students to fund vast bureaucracies that just make life worse for all involved.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link