domain:science.org
Tiber727
10m ago
why is labor so expensive?
Because in a prosperous society people want a lot of money for their labor, because they have a lot of opportunities and everything costs more in a prosperous society.
Why is there the need to exploit interns?
Begging the question. You assume they did it because they needed to, yet somehow most other companies don't do this. Why did those other companies not need to?
The only body who can actually change laws in Congress.
Yes, and said body uses agencies and interns to provide them information, because they are old men whose major skill is campaigning. Hell, Republicans actively campaign on there being too many laws. An agency designed to find laws that are obsolete and clean up spaghetti laws sounds exactly in line with what they claim to want.
The_Nybbler
In the game of roller derby, women aren't just the opposing team; they're the ball.
13m ago
Not only did the bad things social conservatives predicted about gay marriage come to pass, a lot of the stuff social conservatives were jeered at (truly or falsely) for predicting about gay marriage came to pass. The gay marriage pie chart meme is over 15 years old now; since then the terrorists won in Afghanistan, schools have been teaching kids how to have gay sex, various plagues (monkeypox, COVID) have erupted (though no locusts or frogs), and we've got a war in Ukraine (OK this is weak sauce because the meme specified WWIII).
hydroacetylene
15m ago
No comprehensive, up-to-date, source exists.
I'm a social conservative, and the new orthodox faith of the One, True, Catholic Church of Trans Rights is not convincing me to shift on that. All the former gay rights activism that successfully sold the line "if you're not gay, this will have no effect on your life" to the mainstream and the trans activism that piggy-backed on this ("why are those bigoted conservatives so obsessed with bathrooms? no trans person has ever said anything about bathrooms, it's all them!") couldn't maintain the facade.
One of the things that has struck me about the trans backlash, which I think is real, has been its unwillingness to extend the slightest charity to social conservatives qua social conservatives. To put it bluntly and perhaps uncharitably: if the social conservatives warned you that this would happen, and now this has happened, perhaps you ought to consider whether or not they had a point.
So, for instance, I see worries that opposing such-and-such trans issues might overspill into opposing same-sex marriage. But social conservatives at the time said clearly that one of the issues with same-sex marriage was that it would undermine the gender binary. They were right, on facts. They have in fact, regularly been right on the facts. So now that the thing they warned would happen as a result of gay marriage has happened... shouldn't that make their judgement of gay marriage more credible, not less?
The thing is, the push for gay marriage included a number of predictive arguments that have since proven to be incorrect. "Gay marriage will have no effect on your life" was untrue. "Gay marriage is not a stepping stone to more radical activism" was untrue. "The normalisation of and acceptance of homosexuality will not lead more people to identify as homosexual" (deployed in gotchas like "gay marriage won't make you turn gay, why do you care?") was untrue. I suppose you could quibble causation and correlation, but the course seems pretty intuitive. Yet I still see this quite determined hostility to re-evaluating.
iprayiam3
38m ago
More recent models, o1 onwards, have further training with the explicit intent of making them more agentic, while also making them more rigorous, such as Reinforcement Learning from Verified Reward.
Being agents doesn't come naturally to LLMs, it has to be beaten into them like training a cat to fetch or a human to enjoy small talk. Yet it can be beaten into them.
I'm not generally an AI dismisser, but this piece here is worth pausing on. From my experience, ChatGPT has become consistently worse for this effort. It has resulted in extrapolating ridiculous fluff and guesses at what might be desired in an 'active' agentic way. The more it tries to be 'actively helpful', the more obviously and woefully poorly it does at predicting next token / predicting next step.
It was at its worst with that one rolled back version, but it's still bad
hydroacetylene
48m ago
I've talked about this, not at great length, but I've mentioned it, before- cheap labour is cheap for a reason. Interns are a vast improvement over people who'll work for $15/hr in a corporate setting(and BTW, the tenth percentile wage is just above $14/hr. American labour is just really expensive).
There's plenty of working class people who happily make $15-$22 hr. There's a reason they don't get better jobs eventually. Corporate interns generally know things like 'how to keep themselves on track to hit deadlines without constant supervision' and 'how to follow directions correctly without asking for fifteen clarifications every sentence'. These may not be specific skills, but working in a white collar office environment requires abilities like this. Yes, requiring a college degree for this work is excessive, that's why students(who don't yet have one) are doing it as an internship. No, there's not really a solution here(go ahead, name it- no, things like 'flying pigs will provide character references' and 'we'll just kick everyone out of highschool who isn't college material so a diploma counts for the same thing' don't count).
TequilaMockingbird
Brown-skinned Fascist MAGA boot-licker
50m ago
Perhaps it would've been more accurate of me to say "This is part of the reason why LLMs have such difficulty counting..."
But even if you configure your model to treat each individual character as its own token, it is still going to struggle with basic mathematical operations.
Grant_us_eyes
50m ago
I'm curious what you mean by 'low-level', here. I've heard Obsidian described many ways, but I don't think I've heard 'low-level' before.
kky
58m ago
Good post. Interesting to see how your perspective intersects with the other critics of LLMs, like Gary Marcus’ consistently effective methods for getting the systems to spit out absurd output.
