> It seems unfair to call out LLMs for "spam, scams, propaganda, and astroturfing." These problems are largely the result of platform optimization for engagement and SEO competition for attention.

They didn't create those markets, but they're the markets for which LLMs enhance productivity and capability the best right now, because they're the ones that need the least supervision of input to and output from the LLMs, and they happen to be otherwise well-suited to the kind of work it is, besides.

> This isn't unique to models; even we, humans, when operating without feedback, generate mostly slop.

I don't understand the relevance of this.

> Curation is performed by the environment and the passage of time, which reveals consequences.

It'd say it's revealed by human judgement and eroded by chance, but either way, I still don't get the relevance.

> LLMs taken in isolation from their environment are just as sloppy as brains in a similar situation.

Sure? And clouds are often fluffy. Water is often wet. Relevance?

The rest of this is a description of how we can make LLMs work better, which amounts to more work than required to make LLMs pay off enormously for the purposes I called out, so... are we even in disagreement? I don't disagree that perhaps this will change, and explicitly bound my original claim ("so far") for that reason.

... are you actually demonstrating my point, on purpose, by responding with LLM slop?

> In aggregate, however, I believe in the US presidential elections end up voting for their own best interests, as they see it

This is extremely close to one of the early "OK, but maybe there's a reason what we're observing at the individual level isn't so scary" hypotheses explored by political science in the latter half of the 20th century—that individually poor choices would nonetheless produce good outcomes by being in some way chaotic and the good outcomes often manifesting as attractors in that chaotic space, or something like that, or by some "wisdom of the crowds" effect that emerges in aggregate. These approaches have been found untenable despite much trying, though I think there are some limited efforts at it still under way.

HOWEVER! I think after this post I do see what you're actually getting at, which is that if people believe they voted in their own best interests ("as they see it" being key) then they may believe they did in-fact do that indefinitely, even if entirely incorrect, so long as they... well, continue to believe so.

The prisoner voting to remain a prisoner not because they don't want to be free—not because if you describe completely and in detail, leaving nothing out, the conditions they're in-fact in they tell you they would love to live that way (they claim they would hate it!), and then if you also describe free life they claim that is the outcome they would rather have, and if you carefully probe you find that it's not even for some greater-interest purpose they are voting to remain imprisoned (it's not that they believe they'd be a danger to others if free, for example), but because they believe they aren't in prison despite [gestures at their prison cell]—is voting in their own interest.

By that standard, yes, a lot more voters are voting in their own interest than may be reckoned by other standards.

Spam, scams, propaganda, and astroturfing are easily the largest beneficiaries of LLM automation, so far. LLMs are exactly the 100x rocket-boots their boosters are promising for other areas (without such results outside a few tiny, but sometimes important, niches, so far) when what you're doing is producing throw-away content at enormous scale and have a high tolerance for mistakes, as long as the volume is high.

Yeah, I think you nailed it. A ban on playing against "the house" would do it. Taking a fixed amount from each pot (as at poker tables) for play among patrons would still be allowed, but slot machines wouldn't. Your solution's much better than a full ban because it wouldn't drive as much illegal betting (a problem no only because it circumvents the law, but because for gambling in particular but for any black market, really, it tends to become connected with other criminal activity)

> Although, if you believe in that theory, you should reject democracy and aim for some form of aristocracy or monarchy.

Not necessarily! It means that the model of the typical voter's behavior (and of the reasons why elections go the ways they do) isn't what many conceive it to be (or hope it may be), and that democracy's weaknesses, vulnerabilities, strengths, and capabilities may in-fact be at least somewhat different from what one operating from that idealized (and apparently very wrong) model of voter behavior would expect. It could still be the best of a bad lot.

> They did not get "fooled" and bait-and-switched even if they later feel the performance was not great.

They are extremely often operating from incorrect information, either regarding facts about the state of the world, or about probable outcomes of various policies. This can include things that directly affect them (or don't) in ways that one would expect them to notice—one fun form of study that's been run a few times is to ask a population whether a tax increase or decrease that in-fact affected only a tiny sliver of the population but was the subject of substantial propagandizing and/or publicity affected them personally (this is about as direct as it gets!) and the typical result is pretty much exactly what your most-pessimistic guess would be.

Supposing that people very-often hold a bunch of incorrect beliefs about how policies affect them but are also good at voting for their own interests when it comes time to mark the ballot is probably somewhere in the category of wishful thinking—and that's assuming motivations and intentions focused on policies and their outcomes in the first place. There's less-strong but still-quite-strong evidence that, as the kids say, "vibes" are a huge factor in the outcomes of elections, even when those "vibes" come from things that even the extremely politically-ignorant ought to know have nothing much to do with, say, who the President is, like a rash of shark attacks for example. This, of course, doesn't mean that this "vibes-from-irrelevant-stuff" voting makes the difference for anywhere near as many people as incorrect information does (it almost certainly doesn't) but that it has an outsize effect on the true-swing (not self-reported swing, that's mostly bullshit) vote, which tends to consist almost entirely of so-called "low-information voters", with the result that it may not have any effect at all on most voters but elections still turn on it (one of a billion reasons FPTP voting sucks is that it amplifies the power of this effect).

I do think, separately, there are cases of rational trade-offs, of picking (say) an anti-abortion candidate who holds many other positions one dislikes because one's stake in one's position on abortion is that important. That's not the kind of thing I mean, and I don't think it's the kind of thing most people mean when they say people are making mistakes by "voting against their own interests", though the effect of such a choice may well be that one is also in these cases (consciously!) voting against one's own interests on various issues.