In my own experience, the actual current value of neural network systems (and thus LLMs) is fuzzy UIs or APIs. Traditional software relies on static algorithms that expect consistent and limited data which can be transformed in highly predictable ways. They don’t handle rougher data very well. LLMs, however, can make a stab at analyzing arbitrary human input and matching it to statistically likely output. It’s thus useful for querying for things where you don’t already know the keywords - like, say, asking which combination of shell utilities will perform as you desire. As people get more used to LLMs, I predict we will see them tuned more to specialized use cases in UI and less to “general” text, and suddenly become quite profitable for a focused little industry.
LLMs will be useful as a sort of image recognition for text. Image recognition is useful! But it is not especially intelligent.
OliveTapenade
1hr ago
But what RLHF does is create a meta-level reward landscape. The model learns that generating text which corresponds to verifiable facts gets a positive reward, and generating text that gets corrected by users gets a negative reward. It's not learning the "vector for truth." It's learning a phenomenally complex function that approximates the behavior of "being truthful." It is, in effect, learning a policy of truth-telling because it is rewarded for it.
I'm not sure how this makes sense? The model has no access to verifiable facts - it has no way to determine 'truth'. What it can do is try to generate text that users approve of, and to avoid text that will get corrected. But that's not optimising for truth, whatever that is. That's optimising for getting humans to pat it on the head.
From the LLM's perspective (which is an anthropomorphisation I don't like, but let's use it for convenience), there is no difference between a true statement and a false statement. There are only differences between statements that get rewarded and statements that get corrected.
100ProofTollBooth
Dumber than a man, but faster than a dog.
1hr ago
and I might claim what I see as a simpler explanation - they wanted cheap labor but felt bad about it.
Yes! So ask the next logical question; why is labor so expensive? Why is there the need to exploit interns? Why can't they just hire a guy or gal who wants a job for a reasonable price? Because labor laws make it too expensive to hire people cheaply!
The government should have an agency or committee dedicated periodic review of laws to see which laws can be retired, or if multiple overlapping laws can be combined for clarity and brevity.
And then what?
The only body who can actually change laws in Congress. Do you know how many congressionally mandated reports there are? Approximately eleventy billion. Do you know how many congress deeply reviews to implement their recommendations? Approximately negative eleventy billion. What would be the point of your hypothetical new agency? And, in order for it to be created, Congress would have to authorize and fund it. That's not going to happen. You're suggesting a "fix" that is obviously and demonstrably untenable.
100ProofTollBooth
Dumber than a man, but faster than a dog.
1hr ago
I don’t know, im personally of the opinion that there are good and bad ways to achieve any goal and that there are always trade-offs that come with any of it.
...
I just want a country where the bulk of people can live a reasonable lifestyle, and where a setback isn’t fatal.
Define "reasonable." Because what you define as reasonable may be reprehensible to me.
A lot of the tradeoffs we're implicitly talking about have to do with security over possibility. If the shot at a successful and independent life means that I could also, with equal or even higher probability, end up destitute, i'll take the deal so long as I am in control of myself. Sacrificing autonomy and independence so that the government can spoon feed me a "comfortable" (but dependent) life? No thank you.
I’m generally with at least tge idea that whatever the form a government takes, the most important thing is customer service.
For a lot of bedrock constitutional reasons, the American government can never be good at what you term "customer service." The only way is to let the customers help themselves - i.e. less government.
Joyful
1hr ago
I've been journaling off and on since childhood and digital journaling consistently since 2013.
TLDR: Obsidian. I use a template some of the time, and a journal review process I'll describe below.
I started in a program called Liquid Story Binder X, a locally installed program for writing, where you could attach entries to a calendar and it was very rewarding to look at the calendar and see all the dates lit up for the dates you had done them. This was abandonware, however, and did not autosave (at least by default) so I eventually moved to Evernote Remember before one note or google keep or anything when Evernote was a sexy startup unicorn that you could access all of your personal files? And then there were some unpopular changes, they stopped doing their own storage (hence no advantage in avoiding google/amazon) and pared down the capabilities of free accounts so that you could no longer install it on all of your devices for free. I had a paid account some of the time but I just wasn't getting enough usage out of it. So since my journal was on google servers anyway I jumped to Google Docs I brought over all of my files and linked them in a spreadsheet. I also set up my template so that I was creating the journal entry in a google form, when I wanted to. This was when I started doing journal reviews some days. If I feel like doing it, I reread journal entries from the same day. As I stated before I started journaling in 2012, so I have between 4 and 12 entries for any given day. Sometimes I review all of the entries, sometimes only the even or odd years. On review I occasionally delete old entries (if they were sparse) or add a + to the end of the name if I know it is a good one. I became really disillusioned and untrusting of Google, I think it was when they started scanning your personal files for copyrighted material circa 2020 so I jumped to Obsidian I have a dataview table that shows me all of the entries for today (so I can do the review process above) and I store new entries in one big folder in the format 22JUL2025. I have a template with a few brain-dumpy questions (what would make today great? dreams?--logging your dreams helps you remember them) and some memory focused ones(Did anyone say anything funny? What was the best thing about yesterday? Who did you talk to? What are you reading/watching?).
100ProofTollBooth
Dumber than a man, but faster than a dog.
1hr ago
I didn't know these things.
Can you provide any good rec's for a "State of the SSPX vis-a-vis the Vatican" at present?
FarmReadyElephants
1hr ago
The US didn't really put all that much effort into taking out Assad. Turkey did.
Nah, the US oil sanctions were pretty decisive. You can see GDP per capita dive as they go into effect. US support for various jihadists also didn't help. But it was the sanctions that were the killer. I was hopeful that Russian aid could Assad hold on, but it wasn't enough.
IMO the post-Cold War USA record in the Middle East is a giant humanitarian tragedy, and it has wiped out most of the remnants of a Christian culture going back 1900 years. Many other people have made the case, so I'll spare both of us repeating it.
Amadan
Enjoying my short-lived victory
1hr ago
I lean more towards @TequilaMockingbird's take than yours but I agree that his explanation of why LLMs can't count threw me off. (If you ask ChatGPT why it has trouble doing simple math problems or counting r's in "strawberry," it will actually give you a pretty detailed and accurate answer!)
That said, a lot of your objections boil down to a philosophical debate about what "counts" as intelligence, and as far as that goes, I found your fish/bird metaphor profoundly unconvincing. If you define "intelligence" as "able to perform well in a specific domain" (which is what the fish judging birds to be unintelligent is doing) then we'd have to call calculators intelligent! After all, they clearly do math much better than humans.
FoundViaTwitter
1hr ago
There's too much overthinking in this thread.
Nazism is reviled because of the inherent implications for multiracial, multicultural societies. The main thrust of Nazism and Hitler - the enactment of an ethnonationalist society through violence on a country-wide scale - is incompatible in a nation where "less than half of US children under 15 are white".
Communism has an offramp because it's in principal an economic ideology.
Im not sure if it's fair to say it "destroys" anything, but it certainly fails to capture certain sorts of things and in the end the result is the same.
A lot of the frustration I've experienced, stems from these sorts issues where some guy who spends more time writing for thier substack than they do writing code dismisses issues such as those described in the section on Lorem Epsom as trivialities that will soon be rendered moot by Moore's Law. No bro they wont, If you're serious about "AI Alignment" solving those sort of issues is going to be something like 90% of the actual work.
As for the "foom" scenario, i am extremely skeptical but i could also be wrong.
SubstantialFrivolity
I'm not even supposed to be here today
1hr ago
If you can't get past the ridiculous "one unit per type per hex" limit, that's understandable
That change is the best change in the game! Warfare is so boring in Civ 4 because there's no gameplay to it, if you have a stack that counters their stack you win. I am sympathetic to the argument that doom stacks were better because the AI was more competent with them, but can't really understand preferring them as a game mechanic.
I have no reason to disbelieve that the sections you've quoted from the Red Terror entry on Wikipedia actually happened. In fact, I'm inclined to believe all of it.
However, I do have to say that some of it - (1) the Voronezh Cheka rolled naked people around in barrels studded internally with nails, (2) Chinese Cheka detachments placed rats in iron tubes sealed at one end with wire netting and the other placed against the body of a prisoner, with the tubes being heated until the rats gnawed through the victim's body in an effort to escape, (3) the Cheka in Kislovodsk, "for lack of a better idea", killed all the patients in the hospital - reads like the more fantastical and debunked stories of the Holocaust that deniers always trot out to muddy the waters.
I don't want to dwell too much on this topic, but could it be that these more horrific types of tortures were limited to just a handful of people and the rest were summarily executed?
hydroacetylene
2hr ago
I remember WWII history, but far less of it than Holocaust.
sammysoccer411
2hr ago
One can argue 'monetary value is not use-value'. Sure. But name a better predictor.
In practice, prices are literally--like quite literally and exclusively--the result of billions of people voting with their dollars, based on how much utility they believe item x has. What could be a better predictor average of use-value than every persons' opinions on use-value, averaged out?
hydroacetylene
2hr ago
It's totally normal for people to describe shit they think as evil as 'kgb', though. You're correct that people call their political opponents Hitler more than Stalin, but there's always been a token of axiomatic evil in figurative speech- it used to be the biblical pharoah(like from Exodus). Hitler's portrayal during WWII was actually rather buffoonish more than outright evil; the Japs on the other hand...
Now why Hitler gets the title rather than Tojo, that might just be the dominance of Jews in Hollywood. I can remember old folks using terms like 'banzai' to refer to crazy evil, but that was more specific to the crazy part. I can definitely remember, quite recently and by younger people, Stalin used as a metaphor for totalitarian evil. But Hitler definitely takes the generic spot.
roystgnr
2hr ago
Half the reason TheMotte is here is that Scott went viral a decade ago discussing Social Justice And Words, Words, Words.
Probably the morally correct number. But what is the militarily correct number? That is, the number that Israelis should expect to be alive after a successful campaign in which a functional, strongly anti-terrorism regime rules that territory.
I suspect the number approaches zero.
More options
Context Copy link